WO2009022084A1 - Method for automatically composing a personalized ring tone from a hummed voice recording and portable telephone implementing this method - Google Patents

Abstract

The invention relates essentially to a method of automatically composing a personalized ring tone (13) from a recording of a voice signal (5) sung or hummed by a user. In this method, analysis parameters (25) are extracted from the voice signal (5), such as the pitch and/or loudness and/or attack of the notes of the voice signal, and the voice signal (5) is transformed into a ring tone comprising at least one musical track. In accordance with the invention, to transform the voice signal (5) into a ring tone, the voice signal is tuned by transposing said voice signal (5) as a whole of one and the same pitch in such a way as to minimize a distance between the voice signal (5) as a whole and a tempered chromatic range, and the voice signal (5) is tempered by replacing the notes of the transposed voice signal by tempered notes.

Description

 Method for automatically dialing a personalized ringtone from a voice and cell phone recording using this method

[0001]. A method and apparatus for automatically dialing a custom ring from a recording of a voice signal. The voice signal corresponds to a hummed voice, which means in the invention a sung or spoken voice, or various vocal noises, such as whistles, onomatopoeia, or "human beat boxing" which consists of imitation voice of a drum machine, scratch, and other percussive instruments.

[0002]. The invention is intended in particular to automatically develop the ringtone according to the musical intention of the user and a type of music chosen by the user or imposed by the service provider, through a telephone interface , or any other multimedia device, such as a website. The musical intention is defined in particular by the notes and / or the rhythm and / or the timbre and / or the expressiveness of the user's voice (vibrato, dynamics, etc.).

[0003]. The invention can be implemented locally, for example on the mobile phone, or remotely, via a network, on a server.

The invention finds a particularly advantageous application for the development of a ringtone of a mobile phone.

[0004]. Today there is a logic of personalization in the field of mobile phones. Indeed, we see that the colors and shapes of phones, or the presentations of their menus, are increasingly varied, so that the user can choose those that best fit his personality.

[0005]. Regarding the ringtones, several registered styles are offered to the user, the latter can choose the one best suited to the music he likes. However, given the recorded appearance and thus frozen ringtones, personalization is limited to ringtones offered by the phone. There are many methods for downloading ringtones corresponding to tunes of known songs but again the ringtones are limited to existing music tracks.

[0006]. The invention makes it possible to increase the degree of personalization of the ringtones by proposing a method allowing the user to elaborate himself a ringtone from his voice, spontaneous and expressive entry par excellence. The invention thus allows the user to create a ringtone that he himself imagined by humming in his phone, and opens unlimited possibilities in the development of a ringtone.

[0007]. More precisely, in the invention, it is sought to transform the voice signal into a coherent ring. That is to say, we do not necessarily seek to transcribe the voice with absolute fidelity, but rather to extract the musical intentions to turn this voice signal into a music and / or accompany this voice signal. This music may consist of one or more coherent musical tracks. This music and / or accompaniment of this voice signal will be constructed to respect rhythmic and / or melodic musical composition and / or arrangement rules. [0008]. These music composition rules include:

- rhythmic rules, such as prohibiting notes of short duration, or stalling the music on a given tempo or rhythm pattern and / or

melodic rules, such as for example tempering the notes and / or resetting them in a given range and / or concluding an incomplete melodic phrase and / or

rules of arrangement that can be representative of a style, such as for example generating several musical tracks from a monophonic signal while ensuring coherence of the musical tracks with respect to the rhythmic and / or melodic rules and / or musical style.

[0009]. In one example, in the invention, from the monophonic voice signal, corrected if necessary, a set of tracks is created. arranged between them and played preferably by different instruments. These tracks correspond to the musicality of the voice signal and the music style chosen by the user.

[00010]. For this purpose, in the invention, the user hums in the microphone of his phone or computer connected to a site

Internet, or any device or dedicated terminal.

[00011]. Then we extract sound parameters from the voice signal such as the pitch (pitch), the intensity (or velocity, or volume of the voice), the attack (sharp peaks of intensity or consonants separating voiced parts). ), the timbre (roughness and brilliance in particular) and other various expressivity parameters, such as the rhythm or tempo or the range or harmony extracted from the voice. These parameters are part of the invention of the "analysis parameters".

[00012]. Then, we control how the analysis parameters are used for the ringtone synthesis according to a "style" of composition of the ringtone chosen by the user or imposed by the service provider or the manufacturer of the ringtone. apparatus. For example for a style "RnB", we apply some rules of correction and / or composition and / or transformation of the particular voice to this style of music, while for a style "jazz" other rules will be applied. For example, some notes that are considered inappropriate in the "RnB" style may be kept in style

"Jazz".

[00013]. The synthesized ring is then recorded in one of the memories of the object for which it is intended, such as a mobile or fixed telephone or a computer when the invention is used with voice-over-IP applications, or on a peripheral device. dedicated terminal.

[00014]. Terminal refers to any device and / or software used for the composition and / or storage and / or listening ring. A terminal can be for example a mobile phone, a fixed telephone, a computer or a dedicated electronic equipment. The same terminal can be used throughout the process but several terminals can also intervene. For example, a computer can be used to dial the ring, ie to hum from a microphone. This same computer can be used to listen to the ringtone streaming and without having the possibility to load it on the mobile phone. The mobile phone can then be used to store the ringtone which is then sent to the user only for example if he paid the price associated with the ringtone. By extension, the term telephone or mobile phone means any terminal used for the composition and / or listening and / or storage of the ring.

[00015]. Preferably, the ring is synthesized by an arrangement of sound layers in the form of musical tracks that depend on the expressiveness of the voice of the user, and the musical style chosen or imposed. These sound layers can come directly from an audio processing of the voice, or from the playback by a virtual instrument of a MIDI track derived from the voice.

[00016]. One distinguishes the creation of ringtones of the type MIDI and the creation of ringtones of the audio type. A MIDI tone is a MIDI file consisting of a set of MIDI tracks to be played by one or more virtual instruments available on the terminal. An audio-type ringer is an audio file consisting of a set of audio tracks that can correspond to one or more MIDI tracks already played by a virtual instrument and / or to a transformation of the voice and / or pre-existing audio loops. The audio tone is played using an audio player. In the case of the audio ringing, no virtual instrument is necessary for its reading since the audio tracks are readable as such by the audio player. The MIDI format can be replaced by any mode of symbolic representation of the music and in particular by the imelody format.

[00017]. Each track is synthesized separately according to the musical style chosen by the user or imposed (for example a style of ringing type "jazz flute") and all the parameters of analysis. Thus the events of each MIDI track are elaborated from the analysis parameters according to composition rules while sound synthesis of audio tracks is done from analysis parameters and pre-recorded or synthesized sounds, or voice transformations.

[00018]. Voice signal processing can be done remotely on a server, the voice is recorded on the phone and then routed to the server which turns it into a ringtone, the latter being sent back to the phone. This remote processing is an easy solution to implement for telephone operators who only have to install the ringtone software on a single medium (the server) and who can easily control access to this server. authorizing this access in exchange for a fee.

[00019]. In a variant, the hummed voice is directly processed locally on the telephone. In this case, the user is allowed to download the ringtones program for a fee. In this local implementation, the ring is composed almost instantaneously, since there is no exchange of audio file between the phone and the server.

[00020]. Alternatively, the hummed voice is sent directly to the server from the streaming phone, transformed into a ringtone on the server and sent back to the phone as an audio or midi file depending on the type of elaboration chosen ringtone.

[00021]. Alternatively and in the case of use over a network, you can send the ringtone streaming for a preview and then return it as a file once the user has paid the fee associated.

[00022]. Different fees may be charged and for example for access to the service, depending on the time of use of the service, depending on the number of attempts to compose and then listen to the result obtained, or after listening to the result obtained , when the user decides to download or install the result obtained as a ringtone on their mobile phone. Different counters are inserted in the processing chain to allow this billing. [00023]. The invention therefore relates to a method for automatically dialing a ringtone from a recording of a monophonic voice signal, such as a song, a humming by a user, in which: - analysis parameters are extracted from the voice signal, such as the pitch and / or the intensity and / or the attack of the notes of the voice signal, and

the voice signal is transformed into a ring comprising at least one musical track, characterized in that, in order to transform the voice signal into a ring, the voice signal is tuned by transposing the set of said voice signal of a same height so as to minimize a distance between the whole of the voice signal and a tempered chromatic range, and

- the voice signal is tempered by replacing the notes of the transposed voice signal with temperate notes. [00024]. According to one implementation, to tune the voice signal,

determining a central note of the voice signal corresponding to the most frequent note of this voice signal, and

the overall voice signal is transposed in a manner that matches the pitch of the central note to the height of its closest temperate note.

