FR2796486A1 - Chosen word replacement speech editing technique having audio signal statistical word model memory with trial input compared and where match found innocuous word replaced. - Google Patents

Abstract

The speech processing technique recognises incoming words (12) using a trial section input (18). Where a match is found, words are edited (24) and replaced by innocuous words.

Description

 <B> METHODS AND DEVICES FOR SUBSTITUTING </ B> A <B> VOICE </ B> SYNTHESIZED DYNAMICALLY <B> TO IDENTIFIED VOCABULARS </ B> <B> AUTOMATICALLY </ B> Field of the Invention The present invention relates generally to the processing of audio signals and, more particularly, to methods and apparatus for performing dynamic recognition and editing of portions of an audio signal, containing one or more words associated with one or more selected vocabularies, using speech recognition and audio editing techniques, respectively.

BACKGROUND OF THE INVENTION As a result of the official regulation on radio waves, broadcasters must frequently pre-screen or pre-filter films and other entertainment products to find out if the content is acceptable, for example if it is found there. blasphemy, slang, racial or other insinuations or insults, sexual or violent language, etc. Frequently, broadcasters "hide with a bpe" such a lexical term, ie transmit a noisy sound simultaneously to, or instead of the word or words unacceptable (s), so that the word or words words are not heard by an auditor. In the case of live transmission, delays are frequently used to allow the broadcaster to conceal the unacceptable material before transmitting it. Another approach is to completely mute the sound. Other techniques include alternative instantaneous sound tracks using the actors' original track, or even using new actors to practice vocal dubbing.

However none of these conventional approaches is automatic in nature or works unattended. That is, in cases where there is a hiding or a deafening, a broadcaster must actively participate in the filtering process during transmissions or live broadcasts, monitoring the material to be transmitted and then taking a certain disposition to launch the occultation or the deafening when it hears an unacceptable lexical term. Disadvantageously, the approach requires immediate human supervision and manual response when such material is detected. If the person is momentarily distracted or simply not paying enough attention, the unacceptable material, which otherwise would have been "obscured", can be truly transmitted. In addition, the creation of an alternative sound track with voice dubs also needs to be monitored and is not automatic either. Also, none of these conventional approaches are able to function satisfactorily in real time. The delays used when you have live broadcasts are of the order of several seconds or more.

Efforts have been made to develop products that allow parents to automatically control the lexical term they consider inappropriate for their children. Blocking software for the internet and so-called "V-chip" are examples of such products. However, these products are all text-based and, therefore, do not provide, among other things, automatic detection and / or editing of an unacceptable lexical term based on the speech signals.

Therefore, it would be desirable to have methods and devices for unattended and automatic detection and editing of an unacceptable lexical term contained in an input speech signal such as, for example, a speech signal related to a transmission. resignation.

Summary of the Invention The present invention provides methods and devices for automatic and unattended detection and editing of a selected lexical term contained in an input speech signal. The term lexical may be an unacceptable lexical term such as, for example, blasphemy, slang, insults, racial or other insinuations or insults, sexual or violent language, etc. The unacceptable lexical term can be defined by standards of lexical terms, which emanate from the official authorities. According to a broad aspect of the invention, a method of dynamically processing a voice signal firstly comprises the automatic detection that part of the speech signal comprises a lexical term associated with one or more vocabularies of lexical terms. The portion of the speech signal is then automatically edited when such a lexical term is detected, such that the detected lexical term is not audibly perceptible. Automatic detection can be accomplished using speech recognition techniques and preferably using well-known word editing techniques. It should be obvious that the term "voice signal" or "audio signal" as used herein is intended to generally include one or more forms of an audio lexical term, for example, but not limited to spoken words, associated with individual speeches, dialogues, music, video, etc.

According to a first embodiment, the automatic editing step comprises the automatic replacement of the portion of the speech signal containing the lexical term, with a part of the modified speech signal, having at least one voice parameter (for example associated with an excitation of the voice zone, a voice zone filter, etc.) which is a modification of at least one voice parameter in the part of the speech signal containing the lexical term. According to a second embodiment of the invention, the automatic editing step includes automatic muting of the part of the speech signal containing the lexical term. According to a third embodiment of the invention, the automatic editing step comprises the automatic replacement of the voice signal part containing the lexical term by an acoustic signal (for example, a bpe).

