CN112233656A - Artificial intelligent voice awakening method - Google Patents
Artificial intelligent voice awakening method Download PDFInfo
- Publication number
- CN112233656A CN112233656A CN202011075074.3A CN202011075074A CN112233656A CN 112233656 A CN112233656 A CN 112233656A CN 202011075074 A CN202011075074 A CN 202011075074A CN 112233656 A CN112233656 A CN 112233656A
- Authority
- CN
- China
- Prior art keywords
- awakening
- voice
- voice data
- model
- wake
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 23
- 238000001514 detection method Methods 0.000 claims abstract description 37
- 238000012549 training Methods 0.000 claims description 29
- 238000013473 artificial intelligence Methods 0.000 claims description 15
- 238000009499 grossing Methods 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 3
- 230000007547 defect Effects 0.000 abstract description 3
- 238000005516 engineering process Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/02—Feature extraction for speech recognition; Selection of recognition unit
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
- G10L15/063—Training
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/26—Speech to text systems
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
- G10L2015/088—Word spotting
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Computational Linguistics (AREA)
- Quality & Reliability (AREA)
- Signal Processing (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Telephone Function (AREA)
Abstract
The invention relates to voice awakening, in particular to an artificial intelligent voice awakening method, which comprises the steps of obtaining voice data, determining energy characteristics corresponding to the voice data by using a voice detection model, determining text data corresponding to the voice data according to the energy characteristics, judging whether the voice data contains awakening keywords or not, if the voice data contains the awakening keywords, utilizing an awakening judgment model to carry out awakening judgment, and if not, outputting a current situation maintaining instruction by the awakening judgment model; the technical scheme provided by the invention can effectively overcome the defects of low accuracy of voice awakening recognition and incapability of flexibly adjusting awakening words in the prior art.
Description
Technical Field
The invention relates to voice awakening, in particular to an artificial intelligence voice awakening method.
Background
In an intelligent home or a voice interactive system, a voice wake-up (wakeword) technology is widely applied. However, the accuracy and the computation load of voice wakeup recognition greatly reduce the actual application experience, and the requirements on the hardware of the device are improved, for example: if the false wake-up rate of voice wake-up is greater than a threshold in the application, such wake-up false trigger frequency may cause the user to feel the discomfort; on the other hand, if the computation amount of voice wakeup exceeds the computation capability of some low-end chips, the voice wakeup technology is restricted to be used by many products.
In the existing related technologies, a keyword recognition (keyword-spotting) technology is used for voice awakening, a smart small decoding network is built by designing a deep neural network model, and a voice awakening function is realized by matching with some keyword detection technologies.
However, the above voice wake-up technology based on the keyword recognition method has a large number of model parameters, and the design of the filling words needs to be changed for different wake-up words, and the corresponding decoding parameters and the keyword detection technology need to be adjusted, so that it is difficult to have a uniform algorithm to ensure that the wake-up effect of each wake-up word is at a stable level.
Disclosure of Invention
Technical problem to be solved
Aiming at the defects in the prior art, the invention provides an artificial intelligent voice awakening method which can effectively overcome the defects that the voice awakening recognition accuracy rate is low and the awakening words cannot be flexibly adjusted in the prior art.
(II) technical scheme
In order to achieve the purpose, the invention is realized by the following technical scheme:
an artificial intelligence voice awakening method comprises the following steps:
s1, acquiring voice data;
s2, determining energy characteristics corresponding to the voice data by using the voice detection model, and determining text data corresponding to the voice data according to the energy characteristics;
s3, judging whether the voice data contains the awakening keyword;
and S4, if the voice data contains the awakening keyword, utilizing the awakening judgment model to perform awakening judgment, otherwise, outputting a current status maintaining instruction by the awakening judgment model.
Preferably, the determining the energy feature corresponding to the voice data by using the voice detection model includes:
preprocessing voice data, and determining energy values of filter banks in the voice data;
and processing the energy values of each filter bank in the voice data by using the trained voice detection model.
Preferably, the processing the energy values of the filter banks in the speech data by using the trained speech detection model includes:
determining the energy value of each filter group in a certain frame of voice data and corresponding smoothing parameters;
acquiring the energy value of each smoothed filter bank in the previous frame of voice data;
and determining the corresponding smoothed filter bank energy value in certain frame of voice data according to the filter bank energy value, the smoothed filter bank energy value and the smoothing parameter.
