CN1924996B - System and method for selecting audio content by using speech recognition - Google Patents

Abstract

The present invention is a system and method for selecting sound content by using voice recognition, which is used to obtain a sound sentence in a sound content played in sequence, and then process it in a processing system. The system includes: a playback module, which is used to play the sound content; a receiving module, which is used to instantly receive a voice input sentence issued by a user; a buffer module, which is used to temporarily store the sound content in a specified interval played by the playback module and the voice input sentence issued by the user; a recognition module, which is used to capture the sound content and the voice input sentence in the specified interval in the buffer module to perform voice recognition; and a conversion module, which is used to convert a text sentence corresponding to the sound sentence according to the sound sentence that best matches the voice input sentence compared by the recognition module, so as to transmit it to the processing system for processing.

Description

System and method for selecting audio content by using speech recognition

technical field

The present invention relates to a system and method for selecting audio content, in particular, the present invention relates to a system and method for selecting a specific audio segment from audio content by using speech recognition technology for further subsequent processing. the

Background technique

The current form of information expression is mainly based on the content of the written text, and there are often some important or key text sentences in the written text content, for these key phrases (key phrases) can be obtained through the system Actively add marks for easy identification by users, such as highlighting, underlining, quotation marks, different colors or font changes, or are actively moved by users with various selection tools such as keyboard, mouse or stylus The selected key words can be used for advanced search or keyword index. For example, the system can add hyperlinks to key sentences in the content of the web pages on Internet sites, and after clicking, they can be linked to other web pages. After selecting a key sentence with the mouse, it can be forwarded to various search engines on the Internet to find relevant corresponding articles. the

Because most of the current information content is mainly presented in text, there are still a minority of information content that is purely expressed in sound. However, as various mobile devices become more and more popular, due to the limitation of screen size, some information is It is more convenient to change "watch" to "listen". Coupled with the gradual popularization of technologies such as Bluetooth headsets and unlimited Internet access, more and more forms of information expression use "listen" sound content, and for these sound content How to select the key phrase (key phrase) has become a problem that needs to be solved. the

In addition, because the written content of "seeing" is used to express its information in a parallel manner (parallel), while the audio content of "listening" is expressed in a sequential manner (sequential), it is obvious that It is impossible to use the existing selection tools like the written text content, such as hyperlinks or key words selected by the mouse, etc. to select the sound content, so how to effectively interact with the sound content has become an increasing demand for users .

To sum up, since the technology for selecting key sentences of voice content still has deficiencies, the inventors have invented the present invention "system for selecting voice content by using speech recognition and its method". the

Contents of the invention

The main purpose of the present invention is to provide a system and method for selecting voice content by voice recognition, which can utilize the existing voice recognition method and use it properly to achieve effective interaction between the voice content and the user. the

Another object of the present invention is to provide a system and method for selecting sound content by using speech recognition. Speech recognition is performed on the audio content played in the specified interval, and specific audio sentences in this segment of audio content are selected for subsequent processing. the

Yet another object of the present invention is to provide a system for selecting audio content, for obtaining an audio sentence from an audio content played sequentially, and then processing it in a processing system, which includes: a playback module for playing The sound content; a receiving module, used to immediately receive a voice input sentence issued by a user; a buffer module, used to temporarily store the sound content and the sound content in a specified interval played by the playing module. The voice input sentence issued by the user, and the specified interval is the sound content played by the playback module within a last specified time when the receiving module receives the voice input sentence; a recognition module, It is used to extract the voice content in the specified section in the buffer module and the voice input sentence for voice recognition, and then compare and identify the voice content in the specified section that best matches the voice uttered by the user The sound sentence of the input sentence; and a conversion module, connected with the recognition module, in order to convert the corresponding sound sentence according to the sound sentence that is most consistent with the voice input sentence compared by the recognition module A literal statement of , and then provided to the processing system for processing. the

According to the above idea, the system also includes a source database, and the source database can contain multiple text contents, so the conversion module can also be connected with the source database and the playback module to retrieve the source database A text content is converted into the audio content and played through the playback module. the

According to the above idea, the source database may also include a plurality of text contents and voice information, so the playback module captures the voice data of the source database to play the audio content. the