[00025]. According to one implementation, to temper the voice signal, the voice signal having been previously cut into segments,

- for each temperate note, we define an associated note model,

in a loop, for each segment, the tempered note whose model is closest to the height of the segment is determined,

said tempered note is assigned to the segment, and

the model of said temperate note is updated by taking into account the height of the segment, for example by averaging the heights of the notes of the signal which have been associated with this tempered note, these note pitches being able to be weighted, the where appropriate, by their duration or intensity. [00026]. According to one implementation, in the initial note template, each tempered note is modeled by a height of the chromatic scale tuned to the central note.

[00027]. According to one implementation, to temper, we start from the last segment of the voice signal from a temporal point of view and go back to the first.

[00028]. According to one implementation, for transposing the voice signal, a tuning cost is calculated which is equal to the integration, over the duration of the melody of the voice signal, of the product of the instantaneous difference between the signal height. of voice and the nearest temperate height raised to the power p (p strictly positive real) and the intensity of the high voice signal to the power q (real positive q), and the voice signal is transposed so as to minimize the value of the tuning cost.

[00029]. According to one implementation, p is equal to 2 and q is 1.

[00030]. According to one implementation, a range is determined by implementing the following steps:

- we choose probabilities of occurrence of a note in a given range,

the degree of belonging of the melody to several scales is calculated, this degree of membership being a function of the concordance of the notes of the voice signal and the probabilities of occurrence of the notes of the scale, and

- the range with the highest degree of membership is selected.

[00031]. According to one implementation, the degree of membership of the notes of the scale is equal to the sum for all the notes of the voice signal of the product of the duration of each note raised to the power p by the intensity of each high note. to the power q and by the probability of occurrence of each note raised to the power r, pq and r being realities greater than 0 and p being different from 0. [00032]. According to one implementation, to temper the melody of the voice signal,

- one chooses probabilities of occurrence of a note in a given range, - one computes an optimal transposition for each of the candidate ranges which is a function of the concordance of the temperate notes closest to the voice signal and probabilities of occurrence notes of the candidate scale, by calculating a degree of belonging of the melody to each candidate range and for all the possible values of the transposition,

- the range with the highest degree of membership is chosen and

- transpose the notes of the voice signal of the optimal transposition associated with this range, and - tempers by replacing the notes of the voice signal transposed by the temperate notes closest to the transposed voice signal.

[00033]. According to one implementation, the degree of membership of the notes of the scale is equal to the integral for all the notes of the melody of the product of the intensity of each note raised to the power q by the difference of the note. of the voice signal in relation to the note between the two temperate notes closest to the power p by the probability of occurrence of each score raised to the power r, pq and r being realities greater than 0 and p being different from 0.

[00034]. According to one implementation, range knowledge is used to temper by minimizing a distance between the transposed voice signal note and the closest temperate note, this distance being weighted according to the probability of occurrence of the note in the range.

[00035]. According to one implementation, the probability of occurrence of a score in a given range is determined from the Krumhansl & Kessler probetones.

[00036]. According to one implementation, to transform the voice signal into a ring, it removes or groups notes whose duration is less than a reference value, for example 1 ms, and / or whose intensity is less than a reference intensity and / or whose height extraction quality is less than a reference value.

[00037]. According to one implementation, to transform the voice signal into a ring, the voice signal or a musical track already obtained from the voice signal is corrected by cleaning it and / or correcting the melody and / or recalibrating it rhythmically. and / or by deriving a melody from the voice signal.

[00038]. According to one implementation, to rhythmically adjust the voice signal, it tracks the tempo of this voice signal and recalibrates the notes of the voice signal in this tempo.

[00039]. According to one implementation, the rhythm correction is performed by imposing a fixed tempo, for example the average tempo extracted from the voice signal, the voice signal being calibrated rhythmically on the tempo imposed by a method of "Time stretching".

[00040]. According to one implementation, the rhythmic correction is performed by a rhythmic marking by the technique of "Time Warping" in which a tracing of the loud beats of the voice signal is carried out, in order to build a reference rhythm on which the musical tracks are synchronized. [00041]. According to one implementation, the melodic correction is performed by a "pitch shifting" technique in which the notes of the voice are recalibrated in temperate notes and / or in the average range of the hummed voice signal.

[00042]. According to one implementation, to compose a new melody derived from the voice signal, the notes of the voice signal and a rhythm are selected according to the extracted analysis parameters and musical construction rules that depend on a chosen musical style. by the user or imposed.

[00043]. According to one implementation, to compose a new melody deriving from the voice signal, an original melody is produced that is a response to the vocal melody and / or the vocal rhythm calculated at from the analysis parameters to establish a dialogue with a machine, or a sequence to finish the vocal melody correctly according to the chosen style.

[00044]. According to one implementation, the tone, and / or the pitch and / or other characteristics of the voice are modified by transforming the voice signal, or by synthesizing the sound environment from the voice signal.

[00045]. According to one embodiment, to develop one or more musical tracks, one or more pre-recorded rhythm loops are used in the form of an audio signal that is staggered on the tempo extracted from the voice signal.

[00046]. According to one implementation, to create one or more musical tracks, musical samples are selected from musical databases presenting the musical parameters closest to those of the voice signal over a given time interval.

[00047]. According to one embodiment, the volumes of the different musical tracks are adjusted with respect to each other, and / or effects are introduced on selected tracks, such as saturation or a sound compression effect, and, if appropriate, mixes all the tracks into an output track, and / or we introduce global effects on this output track, such as reverb.

[00048]. According to one implementation, an audio file is created from the mixed output track, in an mp3 format.

[00049]. According to one implementation, the ringtone comprises several musical tracks arranged between them according to the extracted analysis parameters and musical composition rules.

[00050]. According to one implementation, the musical composition rules are related to a style of music, such as a rock or blues style, chosen by the user. [00051]. According to one implementation, the voice signal is streamed to a server providing the extraction of analysis parameters and the development of the ringtone.

[00052]. According to one implementation, the musical tracks are obtained from a MIDI and / or audio processing of the voice signal.

[00053]. The invention further relates to a mobile phone implementing the method according to the invention.

[00054]. The invention further relates to a method for associating telephone ringtones with contacts stored in a mobile phone in which:

- we record a phrase sung by the user of the mobile phone or by one of the contacts,

the sung phrase is transformed into a ring by means of the method defined according to the invention, and the ringtone obtained is stored in a memory associated with the contact for which the ring is intended, so that when the contact calls, the ringing his correspondent is played by the phone.

[00055]. The invention further relates to a device for generating real-time music implementing the method defined according to the invention for generating a ringtone from a sung musical phrase, this device comprising means for playing this ringing ring, so that it is possible to sing on this ringtone, or mix it with music tracks to create musical tracks.