Preferably, the training step of the speech detection model comprises:
acquiring training voice data and initializing a voice detection model;
and training the voice detection model by using the training voice data, and determining the trained voice detection model and the corresponding smoothing parameters.
Preferably, the determining whether the voice data includes a wake-up keyword includes:
carrying out feature analysis on voice data of continuous multiple frames, and caching audio features;
comparing each frame of cached audio features with the awakening keywords by using a keyword comparison model, and determining the association degree between the frame of audio features and the awakening keywords;
and determining the confidence level of the voice data containing the awakening key words according to the association degree between each audio feature and the awakening key words.
Preferably, the keyword comparison model is trained first, and then the wake-up decision model is trained.
Preferably, the keyword comparison model is trained using a first speech data training sample set, where the first speech data training sample set includes a wake-up keyword.
Preferably, the awakening decision model is trained by using a second speech data training sample set and awakening keyword text data, and background noise in the second speech data training sample set is greater than background noise in the first speech data training sample set.
Preferably, the making of the wake decision by using the wake decision model includes:
and inputting the awakening keywords detected by the keyword comparison model and the text data in the S2 into the trained awakening judgment model, and comprehensively carrying out awakening judgment by the awakening judgment model according to the detection result of the keyword comparison model and the text data.
Preferably, if the wake-up decision model determines that both the detection result of the keyword comparison model and the text data contain the wake-up keyword, the wake-up decision model outputs a wake-up instruction; otherwise, the awakening judgment model outputs a current status maintaining instruction.
(III) advantageous effects
Compared with the prior art, the artificial intelligence voice awakening method provided by the invention has the advantages that the voice awakening identification accuracy can be effectively improved through the comparison detection among the voice data, the text data converted from the voice data and the awakening keywords, the voice detection model is constructed to convert the voice data into the text data, the keyword comparison model is constructed to judge whether the voice data contains the awakening keywords, the awakening judgment model is constructed to carry out awakening judgment, and the awakening keywords can be flexibly adjusted conveniently through setting the artificial intelligence model.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below. It is obvious that the drawings in the following description are only some embodiments of the invention, and that for a person skilled in the art, other drawings can be derived from them without inventive effort.
FIG. 1 is a schematic flow chart of voice wake-up according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. It is to be understood that the embodiments described are only a few embodiments of the present invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
An artificial intelligence voice awakening method, as shown in fig. 1, acquires voice data; and determining energy characteristics corresponding to the voice data by using the voice detection model, and determining text data corresponding to the voice data according to the energy characteristics.
In the technical scheme of the application, the training step of the voice detection model comprises the following steps:
acquiring training voice data and initializing a voice detection model;
and training the voice detection model by using the training voice data, and determining the trained voice detection model and the corresponding smoothing parameters.
The method for determining the energy characteristics corresponding to the voice data by using the voice detection model comprises the following steps:
preprocessing voice data, and determining energy values of filter banks in the voice data;
and processing the energy values of each filter bank in the voice data by using the trained voice detection model.
The method for processing the energy values of the filter banks in the voice data by using the trained voice detection model comprises the following steps:
determining the energy value of each filter group in a certain frame of voice data and corresponding smoothing parameters;
acquiring the energy value of each smoothed filter bank in the previous frame of voice data;
and determining the corresponding smoothed filter bank energy value in certain frame of voice data according to the filter bank energy value, the smoothed filter bank energy value and the smoothing parameter.
Judging whether the voice data contains a wake-up keyword or not; if the voice data contains the awakening keyword, the awakening judgment model is used for awakening judgment, and otherwise, the awakening judgment model outputs a current status maintaining instruction.
Wherein, judge whether contain the key word of awakening up in the voice data, include:
carrying out feature analysis on voice data of continuous multiple frames, and caching audio features;
comparing each frame of cached audio features with the awakening keywords by using a keyword comparison model, and determining the association degree between the frame of audio features and the awakening keywords;
and determining the confidence level of the voice data containing the awakening key words according to the association degree between each audio feature and the awakening key words.
According to the technical scheme, the keyword comparison model is trained, and then the awakening judgment model is trained.