According to the above conception, the last specified time is 20 seconds. the

According to the above idea, the processing system is a voice dialogue system, an index classification system, a control system or an advanced search system, if the processing system is the advanced search system, it can be retrieved through a retrieval module Output relevant text or voice information corresponding to the text sentence for the user to use. the

Another object of this case is to provide a system for selecting audio content, which is used to obtain an audio sentence in an audio content played sequentially, wherein the audio content has a plurality of audio tags for marking the audio content in the audio content. A plurality of key terms, which include: a playback module, used to play the sound content with the sound mark; a receiving module, used to receive a voice input sentence issued by a user in real time; a recognition module group, performing voice recognition on a plurality of key terms of the voice content and the voice input sentence, and then comparing and identifying the voice sentence that best matches the voice input sentence issued by the user among the key terms; a buffer module set, used to temporarily store the sound content in a specified interval played by the playback module and the voice input sentence issued by the user, wherein the recognition module retrieves the buffer module Identify the voice content in the specified interval and the voice input sentence issued by the user; and a conversion module for matching the voice input according to the recognition module The voice sentence of the sentence is converted into a text sentence corresponding to the voice sentence. the

According to the above idea, the recognition module uses a direct sound wave comparison to find out the closest possible sound waveform of the two sides directly compared or. the

According to the above idea, the identification module is selected from a hidden Markov model (Hidden Markov Model, HMM), a neural network (Neural Networks), a dynamic time calibration (Dynamic Time Warping, DTW) or a Voice template comparison (Template Matching) for voice recognition. the

According to the above idea, the sound mark is to express the key term with different speeds, different tones or different volumes, or the sound mark is a method mark for adding prompt sounds before and after the key term. the

According to the above idea, the text sentence converted by the conversion module is further provided to a processing system for subsequent processing. the

Another object of this case is to provide a method for selecting sound content, for obtaining a sound sentence in a sound content played sequentially, and then performing a follow-up processing procedure, which includes the following steps: (a) playing the sound content ; (b) receiving a voice input sentence issued by a user; (c) performing speech recognition on the voice input sentence and the sound content played in a specified interval; and (d) from the specified interval The voice content is compared with the voice content that most matches the voice input sentence uttered by the user, and then the subsequent processing procedure is performed. the

According to the above idea, the sound content also has a plurality of sound marks, which are used to mark out a plurality of key terms in the sound content, so

According to the above idea, the step (c) further includes performing voice recognition on the voice input sentence and the key words with one of the plurality of key words in the voice content in the specified interval. the

According to the above idea, the step (d) further includes comparing the plurality of key words to find the voice sentence that best matches the voice input sentence uttered by the user. the

According to the above idea, the step (c) is through a comparison of the sound waveforms of the two sides which is the closest possible direct comparison or through a method selected from a hidden Markov model, a neural network method, and a dynamic time calibration The voice recognition is carried out by means of a voice template comparison mode or a voice template comparison mode. the

According to the above idea, the step (d) further includes a step (d1) converting the audio content into a text sentence. the

According to the above idea, the subsequent processing step is an advanced search step, a keyword indexing step, a voice dialogue system or a control program. the

The effect and purpose of this case can be explained through the following embodiments, and it has a deeper understanding. the

Description of drawings

FIG. 1(A) is a schematic configuration diagram of a system for selecting audio content by using speech recognition according to the first preferred embodiment of the present invention. the

FIG. 1(B) is a schematic configuration diagram of a system for selecting audio content by using speech recognition according to a second preferred embodiment of the present invention. the

FIG. 2 is a schematic flowchart of a method for selecting audio content by using speech recognition according to a preferred embodiment of the present invention. the

Detailed ways

For the following descriptions of the present invention, those skilled in the art should understand that the following descriptions are only for illustration purposes and not intended to limit the present invention. the

The following is a description of the system and method for selecting sound content by using speech recognition in the preferred embodiment of this case, but the actual structure and the method adopted do not necessarily completely conform to the described structure and method, and those of ordinary skill in the art will be able to Various changes and modifications may be made without departing from the true spirit and scope of the invention. the

Please refer to FIG. 1 (A) and (B), which are schematic diagrams of a system structure of a system and method for selecting audio content by using speech recognition disclosed in the present invention, respectively. The selection system 10 of the present invention comprises a playing module 11, a receiving module 12, a buffering module 13, an identification module 14, a conversion module 15 and a source database 16, and it is obtained from the playing module 11 Select a voice sentence from the played voice content, and then provide it to a processing system 17 for subsequent processing. the