[00056]. The invention also relates to a method for producing a ringing tone in which several voice lines are recorded successively, the listening of voice lines previously recorded and processed according to the method defined in accordance with the invention being authorized during the recording of a voice. new voice line and analysis parameters that can be extracted from each record or from all records. [00057]. The invention will be better understood on reading the description which follows and on examining the figures which accompany it. These figures are given for illustrative but not limiting of the invention. They show :

[00058]. Figure 1: a schematic representation of the processing chain according to the invention for developing a ringtone from a voice signal of a user;

[00059]. Figure 2: a schematic representation of the voice signal analysis module according to the invention;

[00060]. Figure 3: a schematic representation of the midi and audio synthesis modules according to the invention;

[00061]. Figure 4: a graphical representation of the amplitude of the voice signal of the user as a function of time;

[00062]. Figure 5: a graphical representation of a raw MIDI transcription of the voice signal;

[00063]. Figures 6-8: graphical representations of MIDI musical tracks obtained from the raw MIDI track of the voice signal after application of the processing method according to the invention;

[00064]. Figure 9: a graphic representation of the rhythm marking according to the invention performed on the voice signal to synchronize rhythmically the voice transformed or not, and the instrumental track or tracks;

[00065]. Figure 10: a graphical representation of an audio signal of an RnB style drum loop that can be keyed to the voice to accompany it;

[00066]. Figure 11: a graphical representation of the raw voice signal and the voice signal transposed according to a tuning algorithm (optimal transposition of the melody); [00067]. FIG. 12: a graphical representation of the optimal transposed voice signal and the tempered voice signal according to a "tempering" algorithm of the melody over a tempered chromatic range in "La 440";

[00068]. Figure 13.1: a graphical representation of the tempered voice signal with respect to a range of non-optimal Major;

[00069]. Figure 13.2: a graphical representation of the tempered voice signal with respect to an optimal harmonic min range;

[00070]. Figure 14: a graphical representation of the raw voice signal and the voice signal transposed according to a tuning algorithm taking into account the probabilities of occurrence of the notes in a range;

[00071]. Figure 15: a schematic representation of a simplified processing chain of the invention;

[00072]. Figure 16: a graphical representation of the pitch of a voice signal as a function of time on which a tuning grid has been placed;

[00073]. Figure 17: a histogram of the pitches of the sung notes, following the 1/2 ton bands of the tuning grid;

[00074]. Figure 18: a representation of the initial model of the temperate notes tuned by 1/2 ton on the central note;

[00075]. Figure 19: a representation of the heights of the models of temperate notes, after a first analysis of the last sung notes;

[00076]. Figure 20: a representation of the final heights of the temperate note models, after a complete analysis of the voice signal.

[00077]. Identical elements retain the same reference from one figure to another. [00078]. Figure 1 shows a schematic representation of a processing chain 1 for automatically generating a telephone ring 13 from a voice signal of a user. This user 2 is in connection with an interface 3, such as a mobile phone or a computer or a terminal that is connected to a server 4 via a network, for example of the Internet type. In this implementation, the processing of the voice signal 5 of the user is performed on the server 4.

[00079]. More specifically, the user 2 defines, via the interface 3, input parameters of the method according to the invention. For this purpose, the user emits a voice signal humming in his telephone and chooses the style of music according to which the ringing 13 will be elaborated. The choice of music style is optional and may be imposed on the user.

[00080]. The interface 3 comprises a microphone 9 capable of capturing the voice signal 5 of the user and, if appropriate, interfaces 10 allowing the user to choose the style of music developed and may include a loudspeaker to accompany, s if there is any, musically the user while he is singing. The interface 3 further comprises a memory capable of storing the ringing 13.

[00081]. Alternatively, the ring 13 can be stored on a terminal different from that used to create the ring. For example the creation of the ringtone can be done from a computer while it will finally be stored on a mobile phone.

[00082]. Moreover, the server 4 comprises an analysis module 15 and a synthesis module 17.

[00083]. The module 15 extracts the analysis parameters, that is to say the physical and musical parameters of the voice signal 5, such as the height extracted for example by autocorrelation or by the algorithm described in FIG. French patent document of France Telecom bearing the national registration number 01 07284, the intensity extracted for example from the energy of the voice signal or by the algorithm described in the French patent document of France Telecom bearing the national registration number 01 07284, the quality detection of the height which characterizes the level of reliability of the estimate of the height extracted for example by the algorithm described in the French patent document of France Telecom carrying the national registration number 01 07284, the attacks extracted for example by a High Frequency Content (HFC) type algorithm using the high-frequency content of the voice signal or from the height detection quality, for example by considering that an attack is a sound whose height is estimated with a poor quality, the vowels characterized by an algorithm using for example the spectral centroids consonants characterized for example by a model combining HFC and spectral centroids, the tone of the voice characterized for example by an algorithm using MFCC (MeI Frequency Cepstral Coefficients) , the various vocal noises like the "beat box" extracted for example by an algorithm using equal MFCC.

[00084]. Intensity is the absolute or normalized intensity. Normalized intensity is an intensity that has been normalized, for example, with respect to the strongest intensity detected in the hummed melody.

[00085]. The module 17 provides the synthesis of the ring, that is to say the correction operations, and / or cleaning, and / or transformation and / or orchestration and / or rhythmic registration of the voice signal 5 depending on the analysis parameters extracted by the module 15 and the musical style chosen by the user or imposed.

[00086]. Thus, when the user hums, the telephone 3 picks up the voice 5 of the user using the microphone 9 and sends it to the server 4 in the form of an audio file 14, for example of the mp3 or wav type.

[00087]. The server 4 receives the file 14 and extracts the analysis parameters 25, such as the pitch, the intensity and / or the attack of the voice signal 5. The analysis parameters extracted from the voice signal 5, as well as the voice signal as such, are then transmitted to the synthesis module 17. [00088]. In addition, parameters 16 of the musical style chosen by the user or imposed, called "style parameters" are transmitted to the synthesis module 17. From these style parameters 16, the synthesis module 17 will establish in particular the rules of musical composition, such as correction rules, and / or cleaning, and / or transformation, and / or orchestration to apply to the voice signal 5 to obtain a music presenting the characteristics of the musical style.

[00089]. According to the extracted analysis parameters and the style parameters 16, the module 17 transforms the voice signal 5 into one or more sound tracks (for example one or more tracks

MIDI derived from the voice played by different virtual instruments, and / or one or more audio tracks derived directly from the voice), which are then, if necessary, mixed together to obtain the ring 13 of the mobile phone.

[00090]. Alternatively, the recording of the file containing the voice signal 5 is performed on the remote server 4, the recording of the voice signal 5 on the server 4 can then be performed in "streaming".

[00091]. In a real-time or streaming operation, the analysis parameters can be extracted instantly, or more precisely at the end of each signal observation window.

[00092]. Alternatively, the signal is processed locally by the mobile phone 3.

[00093]. As a variant, the signal is processed both partially locally (for example for calculating the Fourier transform of the voice signal) and partially on the server, in order to relieve the CPU load of the processing performed on the server. In this case, the voice signal and the result of the FFT or any other calculation made locally are transmitted to the server that uses this data to transform the voice signal into a ring. [00094]. FIG. 2 shows a schematic representation of the module 15 according to the invention which analyzes the voice signal 5 picked up by the microphone 9 (represented in FIG. 4). This module 15 performs a low level analysis of the voice signal 5 via the modules 21 and possibly 22 and possibly a higher level analysis via the module 23.

[00095]. The low-level analysis is performed locally almost instantaneously since it is related to the sound that has just been pronounced or to a short time window and for example 10 ms. While the high-level analysis, performed using the low-level parameters, is a global analysis of the voice signal performed a posteriori over several seconds of the voice signal 5 or even on its whole.

[00096]. More specifically, during the low-level analysis, the module 21 extracts instantaneous parameters such as the pitch and / or the intensity of the voice signal, which in particular allow the module

22 to segment the voice signal into sound events (ie for example to determine the moment at which each note was sung and the duration thereof) and / or to classify these sound events is ie to associate each event to a class (which can be for example the different instruments of a battery or the different humming notes). Voice flow events are objects that have a rhythmic and / or melodic meaning, such as notes, syllables or phonemes.

[00097]. For example, the module 21 only extracts the height which is the only parameter used subsequently by the modules 22 and / or 23. As a variant, the module 21 extracts the pitch and the intensity of the voice signal. The intensity may for example be used to influence the intensity of the sounds to be synthesized.

[00098]. As a variant, the module 21 can extract the spectral parameters of the voice signal (based on a Fourier transform or on an MFCC of this signal) which make it possible in particular to characterize the vocal expressivity (in terms of timbre, phonemes, etc.). . [00099]. In this case, and for example for rhythmic applications, the module 21 does not necessarily extract the height and extracts spectral parameters such as, for example, the high frequency content (HFC) used to identify the attacks and derive a mean rhythm, or the First 13 coefficients of the MFCC (MeI Frequency Cepstral

Coefficients) to vocally control a drum machine (Vocal BeatBoxing).

[000100]. Moreover, from the instantaneous parameters extracted by the module 21, the module 22 realizes a low-level musical description, consisting in segmenting the hummed voice signal into sound events, and / or classifying these sound events, as described for example in the French patent application of registration number 0653557.