It should be noted that the automatic detection and editing methods and apparatus of the invention may preferably be used by broadcasters in connection with a live broadcast signal, for example, broadcast signals. television and radio show. Such features, automatic detection and editing, can be implemented to prevent an unacceptable lexical term from being heard by the audience of the program, according to the standards of lexical terms officially imposed. However, one of ordinary skill in the art will consider many and varied applications for such inventive teachings. For example, the invention may be implemented in connection with a box to be connected. In this embodiment, a parent or other user may use the invention to help control the voice-based material that he considers inappropriate for their children. Other objects, features and advantages of the present invention will become apparent upon reading the following detailed description of illustrative embodiments thereof, in conjunction with the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram of a dynamic lexical word recognition and editing system, in accordance with an illustrative embodiment of the present invention; Fig. 2 is a block diagram of a word-correcting speech recognition system according to an illustrative embodiment of the present invention; Fig. 3 is a block diagram of an audio editing system of lexical terms according to an illustrative embodiment of the present invention; FIG. 4 illustrates the sequence of steps of a method of dynamic recognition and edition of lexical terms according to an illustrative embodiment of the present invention; and Figs. 5A and 5B are respective block diagrams of lexical word audio editing systems, according to two other illustrative embodiments of the present invention. Detailed Description of the Preferred Realization Modes.

The present invention will be explained below in the illustrative context of an implementation with a transmission transmission signal. However, it is obvious that the present invention is not limited to such a particular application, nor to such a structural embodiment. Rather than this, the invention is more generally applicable to any situation in which it appears desirable to automatically detect and edit a voice signal which comprises a lexical term which substantially coincides with a lexical term found in one or more vocabularies. previously stored targets, such that the detected lexical term is not audibly perceptible to a potential listener.

Referring initially to Figure 1, there is shown a block diagram of a dynamic lexical recognition and editing system according to an illustrative embodiment of the present invention. The illustrative system 10 receives an input signal 12 to be transmitted. It is obvious that the signal can come from a broadcast studio, for example from a radio or television station. The broadcaster wishes to edit the signal automatically, in order to remove the unacceptable lexical term that may be included in the signal, before it is transmitted, published or otherwise. The system 10 comprises three distinct signal processing components: (i) an intermediate audio and video system including an audio buffer circuit 14 and a video buffer circuit 16; (ii) a word retouch speech recognition system including a training data memory 18, a statistical word template memory 20 and a word retouch processor 22; (iii) a system for audio editing of lexical terms including an audio editing processor 24. After performing detection and automatic editing according to the invention, as will be explained in detail below, the system 10 sends a edited signal 26 which is now advantageously free of any unacceptable audio lexical term.

It should be noted that the term "processor" as used herein is intended to include any processing device such as, for example, a device comprising a CPU (Central Processing Unit). For example, the processor may be a digital signal processor, as is known in the art. Also, the term "processor" can refer to one or more individual processors. The term "memory", as used herein, is intended to include a memory associated with a processor or a CPU, such as, for example, a RAM, a ROM, a fixed memory device (for example a hard disk) , a removable memory device (for example, a floppy disk), etc. Correspondingly, computer software, including instructions or code for executing the methodologies of the invention, as described herein, may be stored in one or more of the associated memory devices (for example, a ROM, a fixed memory or removable) and, when ready to use, be partially or fully loaded (for example, into RAM) and executed by a CPU. In any case, it should be obvious that the elements illustrated in the figures can be implemented in various forms of hardware, software or combinations thereof, for example one or more digital signal processors with associated memories, a or application-specific integrated circuits (ASICs), functional circuitry, one or more general-purpose digital computers appropriately programmed, having associated memory, etc. Given the teachings of the invention set forth herein, those of ordinary skill in the art will be able to consider other implementations of the elements of the invention.

The audio and video intermediate system, which includes an audio buffer circuit 14 and a video buffer circuit 16, is preferably a digital memory buffer which stores the incoming signal to be transmitted in subsequent detection and editing processes according to the present invention. 'invention. In addition, as will be explained, the contents of the buffer may be accessed, on instruction from the word retouching processor 22, for use in performing the audio editing functions. Since the operation of the invention is preferably in real time, the total size of the buffer need not be large. For example, the buffering subsystem may have a total memory capacity of about one second of audio and video signal storage. For a quality audio CD (CD) of this duration, you need about 86 kilobytes of memory. Similarly, one second of quality video can be stored in several megabytes of memory.