The keyword comparison model is trained by utilizing a first voice data training sample set, and the first voice data training sample set contains awakening keywords; and the awakening judgment model is trained by utilizing a second voice data training sample set and awakening keyword text data, wherein the background noise in the second voice data training sample set is greater than that in the first voice data training sample set.
Wherein, the awakening judgment is carried out by utilizing the awakening judgment model, and the awakening judgment comprises the following steps:
and inputting the awakening keywords detected by the keyword comparison model and the text data in the S2 into the trained awakening judgment model, and comprehensively carrying out awakening judgment by the awakening judgment model according to the detection result of the keyword comparison model and the text data.
If the awakening judgment model judges that the detection result of the keyword comparison model and the text data both contain the awakening keyword, the awakening judgment model outputs an awakening instruction; otherwise, the awakening judgment model outputs a current status maintaining instruction.
According to the technical scheme, the accuracy of voice awakening recognition can be effectively improved through comparison detection between the text data converted from the voice data and the awakening keywords, and the voice awakening recognition device is provided with the voice detection model, the keyword comparison model and the awakening judgment model which can be independently trained, so that the awakening keywords can be flexibly adjusted.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; such modifications and substitutions do not depart from the spirit and scope of the corresponding technical solutions.
Claims (10)
1. An artificial intelligence voice awakening method is characterized in that: the method comprises the following steps:
s1, acquiring voice data;
s2, determining energy characteristics corresponding to the voice data by using the voice detection model, and determining text data corresponding to the voice data according to the energy characteristics;
s3, judging whether the voice data contains the awakening keyword;
and S4, if the voice data contains the awakening keyword, utilizing the awakening judgment model to perform awakening judgment, otherwise, outputting a current status maintaining instruction by the awakening judgment model.
2. The artificial intelligence voice wake-up method according to claim 1, characterized in that: the determining the energy characteristics corresponding to the voice data by using the voice detection model comprises the following steps:
preprocessing voice data, and determining energy values of filter banks in the voice data;
and processing the energy values of each filter bank in the voice data by using the trained voice detection model.
3. The artificial intelligence voice wake-up method according to claim 2, characterized in that: the processing of the energy values of the filter banks in the speech data by using the trained speech detection model comprises:
determining the energy value of each filter group in a certain frame of voice data and corresponding smoothing parameters;
acquiring the energy value of each smoothed filter bank in the previous frame of voice data;
and determining the corresponding smoothed filter bank energy value in certain frame of voice data according to the filter bank energy value, the smoothed filter bank energy value and the smoothing parameter.
4. An artificial intelligence voice wake-up method according to any one of claims 1-3, characterised in that: the training step of the voice detection model comprises the following steps:
acquiring training voice data and initializing a voice detection model;
and training the voice detection model by using the training voice data, and determining the trained voice detection model and the corresponding smoothing parameters.
5. The artificial intelligence voice wake-up method according to claim 1, characterized in that: the judging whether the voice data contains the awakening keyword includes:
carrying out feature analysis on voice data of continuous multiple frames, and caching audio features;
comparing each frame of cached audio features with the awakening keywords by using a keyword comparison model, and determining the association degree between the frame of audio features and the awakening keywords;
and determining the confidence level of the voice data containing the awakening key words according to the association degree between each audio feature and the awakening key words.
6. The artificial intelligence voice wake-up method of claim 5, wherein: and training the keyword comparison model, and then training the awakening judgment model.
7. The artificial intelligence voice wake-up method of claim 6, wherein: the keyword comparison model is trained by using a first voice data training sample set, and the first voice data training sample set contains awakening keywords.
8. The artificial intelligence voice wake-up method of claim 7, wherein: the awakening judgment model is trained by utilizing a second voice data training sample set and awakening keyword text data, and background noise in the second voice data training sample set is larger than background noise in the first voice data training sample set.
9. The artificial intelligence voice wake-up method of claim 8, wherein: the method for performing wake-up judgment by using the wake-up judgment model comprises the following steps:
and inputting the awakening keywords detected by the keyword comparison model and the text data in the S2 into the trained awakening judgment model, and comprehensively carrying out awakening judgment by the awakening judgment model according to the detection result of the keyword comparison model and the text data.