[000101]. For this purpose, the module 22 detects the attacks in the voice signal and / or performs a segmentation so as to identify the different events of the voice stream. The notes that have been sung are deduced by the module 22 as a function of the frequencies of the notes of the signal and the position of the measured attacks. The classification consists of determining to which class belongs each of the events. For example, for Vocal BeatBoxing, we associate each sound with one of the percussive instruments of a drums.

[000102]. The module 22 also performs if necessary an analysis of expressiveness of the voice by detecting in particular legato and / or tremolo present in the voice.

[000103]. From the parameters extracted by the modules 21 and if appropriate 22, the module 23 performs a global analysis of the voice signal, said high-level musical description.

[000104]. For this purpose, the module 23 may in particular determine a tempered transposition of the melody ("tuning") and where appropriate the range ("harmony") and / or the tempo ("rhythm") in which the voice signal is hummed . [000105]. For this purpose, the module 23 can perform a rhythmic analysis. The rhythm can be characterized in particular by its tempo, measured in bpm. The tempo is deduced by locating the temporal position of the vocal events (notes, syllables or phonemes). In a typical implementation, the tempo can be extracted from the autocorrelation of the voice signal. In an implementation, the tempo can be extracted by a tracking algorithm like the one proposed by Eric Scheirer (Eric D Scheirer "Tempo and Beat Analysis of Acoustic Music Signals." J. Acoust, Soc., Am., 103 (1), 1998 ).

[000106]. In order to achieve the tuning, the module 23 determines a global transposition for retranscribing the melody optimally in the temperate chromatic range (tuning "The 440") playable by the VST instruments granted in the 440. This optimization depends on criteria as the frequency proximity to the original melody, and can be weighted by other parameters such as intensity. An example of tuning is shown in Figure 11. In the case of a tempered range in Vz tones, it is sufficient to look for the optimal transposition between -1/4 ton and +1/4 ton.

+ Transpo)\ .int ensité(t)^q ,p > 0,q ≥ 0 , p[000107]. For example, we can choose the transposition that allows to modify as little as possible the melodic passages of high intensity, which mathematically amounts to minimizing a "cost of tuning". An example of tuning cost is calculated by the final P formula: Cost (Transpo) =

+ Transpo) \ .int asity (t) ^q , p> 0, q ≥ 0, p

and q real, h (t) being the pitch of the melody at time t, Transpo being the applied pitch variation whose optimum value is sought between -1/4 and +1/4 of a tone for a chromatic scale tempered in Vz tones, where Δ is the instantaneous difference between a height and the nearest tempered height, intensity (t) being the intensity of the melody at time t, where t0 is the beginning of the recording and tfinal is the end of the recording. [000108]. The reference implementation uses this formula with p = 2 and tfinal ² q = 1, which is Cost Trans () = + Transpo) \ .int (t). t

[000109]. It should be noted that the choice of tempered heights (for example the chromatic scale in "The 440") depends on the chosen musical theory, a range in% of tone or variable pitch difference that can for example be chosen for musical theories non-Western.

[000110]. Once the optimal transposition has been carried out, a "tempering" of the optimal transposed melody, consisting in tuning the instantaneous height of the optimal transposed melody to the most probable tempered height is carried out. In a simple example shown in Figure 12, the most likely temperate height is the closest to the melody height.

[000111]. Alternatively, the sung notes can be tempered without optimizing the transposition. For example, each singed note can be transposed or tuned instantly to the nearest or most likely frequency note in the Western chromatic scale.

440Hz ". This instant tuning can be achieved by the module 22.

[000112]. Alternatively, one can implement more sophisticated rules to temper and taking into account for example the melodic profile, the presence of attacks, or the proximity of a temperate note (accuracy of the note). For example, one may decide to tune all the notes of the optimal transposed melody or melody that are within 1/16 ^th of a note just on the nearest note and adopt a more sophisticated strategy. for other notes (false notes). 1/16 ^th of the tone is given here as an example and you can choose any other rule to divide the notes between the note just below and the note just above.

[000113]. For the false notes, one can for example decide to concatenate them to the note of before, that is to say to the note which precedes it temporally, or to the note of after. To concatenate is to replace them with the note before or after. We will decide to concatenate with the note after for example if the false note is preceded by an attack and the note after it is not, which suggests that they constitute one and the same note. It will be decided to concatenate with the note before for example if the false note is not preceded by an attack but is followed by an attack. The other false notes will then be concatenated for example for the first half to the note before and for the second half with the one after.

[000114]. Then, if necessary, the optimal range in which the melody is inscribed is determined from among a set of scales (for example the major and minor diatonic and pentatonic scales).

The inscription of the melody in a range may depend for example on the number of notes of the melody belonging to the scale, their duration, their intensity, and / or the probabilities of notes in the range. An example of determining the optimal range is shown in Figures 13.1 (non-optimal range of A major) and 13.2 (optimal range of A minor). The bold lines represent the notes of the scale while the lean traits represent the notes out of range. Comparing figures 13.1 and 13.2, we notice that the hummed melody corresponds more to notes of the range of A minor than those of A major.

[000115]. By probability of notes in the range, we mean the probability that the note appears in a melody that would be sung in a given range. We then assign to each of the notes of the temperate scale a coefficient which is the probability of the note in the range. The sum of these coefficients must then be normalized to represent a probability in the strict sense that is to say that the total of the coefficients is equal to 1. The probetone (Krusmhansl & Kessler, 1982) provide a good example of probability of the occurrence of a note in a major or minor scale, but one can choose any other distribution of these probabilities, in particular according to the style of music chosen. [000116]. Alternatively, one can increase the coefficient associated with the first and / or last note to increase their weight, because of their high probability of being a fundamental note of the optimal range.

[000117]. For example, one can choose the range which makes it possible to keep as many notes of high intensity in the range as possible, which mathematically amounts to choosing the range that maximizes a harmonic "quality". This quality can be calculated by the formula - Q ^uality ( ^G ) = Σ ^duration ( ^note y ^{Λnt ensited} ( ^note yΛG, noteγ, _P > o, (q, r)> or 'notes'

G being the candidate range, duration (note) being the duration of the note, intensity (note) being the intensity of the note, and 'P (G, note)' being the probability of the note 'note' in the range BOY WUT.

[000118]. The reference implementation uses this formula with p = q = r = 1, and P the probe tone of Krumhansl & Kessler, either

Quality (G) = V duration (note). intity {note) .P (G, note) notes [000119]. As a variant, a range can be imposed for the humped notes without searching for an optimal range. For example, each note singed instantly can be tuned to the nearest frequency note in the range of The Major.

[000120]. Alternatively, the tuning and the determination of the range can be carried out simultaneously, by directly determining the optimal transposition of the original melody in the different candidate ranges.

(eg the major and minor diatonic and pentatonic scales), and choosing the range inducing the best "tuning quality". In this case, the tuning quality is weighted by the probabilities of the range considered.

[000121]. For example, we can choose the melodic transposition and the type of scale that allow to modify as little as possible the notes of great intensity by respecting at best the probabilities of occurrence of the notes in the range, which mathematically amounts to maximizing a "quality". [000122]. This quality can be calculated by the formula:

Quahté (G, Transpo) = \ - -A (h (t) + Transpo) \ mt set (t) ^q P (G, h (t) + Transpo) ^r , p> 0, (q, r)> 0

JA to ^H

, p, q and r real, G being the candidate range, Transpo being the applied pitch variation whose optimal value is sought between -1/4 and +1/4 of a tone for a tempered chromatic range in Vz tones, h (t) being the height of the melody at time t, where Δ is the instantaneous difference between a height and the nearest tempered height, intensity (t) being the intensity of the melody at time t, where t0 is the beginning of the recording and tfinal the end of the recording, and 'P (G, h)' the probability in the G range of the temperate note of pitch closest to the height h.

[000123]. The reference implementation uses this formula with p = 2, q = r = 1, and P the Krumhansl & Kessler probe tone

Quality (G, Transpo) = - -A (h (t) + Transpo) int ensity (t) P (G, h (t) + Transpo). A

⁴ example of tuning taking into account the optimal range is shown in FIG.

[000124]. Once the optimal transposition has been performed, a "tempering" of the optimal transposed melody is performed, which consists in tuning the instantaneous height of the optimal transposed melody to the most probable or closest tempered height in the optimal G range.

In a simple example, the most likely temperate height is the closest frequency to the pitch of the melody without taking into account the optimal G range which was then used only to define the optimal transposition.