It is obvious that for the illustrative embodiments of the invention, no video editing is performed. Instead, the video signal is stored in the video buffer 16 to maintain synchronization with the video signal which is subsequently analyzed and possibly edited. It should be noted that although this is not shown in FIG. 1, conventional methods of separating the original audio and video signals for storage and recombination of the audio signal (possibly edited) and the original video signal can be used. In addition, for radio transmission signals, for example, the video buffer is unnecessary, since there is no associated video signal.

It is also evident that although a buffering system is used in the illustrative embodiments, this is not necessarily necessary. That is, the word editing system and the audio editing system can perform their functions directly on the voice signals received from their sources. However, depending on the type of audio editing operation used, it may be preferable to buffer the signals to satisfy the processing time.

The word retentive speech recognition system which includes a training data memory 18, a statistical word template memory 20 and a word retouch processor 22, is a voice recognition system which is preferably trained to mark. the beginnings and ends of a predefined vocabulary. It should be noted that there are numerous examples of conventional voice identifiers, known to those skilled in the art, that use this recognition mode, called word editing. For example, such a system, as described in J. G. Wilpon et al., "Automatic Recognition of Keywords in Constrained Speech Using Hidden Markov Models," IEEE Trans. Acoust. Speech Signal Processing, Vol. 11, pages 1870-1878, November 1990, can be used. FIG. 2 illustrates an illustrative word-processing voice recognition system for use in accordance with the present invention.

Referring now to FIG. 2 the word retouching processor 22 is described in more detail as comprising a feature extraction subsystem 28 and a pattern recognition subsystem 30. As is known in the art, the subsystem feature extraction 28 samples the voice signal and subdivides the signal into frames, having a duration for example of about 10 milliseconds, so that each frame can be treated discretely by the rest of the system. The feature extraction subsystem then proceeds to extract the spectral characteristics from the signal for each frame. The spectral characteristics are in the form of feature vectors (signals), which are then passed over the pattern recognition subsystem 30. As is known, the pattern recognition subsystem 30 uses statistical words (from a model memory 20) for calculating the probability values associated with the input speech signal. The voice recognition subsystem preferably comprises a plurality of word patterns that are passed to training using a large number of pronunciations representative of the target vocabulary. The target vocabulary used for the training can be stored in a learning data memory 18. It should be noted that the target vocabulary is the group of words or sentences that must be avoided from being audibly perceptible to an auditor of a show. For example, it may include words or phrases that are considered unacceptable. You can also select more than one target vocabulary. Target vocabularies can be generated in advance by broadcasters, based on official standards or other standards. Also, in an application using a plugged-in box, a parent or other user can generate one or more custom vocabularies to be used with the system 10. Also, the target vocabularies can be selected according to flags that can be set by the user. The user, as described in the US patent application, identified under Serial No. 09 / 190,974, registered November 12, 1998, and entitled "Disabling and Enabling of Subvocabularies in Speech Recognition Systems". Preferably, the vocabulary can be defined phonetically, i.e., the vocabulary is represented by a phonetic alphabet, the training data being used for the purpose of learning the phones.

Correspondingly, the portion of the audio signal "transmit, obtained from the audio buffer 14, is divided into time segments or frames and processed by the feature extraction subsystem 28. The feature vectors relating to a frame are then provided with of a quantitative value by the pattern recognition subsystem 30, using the statistical models to generate probability values for observing the words in the target vocabulary When the probability value of a word exceeds one specified threshold value, this indicates that the word is part of the target vocabulary, the frames associated with the target word are labeled and are provided to the lexical word audio editing system by the pattern recognition subsystem 30. Thus the system Lexical word audio editing is informed of the start and end positions of the target word, in the audio signal. As will be explained, the pattern recognition subsystem may send a validation / invalidation signal to the audio editing system of lexical terms, in order to influence the editing process. It should be noted that the automatic edition of the lexical term by the audio editing processor 24 can be performed in many different ways. In an illustrative embodiment, as will be explained, a predefined transformation algorithm is used to convert unacceptable words to a similar but less offensive tone phrase. This approach uses a vocal synthesis of parametric or adaptive nature. In another illustrative embodiment, as will be explained, the audio element associated with the unacceptable lexical term is muted for the duration of the one or more words in the audio buffer. According to another illustrative embodiment, as will be explained, the unacceptable lexical term is replaced or obscured by a beep, with a sound or other appropriate sound effect, for the duration of the word (s) in the audio buffer. .