Wherein, the playing module 11 is used to play the sound content so that a user can hear the sound content according to time sequence, and the receiving module 12 is used to receive a voice from the user in real time Input sentence, in addition, this buffering module 13 is to temporarily store this sound content in a specified interval played by this playing module 11 and this voice input that this user that receives by this receiving module 12 sends out sentence, therefore, the identification module 14 extracts the voice content in the designated interval in the buffer module 13 and the speech input sentence for speech recognition, and then compares and recognizes the best voice content in the designated interval. The voice sentence that matches the voice input sentence issued by the user, so the conversion module 15 is used to convert the voice sentence that best matches the voice input sentence that the recognition module 14 compares. A text sentence corresponding to the sound sentence, and the source database 16 provides the source of the sound content played by the playing module 11 . the

In addition, according to the different types of information stored in the source database 16 , the structure of the selection system 10 is also slightly different. the

Then, referring to Fig. 1 (A), it is the selection system 10 of the first embodiment of the present case, wherein the source database 16 contains a plurality of text contents, so the conversion module 15 can also be connected with the source database 16 and the player The modules 11 are connected to each other, and the conversion module 15 can extract a text content among a plurality of text contents in the source database 16 and convert it into the audio content and play it out through the playback module 11. At the same time, through The conversion module 15 can also store the audio content to be played in the buffer module 13 . the

In addition, if the source database 16 includes multiple text content and voice information, in this case, please refer to FIG. 1(B), the source database 16 does not need to be connected with the conversion module 15, but directly The sound content can be played by the playback module 11 from the voice data in the source database 16 , and the source database 16 can also store the sound content to be played in the buffer module 13 . the

And because the user hears the sound content in chronological order, the voice input sentence sent by the user usually belongs to the sound content just heard, so the present invention sets the designated interval as when the receiving module 12 When receiving the voice input sentence, the playing module 11 plays the sound content within a last designated time, and temporarily stores the sound content of the designated interval in the buffer module 13, wherein the last designated time can be set Set to 20 seconds or any other arbitrary time. In addition, when the receiving module 12 receives the voice input sentence issued by the user, the voice input sentence will also be stored in the buffer module 13, so the recognition module 14 only needs to retrieve the buffer module 13 The stored voice content is compared with the voice input sentence by using voice recognition technology to select the voice sentence that best matches the voice input sentence issued by the user in the voice content of the specified interval. Through the conversion module 15, the voice sentence selected by comparison is converted into a text sentence, and then provided to the processing system 17 for processing. the

Wherein the processing system 17 can be a voice dialogue system, an index classification system, a control system or an advanced search system, etc., and can carry out different follow-up processing procedures according to different needs, for example: the voice dialogue system can be based on The meaning of the text sentence is used to carry out a voice dialogue, the index classification system can perform keyword indexing program on the sound content, the control system can control other programs by understanding the meaning of the text sentence, or the processing The advanced search system can pass the text sentence through a retrieval module (not shown in the figure) to retrieve relevant text or voice information corresponding to the text sentence for the user to use. the

And because the processing system 17 performs different follow-up processing procedures in response to different needs, for example: if the processing system 17 is the index classification system, then only the selection system 10 is required to provide the audio content to perform index classification, and if If the processing system 17 is the voice dialogue system, the control system or the advanced search system, the selection system 10 may be required to provide the text statement for further judgment and analysis by the processing system 17 . Therefore, the selection system 10 can transmit the voice sentence or the text sentence to the processing system 17 for subsequent processing according to the different types of the processing system 17, and in its actual information flow transmission process, if the selection If the system 10 wants to send the voice sentence to the processing system 17, the recognition module 14 can send the voice sentence to the processing system 17; otherwise, if the selection system 10 wants to send the text sentence to the processing system 17 In the process, the converted text statement can be sent to the processing system 17 through the conversion module 15 . the

Furthermore, the identification module 14 performs speech recognition through a direct acoustic wave comparison method or an acoustic model comparison method, wherein the direct acoustic wave comparison method is to directly compare the sound waveforms of both parties, and compare Match the closest possibility, and the acoustic model comparison method is through a hidden Markov model (Hidden Markov Model, HMM), a neural network (Neural Networks), a dynamic time calibration (Dynamic Time Warping, DTW) Or a variety of acoustic models such as template matching (Template Matching) for speech recognition. the