[000125]. As a variant, the optimum gamma-gamma can be taken into account in the "tempering" step by relying on the notion of a zone of attraction of a score weighted by its probability of occurrence in the optimal gamma range.

[000126]. Consider a note between do and do # for a range of C major. Integrating the notion of attraction zone of a weighted score by its probability of occurrence amounts to saying that the do (note fundamental) has a greater power of attraction than the do # (secondary altered note), that is to say that we must choose the do more often than if we did not know that we are in range of do, without excluding the do # when we are very close. [000127]. In an implementation, the distance between the note played and each of the two notes with which it is closest is calculated at each instant and this distance divided by the probability of occurrence of the note in the range P (G, h) . We then obtain a notion of weighted distance. Alternatively, one can divide by P (G, h) ^r with r> 0. Then choose the one of the two notes which is at the closest weighted distance.

[000128]. Alternatively, weighted distance tempering strategies may be combined with those taking into account the presence of attacks.

[000129]. Alternatively, the notion of weighted distance can be used to perform the tuning and range determination operation.

(Fig.14). In this case, Δ represents the weighted distance.

[000130]. As a variant, in a so-called real-time implementation, the end of the recording is not expected to perform the high-level musical description 23. It is then possible to carry out this high-level analysis 23 permanently in order to provide at each instant an estimate of the high level parameters (rhythm, tempo, range in particular). At the instant t, the analysis is carried out for example on a window whose size increases and starts from the beginning of the recording and up to t, or on a window of constant size and taking into account the last 5 seconds of singing. This implementation would find a particularly advantageous application in the context of a video game or any other interactive application real time.

[000131]. The set of analysis parameters extracted by the modules 21-23 and possibly the style parameters 16 are sent to the synthesis module 17 represented in FIG. 3. This synthesis module 17 is composed of a MIDI processing module. 26 and / or an audio processing module 27. The analysis parameters 25 as well as that the style parameters 16 are transmitted to these two modules 26, 27. In addition, the voice signal 5 and / or prerecorded instrument loops 53 are applied to the input of the module 27.

[000132]. The transmitted analysis parameters depend on the type of synthesis envisaged. In fact, to simply transcribe the sung voice as a MIDI ringtone, only the beginnings of notes, the durations, the heights and possibly intensities of the sung notes are sent to the module 17. On the other hand, to add a rhythmic loop on the voice, the tempo extracted from the voice is sent to the module 17 in order to stall the rhythmic loop on the voice. While to correct the melody or add a coherent melodic accompaniment, the parameters related to the harmony of the voice signal 5 are further sent so that a synthesized instrument can correspond to this harmony.

[000133]. As a variant and for example in the case where it is desired to capture only rhythmic information, the height is not analyzed, but spectral order information, such as for example the HFC, to identify precisely the attack instants or the first 13 coefficients of the MFCC for a voice control of a battery from a beat boxing.

[000134]. During the MIDI processing performed by the module 26, the speech signal is transformed from the analysis parameters 25 into a raw MIDI track 29 shown in FIG. 5, this track 29 having for the melodic tracks the temperate notes representing the phrase sung determined from the modules 21-22.

[000135]. In a variant, the raw midi track 29 is optimally tempered by integrating the results of the tuning and, if appropriate, the optimum range determined in the module 23 and possibly the style parameters.

[000136]. The raw MIDI track 29 is then used to synthesize all MIDI instrumental tracks, such as the piano, bass, drums, or synthesizer tracks. In other words, in the invention, from a monophonic line (the voice line), one produces a polyphonic ringtone, that is to say a ring which is a superposition of several musical tracks, and poly-instrumental, that is to say that the ring tracks can correspond to different instruments.

[000137]. As shown in Fig. 8, tracks 31-33, 35-38 defining the different musical tracks of the ring may be obtained by processing and / or transforming the raw MIDI track 29 of the voice signal. These treatments and transformations that are part of the musical composition rules depend in particular on the type of instrument to be synthesized and the parameters of style.

[000138]. Treatments can be track cleansing treatments, and / or melodic correction, and / or rhythmic registration. The transformations consist in automatically composing a new melody derived from the sung melody: the choice of the notes and the rhythm is made according to the parameters of analysis, and rules of musical construction which depend on the parameters of style (the rules of composition applied for RnB music being different from the composition rules for jazz music).

[000139]. To carry out the cleaning, we eliminate unwanted notes while keeping the notes of expressiveness if necessary. For example, a note may be considered expressive if it is written in the melodic continuity of the sentence (close to the preceding and following notes) and if it has sufficient intensity to support the notes it accompanies.

[000140]. The cleaning consists for example in applying various treatments to the track 29, such as deleting or grouping the notes whose duration is less than a reference duration (for example 1 ms), and / or the smoothing of the unstable notes, and / or deleting notes whose intensity is below a threshold, and / or deleting notes whose height detection quality is below a threshold. FIG. 6 shows, for example, the processing performed on the raw MIDI track 29 to obtain the track 30 played for example by a piano. [000141]. In order to perform a melodic correction, the pitch of the notes of the raw MIDI track 29 is adjusted according to the supposed harmony, if necessary by preserving all or part of the expressive notes. A sung note that is not in the calculated range is either replaced by the nearest or most probable note of that scale, or kept and highlighted as a note of expressiveness (for example as a trill or for the setting place of a portamento). The correction of the notes can thus be controlled, depending for example on their level of expressiveness, their intensity and their duration. FIG. 6 shows a melodic correction performed on the cleaned track 30 of FIG. 5 to obtain a track 31 recalibrated in a major range. For this purpose, the notes of the track 30 are recalibrated in the optimal range calculated previously.

[000142]. To perform a rhythmic adjustment, the notes detected during the analysis phase are shifted so that their attacking moments respect a given rhythm pattern. This rhythmic pattern can be simply a regular rhythm (for example the notes are positioned on the eighth notes determined according to the tempo extracted by the module

23), or more complex (for example a bossa-nova rhythm), or corresponding to the instants of attack of the sung melody.

[000143]. To perform a transformation (automatic composition of a new melody or a rhythmic track derived from the voice), many methods and algorithms are possible: transposition, extraction of notes on the highlights (identified thanks to the tempo extracted by the module 23), application of rhythmic patterns, derivation of notes of harmony.

[000144]. Thus in one example, the track 33 corresponding to an oem ά. voice is obtained by transposing track 31 three octaves higher. For this purpose, three octaves are added to the pitch parameter contained in the MIDI file of the corrected voice signal.

[000145]. In one example, the track 35 corresponding to a chorus is obtained by taking again the notes of the track 31 on the strongest times. For this, we modify the MIDI track by eliminating all the notes that do not are not on the times, and changing the duration of the remaining notes so that they last the entire time.

[000146]. The track 36 corresponding to a bass is obtained by keeping only the notes on time and alternating fundamental and fifth determined according to the knowledge of harmony. For that, one modifies the MIDI track by eliminating all the notes which are not on the times, and modifying the height of the remaining notes by choosing the note closest to the chord considered (fundamental or fifth).

[000147]. The track 37 corresponding to a second bass more complex than the first is based on the transposition of a riff (repetitive rhythmic-melodic pattern). For this, we modify the MIDI track by eliminating all the notes that are not on the beats (or several times according to the duration of the riff), and by adding the notes of the riff transposed to the height of the remaining notes.

[000148]. The track 38 corresponding to the battery is obtained by setting a rhythmic pattern characteristic of the RnB style on the tempo of the voice (extracted by the module 23). For this, we load a MIDI drum track characteristic of the style, which is repeated a sufficient number of times to cover the entire melody, and we add additional notes to the expressive moments of the voice.

[000149]. Complementary percussive sounds (for example cymbal strokes) may be added depending on the expressiveness of the voice (analysis parameters extracted by the modules 22-23), for example on notes of expressiveness.

[000150]. It should be noted that the consistency in the arrangement of the tracks is guaranteed by the analysis parameters 25 which make it possible in particular to ensure that all the tracks are in the same tempo and / or in the same harmony.

[000151]. Each MIDI track of the ringtone can be created from the raw MIDI track of the voice signal being processed and / or transformed, or from a MIDI track deriving from the raw MIDI track (ie a MIDI track already processed and / or transformed), or from the analysis parameters and pre-existing patterns. For example, a violin track can be derived directly from the piano track, and an RnB drum track can be built directly from the tempo and a rhythmic pattern.