Referring now to FIG. 3, there is shown a block diagram of an audio editing system of lexical terms according to a first illustrative embodiment of the present invention. In particular, FIG. 3 discloses an implementation of the audio editing processor 24, wherein the offending language, having been detected by the word retoucher 22, is processed and is preferably used to construct one or more appropriate euphemistic substitution words. The substitution word (s) are then inserted into the original audio signal, instead of the portion of the signal containing the unacceptable target lexical term, to form an edited voice signal to be transmitted. Advantageously, as will be explained, using the technique of parametric or adaptive speech synthesis and vocal character transformation, the original words are preferably decomposed into a parametric representation of voice zone excitation and zone filter. voice. Then, by making appropriate adjustments in the parametric representation, a substitute word or phrase is formed and will replace the target word or phrase. It should be noted that adaptive speech synthesis techniques are known in the art. For example, adaptive speech synthesis techniques are described in N. Iwahashi et al., Speech Spectrum Transformation by Speaker Interpolation, IEEE International Conference on Acoustics, Speech and Signal Processing, pages I-461, I-464, 1994; H. Valbret et al., "Voice Transformation Using PSOLA Technique," IEEE International Conference on ACOUSt1CS, Speech and Signal Processing, pages I-445-I-448, 1992; and I1-Hyun Nam, "Voice Personality Transformation," Ph.D. Thesis, Rensselaer Polytechnic Institute, 1991.

As illustrated in FIG. 3, the input speech signal 32 is supplied to the audio editing processor 24 from the audio buffer 14 (FIG 1). The speech signal is then analyzed at the spectral analysis block 34. The spectral analysis provides the voice zone parameters (voice zone filter) 36 and the excitation parameters 38. The spectral analysis techniques to generate the parameters of Voice zone filter and excitation are well known in the art, for example as described in the article by N. Iwahashi et al., already cited above.

The voice zone parameters 36 and the excitation parameters 38 are then supplied to a voice zone modification block 40 and an excitation modification block 42, respectively. The parameter modification blocks 40 and 42 modify the parameters associated with the portion of the input signal that contains the target word or sentences (ie the unacceptable lexical term), according to a rules database. A purpose of the parameter change is to modify one or more spectral components of the target word to form a substitution word, which is then inserted into the speech in place of the target word. It should be noted that the excitation parameters of the part of the speech do not have to be modified in each case to form a word of substitution. For example, in cases where the main difference between the target word and the substitution word resides in a vowel change, the excitation parameters can remain unchanged.

The modification rule database 44 contains a set of rules previously stored, governing the modification of the target words by substitution words. The database 44 may be automatically derived from a large database containing learning data 46, including examples of target words and their corresponding (final) substitution words. A training machine algorithm is used to form the change rule database 44 from the training data 46. It should be noted that an algorithm for a training machine is an algorithm that is capable of creating generalized mapping or matching between examples, such as neural networks. An algorithm, which generalizes when there are several examples, is described in the II-Hyun Nam thesis cited above.

The output 50 of the word retouch processor 22 is supplied to the edit rule database block 44. It should be noted that the output of the word retouching processor is preferably a tagged frame containing the word or the sentence identified by the pattern identifier 30 (FIG.2), as part of the target vocabulary. The modification database then provides the appropriate modification rules to the voice zone modification block 40 and, if necessary, to the excitation modification block 42, which in turn modifies the appropriate spectral parameters to form the substitution word or phrase. The substitute word or phrase is then inserted into the appropriate part of the speech in place of the target word or phrase. The modified speech is then fed to a speech synthesis block 52 which, as is well known, generates a synthesized speech signal. The output of the synthesizer 52 consists of the edited voice signal. Advantageously, the voice signal now contains substitution words (for example a lexical term that is not unacceptable), instead of the target words (for example an unacceptable lexical term).

It should be noted that the functional blocks that are in FIG. 3 can operate in streaming mode or on an entire say buffer, depending on the need for real-time operations. It should also be noted that the modification rule system described above may, for example, be implemented using neural networks which are discussed in the aforementioned IIHyun Nam thesis.