Please refer to FIG. 2 again, which is a flow chart of the implementation method of the system and method for selecting audio content by using speech recognition according to the present invention. In the method of the present invention, the system first plays a section of audio content 21, then receives the voice input sentence 22 sent by the user, and performs voice recognition 23 on the voice input sentence and the audio content in a specified interval of the playback audio content. , and compare and select the voice content 24 that best matches the voice input sentence issued by the user from the voice content in the specified interval, and then perform a follow-up processing procedure 25, wherein the follow-up processing procedure can be a Advanced search steps, a keyword indexing step, a voice dialogue system or a control program, and as described above, when the subsequent processing procedure needs to be carried out using text information, the method of the present invention can also use the voice The content is converted into a text statement for the subsequent processing program to process. the

In addition, in order to make speech recognition more efficient, the present invention can also actively mark the voice content, so that the voice content has multiple voice marks to mark multiple key terms in the voice content, so that The user knows that this is a key term when listening to it, and the sound mark is a method mark for expressing the key term with different speeds, different tones or different volumes or adding prompting sounds before and after the key term. the

Wherein the sound mark can be stored in the source database 16 shown in Figure 1 (A) and (B), no matter the storage of the source database 16 is pure text content or has text content and voice information at the same time, as long as through The simple setting of the system (for example: in the voice information, the voice key sentence with a specific sound mark can be directly stored, and in the text content, the sound form to be marked can be directly marked on the specific text segment in the text content , so that the specific sound mark can be played during text-to-speech later), the sound content with the sound mark can be played. the

Therefore, the voice recognition method can only perform voice recognition on the voice content with the voice mark in the specified interval, so not only the recognition time is effectively saved, but also the recognition rate is relatively improved. However, if it is simply discussed in terms of technology, the selection system of the present invention can also directly compare all the voice content with its voice input sentence without specifying the interval of the voice content, or compare all the voice content with the The key words of the sound mark are compared with the voice input sentence. the

Therefore, according to the sound content selection technology provided by the present invention, the appropriate sound sentence is selected in real time, which provides a convenient interaction mechanism to allow the user to interact effectively with the (sequential) sound content presented in a sequential manner, greatly improving the In the past, users could only listen to the audio content in a passive position to extract information, and the improved audio content in the past cannot have as many tools as the written text content presented in parallel to help people and content of interaction. the

Therefore, in practical applications, the present invention can be applied to various interactive devices (such as mobile devices, bluetooth devices, or Internet access devices) that convey information through sound content. As long as the sound content selection mechanism provided by the present invention is used, the It allows the user to easily select the desired voice sentence from the voice content, and then provide it as a follow-up related processing or service item, and the user does not need special training or memorization of special operation instructions. the

To sum up, this case can indeed provide a system and method for selecting voice content by using voice recognition. And with the appropriate information access technology and special voice marking mode, the voice input sentence issued by the user and the sound content played can be recognized by voice, and then the specific voice sentence in this piece of voice content can be selected, and then carried out For subsequent various processing procedures, this technology does not need to add many complicated software and hardware devices, and the implementation cost is extremely low. Therefore, the technology of the sound content selection system and the method for selecting sound content of the present invention is relatively simple but can provide extremely high convenience, and the user does not need special training or learning and can be applied to various fields of expressing information with sound, and It can effectively promote the progress of the industry. The technology of the invention is simple, it can be used in a wide range of fields, and has real industrial value. Therefore, an application for an invention patent is filed according to law. the

The above description utilizes the preferred embodiments to illustrate the present invention in detail, rather than limit the scope of the present invention, so those of ordinary skill in the art should be able to understand that it is appropriate to make some minor changes and adjustments without losing the gist of the present invention Therefore, it should be regarded as a further implementation of the present invention. the

The claimed scope of the present invention should be determined by what is stated in the claims.