RnB.

[000152]. In addition, MIDI-specific effects, such as dubbing notes, arpegiation, delaying, can be applied to the resulting tracks 30, 31, 33, 35-38. These effects are applied according to the musical style chosen by the user or imposed and / or the tone of the voice of the user. For example, it will be possible to transform the notes of a track into chords played in arpeggios, these arpeggios being played according to the harmony previously determined and the speed of speech. In another example effect, all expressiveness notes can be transformed into trills.

[000153]. In addition, an original melody can be derived from the voice, not to substitute, but to continue it. For example, the system can compose a response to the vocal melody (for a dialogue with the machine), or a sequence (to finish the melody correctly according to the chosen style). For example, if the bass line alternates a fundamental and a fifth, the fifth introducing a musical tension and the fundamental closing this tension, and ending on a fifth, the process will continue the musical phrase by introducing a fundamental to close the musical tension. The hummed melody can also be sent with its analysis parameters including the harmonic and / or rhythmic parameters to an automatic composition software, so that this software composes an end or an answer.

[000154]. Once obtained, the MIDI tracks 30, 31, 33, 35-38 are integrated into a single MIDI file that can be played by any MIDI-compliant material. This file can be sent to the phone or any other device or software as a MIDI ringtone, these devices with basic MIDI instruments to play the MIDI tracks it contains.

[000155]. As a variant, the MIDI track or tracks 30, 31, 33, 35-38 are transmitted to a module 43 comprising virtual instruments, for example of the VST type, which make it possible to synthesize the audio tracks associated with each instrument. For example, the MIDI bass track is sent to the VST instrument named "Virtual Bass" which turns this symbolic MIDI track into a bass audio track. All audio tracks synthesized from the MIDI tracks are then sent to a module 47 which mixes the generated musical tracks.

[000156]. Furthermore, the module 27 develops tracks 49-51 of audio type, derived from the direct transformation of the voice signal and / or pre-recorded sounds, without passing through a symbolic representation of MIDI type. These tracks 49-51 are obtained by transforming the voice signal and / or by sound synthesis from the extracted analysis parameters and pre-recorded sounds.

[000157]. The transformations of the voice can be either corrective transformations (melodic and / or rhythmic), or timbral transformations, or these two types of transformations successively applied.

[000158]. The melodic correction can be performed for example by a so-called "pitch shifting" technique in which the notes of the voice are recalibrated on the temperate notes or of the range, optimal or not, determined by the analysis module 15, or on the notes of a melody derived from the voice signal.

[000159]. The rhythmic correction of the voice consists of a rhythmic synchronization of the voice on a rhythm of reference. This rhythmic correction can be performed by rhythmically stalling the voice on an imposed rhythm (for example a fixed tempo). The timing is done by "Time stretching", changing the duration of sound passages that are not in rhythm to stall on the imposed rhythm. This rhythm can be determined by the parameters of analysis or by an imposed rhythmic pattern.

[000160]. Alternatively, this rhythmic correction may consist of a rhythmic synchronization on the rhythm of the voice, on which are instrumental tracks. In one example, a rhythm marking is carried out by the "Time Warping" technique in which the strong points of the voice signal represented by the vertical lines 55 in FIG. 9 are recorded in order to construct a reference rhythm on which all the music tracks are synchronized. This reference rate will then be part of the analysis parameters 25 and may be used to build the midi and / or audio tracks. This reference rhythm can also be used as the input of an automatic composition software that will automatically generate a melody whose rhythm will coincide with this reference rhythm.

[000161]. The timbral transformations are based on the modification of the tone of the voice obtained by applying filters on the voice signal. This modification of the timbre can be performed by transforming the voice signal 5 into another realist voice or not, such as for example an ogre voice or monster, using for example a vocoder phase. The transformed voice can remain intelligible or be transformed into a musical instrument or its unintelligible. The voice can also be transformed into a sound environment (sound sheet obtained for example by concatenative synthesis from various sounds, musical or not).

[000162]. Furthermore, audio tracks can be obtained by selecting notes of the voice signal and emphasizing them according to the expressiveness of the voice or the parameters of analysis 25, and / or by using pre-recorded rhythm loops that the We hold on the measured (and if necessary modified) tempo of the voice signal.

[000163]. In one implementation, the voice is used to select a sound object in a music database, in order to compose one or more audio tracks. For this purpose, the analysis parameters, such as height, intensity and attack, are extracted from samples and stores these musical samples and associated analysis parameters in the database. Then, analysis parameters of the voice signal are extracted over a given period of time and the musical samples having the closest analysis parameters of the voice signal over the time interval are selected in the database. given. For example, to generate the drum track, one can search for the rhythmic RnB corresponding best to the rhythm expressed by the voice.

[000164]. Musical samples can have multiple levels of granularity. Indeed, musical samples can be small musical objects, such as a note of a synthesized instrument, or large musical objects, such as musical loops. The analysis parameters will be extracted over a larger period and in greater numbers in the case of a selection of large musical objects.

[000165]. The audio track or tracks 49-51 thus derived from the voice signal are transmitted to the mixing module 47.

[000166]. The mixing module 47 then performs a sound processing and mixing audio tracks 30 ', 31', 33 ', 35'-38, 49-51' from MIDI processing and / or audio processing. For this purpose, the module 47 adjusts the volumes of the different tracks relative to each other and / or introduces individual effects on each track (such as saturation or sound compression). The module 47 then realizes a single audio output track, mixed from all the individual instrumental tracks. The module 47 is also able to apply global sound effects on this output track (for example a global reverberation or any room acoustics effect such as for example a "church effect").

[000167]. Alternatively, the user can change the original mix by choosing for example to increase the volume on the bass or reduce it on another instrument. [000168]. The sound signal 59 from the mix is then input to a module 60 which transforms it into an audio format readable by the user's phone (eg mp3). At the output of the module 60, there is obtained a ring 61 which is recorded in a memory of the mobile phone.

[000169]. Alternatively, the voice signal can also be transmitted as such to the mixing module 47 to be injected into the ring and then be considered an audio track.

[000170]. Figure 15 shows a simplified processing chain of the invention. The analysis module is limited to module 21 which extracts only the pitch and intensity parameters to generate the raw MIDI track 29. Only two coherent MIDI tracks are derived: track 1 which is the cleaned raw MIDI track, and the track 2 playing this same track an octave higher. [000171]. The invention also relates to mobile telephones comprising a memory zone associated with each of the contacts of the telephone, this memory zone being intended to store a ring produced using the method according to the invention.

[000172]. Thus the user or a contact, can record in the phone a sung phrase having previously selected a style of music. For example, a contact named Ursula records a sung phrase like "hello is Ursulaaaϋ" by selecting a rock style.

[000173]. The method according to the invention described above then transforms the sung phrase into a ring.

[000174]. The user then associates the ringtone obtained with the contact for which it is intended, by storing it in the memory zone associated with this contact.

[000175]. So when the contact will call, the phone will play the ringtone corresponding to that contact. [000176]. The invention finds a particularly advantageous application for the development of a ringtone of a mobile phone. However, the invention could also be used more generally in the field of music creation. By extension, we call "ringtone" any musical object having a melodic and / or rhythmic coherence, and containing or not the voice. Thus an instrumental riff, a jingle, a telephone ring, an answering machine message are considered as ringtones within the meaning of the invention.

[000177]. In addition, it is possible to consider an application called "Voxgroovebox" in which a musical phrase is recorded by pressing a button on the telephone keypad, and as soon as the button is released, the ringtone generated by the method according to the invention from the musical phrase begins to loop on itself. Then, it is possible to sing on this ringtone, or to mix it with pieces of music stored in the phone. Depending on the musical intention, you can activate all or part of the tracks: rhythmic only, bass only, lead alone, two of these tracks or all three.

[000178]. It is also possible to consider an application called "Answering machine augmented" in which one records, declaims or sings an answering machine, and the system automatically performs the correction and / or musical accompaniment of this message. In one example, the voice signal is transcribed alone but it is musically coherent because it has been re-calibrated at a given rate.