Referring now to FIG. 4, there is shown a sequence flow diagram of a method for dynamically recognizing and editing lexical terms according to an illustrative embodiment of the present invention. In particular, the illustrative method comprises parametric speech synthesis, as explained above, as part of the audio editing operation. In step 60, a signal to be transmitted is received and its corresponding audio signal and video signal (or parts thereof, depending on the size of the memory) are buffered. Then, in step 62, the audio signal is subdivided into frames. Characteristic vectors are then extracted from the frames and decoded, for example as described above. If one or more target words are detected in a frame, the frame is then tagged (step 64).

In step 66, the input audio or speech signal is also spectrally analyzed, for example as described above, to decompose the signal into parametric representations. As mentioned above, such parametric representations may include voice zone parameters and excitation parameters. The spectral parameters of the detected target words found in the tagged frames are then modified according to pre-established modification rules, to form substitution words to replace the target words (step 68). The modified utterance is then passed through a vocal synthesis, in step 70, the lexical term of substitution resulting from the synthesis replacing the original target lexical mound. The edited audio signal (synthesized speech signal) is then combined with the original video signal for transmission in step 72.

Referring now to FIGS. 5A and 5B, respective embodiments of the audio editing processor 24 are shown. In FIG. 5A, the audio editing operation produces the muffling of the output speech signal while the word (s) whose word processing processor (22) has determined that it is part of the target vocabulary). In this embodiment, rather than the pattern recognition subsystem 30 providing the tagged frames to audio editing processor 24, a enable / disable signal is provided to processor 24 by subsystem 30. audio editing in this case is an audio output buffer 80. The enable / disable signal is generated by the pattern recognition subsystem 30 and serves to disable the audio output buffer 80 for the duration of the word or words detected. Correspondingly, the portion of the audio signal containing the unacceptable material is muted for transmission.

FIG.

5B is another illustrative embodiment of the audio editing processor 24. This is similar to the embodiment of FIG.

5A, wherein the pattern recognition subsystem 30 generates a enable / disable signal to turn on and off the output of the output audio buffer 80. However, the embodiment of FIG.

5B also includes a sound generator 82 which is validated by the subsystem 30 when the audio buffer 80 is disabled and vice versa. In this case, rather than mute the output when a target word or words are detected, a sound or other appropriate sound effect, generated by the tone generator 82, is inserted into the audio signal, during which time ) the detected word (s). <U> Text </ U> <U> FIGURE 1 </ U> 12 SIGNAL TO TRANSMIT 14 AUDIO BUFFER 16 VIDEO BUFFER 18 LEARNING DATA MEMORY 20 MEMORY OF STATISTICAL WORDS MODELS 24 AUDIO EDITING PROCESSOR 22 WORD RETOUCHING PROCESSOR 26 EDITED SIGNAL <U> FIGURE 2 </ U> SPEECH SIGNAL: VOICE SIGNAL 22 WORD RETRIEVAL PROCESSOR 28 CHARACTERISTIC EXTRACTION 30 IDENTIFICATION OF PATTERN 18 MEMORY <B> OF </ B> DATA OF LEARNING 20 MEMORY OF STATISTICAL WORDS TEMPLATES LABELED FRAMES: LABELED FRAMES <U> FIGURE 3 </ U> 32 VOCAL INPUT SIGNAL 34 SPEECH SPECTRAL ANALYSIS 36 VOCAL AREA PARAMETERS 38 EXCITATION PARAMETERS 40 VOCAL ZONE MODIFICATION 42 CHANGES TO EXCITATION 52 VOICE SYNTHESIS 26 OUTPUT VOCAL SIGNAL 50 IDENTIFIER OUTPUT (TARGET) 44 DATABASE MODIFICATION RULES 46 LEARNING DATA 48 LEARNING ALGORITHM TARGET MACHINE 1: FINAL 1 TARGET N: FINAL N <U > FIGURE 4 </ U> 60 TAM AUDIO PON AND RECEIVED VIDEO SIGNALS AS PART OF THE SIGNAL TO TRANSMIT 62 AUDIO INPUT SIGNAL FRAMING SUBDIVISION, EXTRACTION OF FEATURE VECTORS AND DECODING EACH FRAME 64 IF ONE OR MORE TARGET WORDS ARE DETECTTED IN THE FRAME, THEN TAG THE FRAME 66 SPECTRALLY ANALYZE THE ENTRY VOCAL SIGNAL CORRESPONDING TO THE LABELED FRAMES 68 MODIFY THE SPECTRAL PARAMETERS OF THE TARGET WORDS, IN ACCORDANCE WITH THE PRE-ESTABLISHED CHANGE RULES, IN ORDER TO FORM SUBSTITUTE WORDS, TO REPLACE THE TARGET WORDS 70 SYNTHESIZE A VOICE SIGNAL BY SUBSTITUTION WORDS, REPLACE TO PRODUCE AN AUDIO EDITE SIGNAL 72 COMBINE THE AUDIO SIGNED EDIT WITH THE ORIGINAL VIDEO SIGNAL FOR TRANSMISSION <U> FIGURE 5A </ U> INPUT SPEECH SIGNAL: VOICE SIGNAL OUTPUT SPEECH SIGNAL: SIGNAL ENABLE / DISABLE OUTPUT VOCAL SIGNAL: VALIDATION / INVALIDATION OF SIGNAL <U> FIGURE 5B </ U> INPUT SPEECH SIGNAL: VOCAL INPUT SIGNAL OUT PUT SPEECH SIGNAL: VOICE SIGNAL OUTPUT ENABLE / DISABLE SIGNAL: VALIDATION / INVALIDATION OF SIGNAL 82 SOUND GENERATOR