Claims (8)

1. system that chooses sound-content in order to obtaining a sound statement in a sound-content of playing in regular turn, and then handles in a disposal system, comprises:

One plays module, in order to play described sound-content;

One receives module, a phonetic entry statement that is sent in order to instant reception one user;

One buffering module, the described phonetic entry statement that described sound-content between a designation area of being play in order to temporary described broadcast module and described user are sent, and be between this designation area when this reception module receives this phonetic entry statement, the sound-content of having listened that this broadcast module was play in a last fixed time;

One identification module, in order to the described sound-content of having listened in capturing described in the described buffering module between designation area and described phonetic entry statement and carry out speech recognition, and then comparison picks out the described sound statement in the described sound-content between described designation area, and wherein this sound statement meets the described phonetic entry statement that described user sends most; And

One conversion module is connected with described identification module, changes out a literal statement of corresponding described sound statement in order to the described sound statement of being compared out according to described identification module, and then offers this disposal system and handle.

2. the system as claimed in claim 1, it is characterized in that also comprising: one comes source database, this comes source database to have a plurality of word contents, wherein said conversion module also comes source database to be connected with described broadcast module with described, describedly comes a word content of source database and converts described sound-content to and see through described broadcast module and play in order to capture; And/or one come source database, and this comes source database to have a plurality of word contents and voice information, and wherein said broadcast module acquisition is described to come the speech data of source database to broadcast described sound-content.

3. the system as claimed in claim 1 is characterized in that the described last fixed time is 20 seconds; Described disposal system is an Advanced Search system, and wherein said disposal system sees through related text or the voice information of a retrieval module to retrieve corresponding described literal statement

Use for this user; And/or described disposal system for be selected from a speech dialogue system, an index classification system and a control system one of them.

4. system that chooses sound-content, in order to obtain a sound statement in a sound-content of playing in regular turn, wherein said sound-content also has a plurality of voice marks, and in order to mark a plurality of crucial term in the described sound-content, this system comprises:

One plays module, has the described sound-content of described voice mark in order to broadcast;

One receives module, a phonetic entry statement that is sent in order to instant reception one user;

One identification module, a plurality of crucial term and described phonetic entry statement to described sound-content carry out speech recognition, and then compare the described sound statement that picks out in described these crucial terms, wherein this sound statement meets the described phonetic entry statement that described user sends most;

One buffering module, the described phonetic entry statement that described sound-content between a designation area of being play in order to temporary described broadcast module and described user are sent, wherein said identification module capture the described phonetic entry statement that described sound-content interior between this designation area in the described buffering module and described user sent and carry out identification; And

One changes module, changes out a literal statement of corresponding described sound statement in order to the described sound statement of being compared out according to described identification module.

5. system as claimed in claim 4 is characterized in that described identification module sees through one and directly compares both sides' sound waveform mode to compare the most close sound waveform or to carry out speech recognition through being selected from a concealed markov model mode, a neural network mode, a dynamic time calibrating mode or a voice masterplate comparison mode; This voice mark of representing this key term is represented as different speeds, not same tone or different volume, or the front and back that this voice mark is represented as described crucial term add prompt tone, and described literal statement offers a disposal system to carry out subsequent treatment.

6. method of choosing sound-content in order to obtaining a sound statement in a sound-content of playing in regular turn, and then is carried out a subsequent processing steps, and this method comprises the following step:

(a) play described sound-content;

(b) receive the phonetic entry statement that a user is sent;

(c) with described phonetic entry statement and the described sound-content of being play between a designation area carry out speech recognition; And

(d) compare out the described sound-content that meets the described phonetic entry statement that described user sends most the described sound-content in between designation area, and then carry out described subsequent processing steps; Changing described sound-content is a literal statement; And/or described subsequent processing steps is an Advanced Search step, a key word index step, a speech dialogue system or an operating steps.

7. method as claimed in claim 6 is characterized in that described sound-content also has a plurality of voice marks, in order to mark a plurality of crucial term in the described sound-content; The described sound-content of described step (c) comprises described a plurality of crucial term; Described step (d) also comprises by comparing out described sound statement in described a plurality of crucial terms, and wherein this sound statement meets the described phonetic entry statement that described user sends most; And/or the described voice mark of representing described crucial term is represented as different speeds, not same tone or different volume, or the front and back that this voice mark is represented as described crucial term add prompt tone.

8. method as claimed in claim 6 is characterized in that described step (c) sees through one and directly compares both sides' sound waveform mode to compare the most close sound waveform or to carry out speech recognition through being selected from a concealed markov model mode, a neural network mode, a dynamic time calibrating mode or a voice masterplate comparison mode.

Images