[000179]. In another example, an intelligible voice is retained and an average tempo or a reference rate of the telephone message is determined. Then add to the voice signal an instrumental loop, consisting of one or more tracks and whose tempo or rhythm corresponds to the one that was analyzed.

[000180]. In a variant, the user has the possibility of modifying each of the tracks. In this case, the user receives for example a poly-instrumental score corresponding to the re-orchestrated melody of what he has hummed or a simplified symbolic representation of the score. He can then read each of the tracks, for example using a partitions and / or manually modify each of the tracks. It can for example manually modify the rhythm, the notes, or any parameter of analysis. Any interface and for example a stylus or buttons can be used to edit tracks and / or remix.

[000181]. Alternatively, the composition may be performed in several steps or passes. The user begins, for example, by vocally dialing the humming melodic lead. In this case, only one track is generated: the melodic lead corrected from the analysis parameters. He hums then for example the bass line. To help him hum in rhythm and harmony, he will be able to hum the bass line while listening to the melodic lead. The hummed bass line will then be corrected by applying the parameters of analysis of the melodic lead. He could also choose to apply to each of the two voices a mixture of parameters from the two humps. For example, the rhythm could come from the bass line and the harmony of the melodic line. In another example we will compute a global harmony resulting from the two humps and calculated by applying the harmony analysis algorithms to a voice recording file consisting of the temporal juxtaposition of the two voice recordings (ie say by temporally juxtaposing the two files to create a single file with a length equal to the sum of their two lengths).

[000182]. An implementation of the invention is described below for extracting a melody from a voice signal 5.

[000183]. Firstly, an algorithm is implemented that makes it possible to obtain the instantaneous height, energy and quality fluxes of the voice signal 5, for example using the algorithm described in France Telecom's French patent document. bearing the national registration number 01 07284.

[000184]. Preferably, the stream of the instantaneous notes of the voice signal 5 is cleaned by means of a median filter, eliminating out-of-range notes, and / or low energy notes, and / or low quality notes, and / or or notes that are too short (noisy), for example for segments shorter than 50ms. [000185]. Segmentation of the voice signal 5 is then performed, each segment of the signal corresponding to a piece of voice signal between two attacks corresponding to significant variations in pitch and / or energy of the voice signal and / or energy in the high frequencies, which can for example be detected by implementing a technique of HFC (High Frequency Content) type, described in application PCT / FR2007 / 051807. Then, the signal is preferably reset to zero from the instant of the first detected segment.

[000186]. The height of each segment S1-S8 corresponding to the median height of each segment is determined. Thus, in FIGS. 16, 18, 19 and 20, the segments S1-S8 are separated by a large variation in height.

[000187]. This is followed by a global tuning of the voice signal 5. This tuning consists in determining the note, called "central note" NC, which is the most frequent note (the most played on 100 levels of overall tuning), starting from non-tempered notes of each segment S1-S8, then to globally transpose the whole of the voice signal 5, so as to match the "central note" NC to its closest temperate note.

[000188]. To this end, we consider the melodic flow, ie the flow of the set of heights h of sung notes, as shown in Figure 16. We divide the height axis into B bands corresponding to the notes of the temperature range, so as to obtain a tuning grid G placed on the voice signal 5 built in intervals of 1/2 tones.

[000189]. A histogram is made of the duration t of the sung notes as a function of the heights h of all the notes of the signal 5, in these bands, as shown in FIG. 17. The central note NC is the average of the heights of the band. highest of the histogram

(called mode). [000190]. One then seeks the transposition of the bands (between -1/2 and +1/2 ton) making it possible to obtain an optimal histogram, in which the mode is maximum, in other words the mode in which the central note NC is the most predominant. This optimal transposition corresponds to the overall tuning of the melody. The whole of the voice signal 5 is then transposed by this optimum value so that the central note NC is transposed to a moderate height. For example, in FIG. 18, the transposition makes it possible to align the central note CN of initial height 54.2 with the nearest tempered note of height 54.0.

[000191]. A tempering of the voice signal is then performed which consists in assigning for each segment S1-S8 of the voice signal 5 a tempered height, starting from the height actually sung. For this, we consider the tuned melody, its central note NC, and the 128 temperate notes corresponding to the 128 Midi notes.

[000192]. For each of these 128 Midi notes, we create a model of the actually sung height (more precisely the height difference between the middle note and the sung notes). In the initial model, each Midi note is modeled by a height of the chromatic scale tuned to the central note. Thus, as represented in FIG. 18, the initial model of the Midi notes corresponds to the grid G of notes at constant intervals e placed on the voice signal 5 centered with respect to the central pitch note 54.0, the initial model of the note. Corresponding Midi 54 in our example at height 54.0, the initial pattern of note Midi 55 at height 55.0, and so on.

[000193]. The height of the NC central note remaining invariable, it gradually adjusts the model of each tempered note with respect to the heights actually sung.

[000194]. Preferably, one starts by modeling the heights of the last sung notes, then one refines the model by going back to the beginning of the melody according to the direction of analysis SA. This technique is based on the assumption that the accuracy of the song improves during the melody (the voice is placed), and that the height of the last sung notes is more stable: it allows to use a robust model for determine the pitch of the notes, generally imprecise, sung at the beginning of the melody.

[000195]. The assignment of a tempered note for each segment S1-S8 follows the following algorithm: in a loop, for each segment Sj, starting preferably from the last to the first one, is determined, starting from the current model, the temperate rating most likely for segment Sj. For example, we consider that the most likely temperate note is the tempered note whose model is closest to the sung note, - we assign this temperate note to the segment Sj, and

- we update the model of this note Midi tempered, taking into account the pitch sung on the segment Sj: the height of the model of this temperate note evolves. For this purpose, the model of the tempered note may for example be equal to the average of the heights of the actual scores (sung) to which the score of the model has been assigned, these note pitches being able to be weighted if necessary by their duration or their duration. intensity.

[000196]. For example, at the beginning, the model of the tempered note 54 corresponds to the height 54.0 (the central note) and the model of the tempered note 55 corresponds to the height 55.0. If an analyzed segment Sj has a pitch of 54.2, it will be associated with the tempered note 54, since the difference between the analyzed height 54.2 and the model of the note 54 of height 54.0 is smaller than the difference between height analyzed 54.2 and the model of the note 55 of height 55.0 (54.2-54.0 = 0.2 smaller than 55.0-54.2 = 0.8). Then, we update the model of the note 54 which is equal to the average m of the notes to which we assigned the tempered note 54, namely in the example m = (54.0 + 54.2) /2=54.1. The model of the tempered note 54 thus corresponds to the height 54.1, whereas it corresponded to the height 54.0 before analysis of the segment Sj. We start again for a new segment.

[000197]. Continuing the same example (Figure 19), the model of the tempered note 57 corresponds to the note 57.0 and the model of the tempered note 56 corresponds to the note 56.0. If a new segment analyzed Sk has a note pitch of 56.8, it will be associated with the tempered note 57, since the difference between the analyzed height 56.8 and the 57.0 pitch 57.0 model is smaller than the difference between the analyzed height 56.8 and the model of the note 56 of height 56.0 (57.0-56.8 = 0.2 smaller than 56.8-56.0 = 0.8). Then, we update the model of the note 57 which is equal to the average m of the notes to which we have assigned the tempered note 57, namely in the example m = (57.0 + 56.8) /2=56.9. The model of the tempered note 57 now corresponds to the height 56.9, whereas it corresponded to the height 57.0 before analysis of the segment Sk. It then starts again for the following segment, and so on until analysis of the whole segments.

[000198]. Thus, as shown in FIG. 20, for the processed portion 61 of the end of the signal 5, the height of the pattern of the note 54 rises and the height of the pattern of the note 57 goes down to the actually sung heights.

[000199]. At the end of the algorithm, a tempered note was assigned to each S1-S8 segment. We have thus defined a scale G 'whose gaps e' height no longer correspond to semitones, but to the heights actually sung, as shown in Figure 20.

[000200]. The note pattern has been described for MIDI notes but it is clear that it can be set for any type of temperate note that can be associated with a signal pitch.