Claims (22)

A method of dynamically processing a speech signal, the method comprising the steps of automatically detecting whether a portion of the speech signal comprises a lexical term associated with one or more vocabularies of lexical terms; and automatically editing the portion of the speech signal when the lexical term is detected, such that the detected lexical term is not audibly perceptible. 2. The method according to claim 1, wherein the automatic editing step comprises automatically substituting the portion of the speech signal containing the lexical term with a modifiable modified voice signal portion. The method of claim 1, wherein the automatic editing step comprises automatically muting the portion of the speech signal containing the lexical term. 4. The method of claim 1, wherein the automatic editing step comprises automatically replacing the portion of the speech signal containing the lexical term with an acoustic signal. A method of dynamically processing a speech signal, the method comprising the steps of automatically detecting whether a portion of the speech signal comprises a lexical term associated with one or more vocabularies of lexical terms; and automatically substituting the portion of the speech signal, when the lexical term is detected, with a modified voice signal portion having at least one voice parameter that is a modification of the at least one speech parameter in the portion of the speech signal containing the term lexical. The method of claim 1 or 5, wherein the step of automatically detecting includes decoding the speech signal using a voice recognition system, according to one or more vocabularies of lexical terms. The method of claim 6, wherein the voice recognition system is a word retouching system. The method of claim 5, wherein the automatic substitution step comprises analyzing the portion of the speech signal containing the lexical term to produce the at least one voice parameter; and adjusting the at least one voice parameter to generate the modified voice signal portion for substitution. The method of claim 1 or 5, wherein the voice signal is a part of a transmission signal. The method of claim 9, wherein the transmission signal is a part of a live transmission. 11. The method of claim 1 or 5, wherein one or more vocabularies of lexical terms contain, respectively, one or more unacceptable words. The method of claim 11, wherein the one or more vocabularies of lexical terms are formed according to official or governmental standards. Apparatus for dynamically processing a speech signal, comprising at least one processor operable to automatically detect that a portion of the speech signal comprises a lexical term associated with one or more vocabularies of lexical terms, and to automatically edit the portion voice signal when the lexical term is detected, such that the detected lexical term is not audibly perceptible. The apparatus of claim 13, wherein the automatic detection operation comprises decoding the speech signal using a voice recognition system, according to one or more vocabularies of lexical terms. . The apparatus of claim 14, wherein the voice recognition system is a word retouching system. The apparatus of claim 13, wherein the automatic editing operation comprises automatically substituting the portion of the speech signal containing the lexical term with a modified voice signal portion at the parameter level. The apparatus of claim 13, wherein the automatic editing operation comprises automatically muting the portion of the speech signal containing the lexical term. The apparatus of claim 13, wherein the automatic editing operation comprises automatically replacing the portion of the speech signal containing the lexical term with an acoustic signal. The apparatus of claim 13, wherein the voice signal is part of a transmission signal. The apparatus of claim 19 wherein the transmission signal is part of a live transmission. The apparatus of claim 13, wherein one or more vocabularies of lexical terms contain, respectively, one or more unacceptable words. 22. The apparatus of claim 21, wherein one or more vocabularies of lexical terms are formed according to official or governmental standards.