Claims

A method for automatically dialing a ring (13) from a recording of a monophonic voice signal (5), such as a song, a user humming, wherein: extracting analysis parameters (25) of the voice signal (5), such as the pitch and / or the intensity and / or the attack of the notes of the voice signal, and the voice signal (5) is transformed into a ring comprising at least one musical track, characterized in that, to transform the voice signal (5) into a ring, the voice signal is tuned by transposing the set of said voice signal (5) to the same height so as to minimize a distance between the whole of the voice signal (5) and a tempered chromatic range, and - moderates the voice signal (5) by replacing the notes of the transposed voice signal with temperate notes. 2. Method according to claim 1, characterized in that, in order to tune the voice signal, a central note (NC) of the voice signal (5) corresponding to the most frequent note of this voice signal is determined, and - The whole of the voice signal is transposed globally so as to match the height of the central note (NC) to the height of its closest temperate note. 3. Method according to claim 1 or 2, characterized in that for tempering the voice signal (5), the voice signal (5) having been previously cut into segments (S1-S8), for each temperate note, a model of associated note is defined; in loop, for each segment (Sj), the tempered note whose model is closest to the height of the segment (Sj) is determined, said tempered note is assigned to the segment (Sj), and the model of said temperate note is updated by taking into account the height of the segment (Sj), for example by averaging the heights of the notes of the signal which have been associated with this tempered note, these heights of notes may be weighted, where appropriate, by their duration or intensity. 4. Method according to claims 2 and 3, characterized in that, in the initial note template, each tempered note is modeled by a height of the chromatic scale tuned to the central note. 5. Method according to claim 3 or 4, characterized in that to temper, we start from the last segment of the voice signal (5) from a time point of view and go back to the first. 6. Method according to one of claims 1 to 5, characterized in that for transposing the voice signal (5), a tuning cost is calculated which is equal to the integration, over the duration of the signal melody. of voice, the product of the instantaneous difference between the pitch of the voice signal (5) and the nearest temperate height raised to the power p (p strictly positive real) and the loudness of the voice signal raised to the power q (real positive q), and the voice signal (5) is transposed so as to minimize the value of the tuning cost. 7. Method according to claim 6, characterized in that p is 2 and q is 1. 8. Method according to one of claims 1 to 7, characterized in that a range is determined by implementing the following steps: - we choose probabilities of occurrence of a note in a given range, the degree of belonging of the melody to several scales is calculated, this degree of membership being a function of the concordance of the notes of the voice signal and the probabilities of occurrence of the notes of the scale, and - the range with the highest degree of membership is selected. 9. A method according to claim 8, characterized in that the degree of membership of the notes of the range is equal to the sum for all the notes of the voice signal of the product of the duration of each note raised to the power p by l intensity of each note raised to the power q and by the probability of occurrence of each note raised to the power r, pq and r being realities greater than 0 and p being different from 0. 10. Method according to one of claims 1 to 9 characterized in that to temper the melody of the voice signal, - one chooses probabilities of occurrence of a note in a given range, an optimal transposition is calculated for each of the candidate ranges which is a function of the concordance of the temperate notes closest to the voice signal and the probabilities of occurrence of the notes of the candidate scale, by calculating a degree of membership of the melody to each candidate range and for all the possible values of the transposition, selecting the range having the highest degree of membership and transposing the notes of the voice signal of the optimal transposition associated with this range, and - Tempers by replacing the notes of the voice signal transposed by the temperate notes closest to the transposed voice signal. 11. Method according to claim 10, characterized in that the degree of membership of the notes of the scale is equal to the integral for all the notes of the melody of the product of the intensity of each note raised to the power q by the difference of the note of the voice signal relative to the note situated between the two temperate notes closest to the power p by the probability of occurrence of each note raised to the power r, pq and r being higher realities at 0 and p being different from 0. 12. The method of claim 10 or 11, characterized in that one uses the knowledge of the range to temper by minimizing a distance between the note of the transposed voice signal and the most temperate note. near, this distance being weighted according to the probability of occurrence of the note in the range. 13. Method according to one of claims 8 to 12, characterized in that the probability of occurrence of a note in a given range is determined from the probetones of Krumhansl & Kessler. 14. Method according to one of claims 1 to 13, characterized in that to turn the voice signal (5) into a ring, it removes or groups notes whose duration is less than a reference value, for example 1 ms , and / or whose intensity is less than a reference intensity and / or whose height extraction quality is less than a reference value. 15. Method according to one of claims 1 to 14, characterized in that to transform the voice signal into a ring, the voice signal (5) or a musical track already obtained from the voice signal is corrected by cleaning it and / or correcting the melody and / or recaling it rhythmically and / or deriving a melody from the voice signal (5). 16. The method of claim 15, characterized in that to rhythmically recalibrate the voice signal (5), it tracks the tempo of this voice signal (5) and recalibrates the notes of the voice signal in this tempo. 17. The method of claim 15 or 16, characterized in that the rhythmic correction is performed by imposing a fixed tempo, for example the average tempo extracted from the voice signal, the voice signal (5) being set rhythmically on the imposed tempo. by a method of "Time stretching". 18. A method according to one of claims 15 to 17, characterized in that the rhythmic correction is performed by a rhythmic marking by the technique of "Time Warping" in which a tracing of the high points of the voice signal (5) is carried out. , in order to build a reference rhythm on which the musical tracks are synchronized. 19. A method according to one of claims 15 to 18, characterized in that the melodic correction is performed by a technique of "pitch shifting" in which the notes of the voice (5) are recalibrated in temperate notes and / or in the average range of the hummed voice signal. 20. Method according to one of claims 15 to 19, characterized in that to compose a new melody derived from the voice signal, the notes of the voice signal (5) and a rhythm are selected according to the analysis parameters ( 25) excerpts and musical construction rules that depend on a musical style chosen by the user or imposed. 21. Method according to one of claims 15 to 20, characterized in that to compose a new melody deriving from the voice signal, an original melody is elaborated which is a response to the vocal melody and / or to the vocal rhythm calculated from Analysis parameters to establish a dialogue with a machine, or a sequence to end the vocal melody correctly according to the chosen style. 22. Method according to one of claims 1 to 21, characterized in that one carries out a modification of the tone, and / or the height and / or other characteristics of the voice by transformation of the voice signal (5), or by sound environment synthesis from the voice signal (5). 23. Method according to one of claims 1 to 22, characterized in that to develop one or more musical tracks, one or more pre-recorded rhythm loops in the form of an audio signal that is staggered on the tempo extracted from the signal of voice (5). 24. Method according to one of claims 1 to 23, characterized in that to create one or more musical tracks, one selects in a musical database musical samples having the musical parameters closest to those of the voice signal on a given time interval. 25. Method according to one of claims 1 to 24, characterized in that the volumes of the different musical tracks are adjusted with respect to each other, and / or are introduced effects on selected tracks, such as saturation or a sound compression effect, and if necessary mixes all the tracks into an output track, and / or we introduce global effects on this output track, such as reverb. 26. The method of claim 25, characterized in that one creates an audio file from the mixed output track, in an mp3 type format. 27. Method according to one of claims 1 to 26, characterized in that the ringing comprises several musical tracks arranged between them according to the extracted analysis parameters and musical composition rules. 28. Method according to one of claims 1 to 27, characterized in that the musical composition rules are related to a style (6) of music, such as a rock or blues style, chosen by the user. 29. Method according to one of claims 1 to 28, characterized in that the voice signal (5) is streamed to a server (4) providing the extraction of analysis parameters (25) and the elaboration the bell (13). 30. Method according to one of claims 1 to 29, characterized in that the musical tracks are obtained from a MIDI and / or audio processing of the voice signal (5). Mobile phone implementing the method according to one of claims 1 to 30. 32. A method for associating telephone ringtones with contacts stored in a mobile phone in which: - we record a phrase sung by the user of the mobile phone or by one of the contacts, the sung phrase is converted into a ring signal using the method defined in one of claims 1 to 30, and the ringing obtained is stored in a memory associated with the contact for which the ring is intended, so that when the contact calls, the corresponding ringing is played by the phone. 33. Device for generating real-time music implementing the method defined according to one of claims 1 to 30 for generating a ringtone from a sung musical phrase, this device comprising means for playing this ringing ring, so that it is possible to sing on this ringtone, or mix it with music tracks to create musical tracks. 34. A method for producing a ringtone in which several voice lines are recorded one after the other, listening to the voice lines previously recorded and processed according to the method defined by one of the claims 1 to 30 being authorized during the recording of a voice. a new voice line and analysis parameters (25) that can be extracted from each record or from all records.