10. The artificial intelligence voice wake-up method of claim 9, wherein: if the awakening judgment model judges that the detection result of the keyword comparison model and the text data both contain the awakening keyword, the awakening judgment model outputs an awakening instruction; otherwise, the awakening judgment model outputs a current status maintaining instruction.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011075074.3A CN112233656A (en) | 2020-10-09 | 2020-10-09 | Artificial intelligent voice awakening method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011075074.3A CN112233656A (en) | 2020-10-09 | 2020-10-09 | Artificial intelligent voice awakening method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112233656A true CN112233656A (en) | 2021-01-15 |
Family
ID=74120504
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011075074.3A Pending CN112233656A (en) | 2020-10-09 | 2020-10-09 | Artificial intelligent voice awakening method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112233656A (en) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105632486A (en) * | 2015-12-23 | 2016-06-01 | 北京奇虎科技有限公司 | Voice wake-up method and device of intelligent hardware |
CN105654943A (en) * | 2015-10-26 | 2016-06-08 | 乐视致新电子科技(天津)有限公司 | Voice wakeup method, apparatus and system thereof |
CN107221326A (en) * | 2017-05-16 | 2017-09-29 | 百度在线网络技术(北京)有限公司 | Voice awakening method, device and computer equipment based on artificial intelligence |
CN107346659A (en) * | 2017-06-05 | 2017-11-14 | 百度在线网络技术(北京)有限公司 | Audio recognition method, device and terminal based on artificial intelligence |
CN110364143A (en) * | 2019-08-14 | 2019-10-22 | 腾讯科技(深圳)有限公司 | Voice awakening method, device and its intelligent electronic device |
-
2020
- 2020-10-09 CN CN202011075074.3A patent/CN112233656A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105654943A (en) * | 2015-10-26 | 2016-06-08 | 乐视致新电子科技(天津)有限公司 | Voice wakeup method, apparatus and system thereof |
CN105632486A (en) * | 2015-12-23 | 2016-06-01 | 北京奇虎科技有限公司 | Voice wake-up method and device of intelligent hardware |
CN107221326A (en) * | 2017-05-16 | 2017-09-29 | 百度在线网络技术(北京)有限公司 | Voice awakening method, device and computer equipment based on artificial intelligence |
CN107346659A (en) * | 2017-06-05 | 2017-11-14 | 百度在线网络技术(北京)有限公司 | Audio recognition method, device and terminal based on artificial intelligence |
CN110364143A (en) * | 2019-08-14 | 2019-10-22 | 腾讯科技(深圳)有限公司 | Voice awakening method, device and its intelligent electronic device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110491416B (en) | Telephone voice emotion analysis and identification method based on LSTM and SAE | |
CN110415699A (en) | A kind of judgment method, device and electronic equipment that voice wakes up | |
US20180158449A1 (en) | Method and device for waking up via speech based on artificial intelligence | |
WO2022134833A1 (en) | Speech signal processing method, apparatus and device, and storage medium | |
CN111524527B (en) | Speaker separation method, speaker separation device, electronic device and storage medium | |
CN110767218A (en) | End-to-end speech recognition method, system, device and storage medium thereof | |
CN111223488B (en) | Voice wake-up method, device, equipment and storage medium | |
CN105632486A (en) | Voice wake-up method and device of intelligent hardware | |
CN106653031A (en) | Voice wake-up method and voice interaction device | |
CN103258535A (en) | Identity recognition method and system based on voiceprint recognition | |
CN111161726B (en) | Intelligent voice interaction method, device, medium and system | |
CN110414005B (en) | Intention recognition method, electronic device and storage medium | |
CN102945673A (en) | Continuous speech recognition method with speech command range changed dynamically | |
CN114550703A (en) | Training method and device of voice recognition system, and voice recognition method and device | |
CN111508493A (en) | Voice wake-up method and device, electronic equipment and storage medium | |
Liu et al. | Loss and double-edge-triggered detector for robust small-footprint keyword spotting | |
CN111179944B (en) | Voice awakening and age detection method and device and computer readable storage medium | |
CN112825250A (en) | Voice wake-up method, apparatus, storage medium and program product | |
CN110930997B (en) | Method for labeling audio by using deep learning model | |
CN112233656A (en) | Artificial intelligent voice awakening method | |
CN116564315A (en) | Voiceprint recognition method, voiceprint recognition device, voiceprint recognition equipment and storage medium | |
CN112017676A (en) | Audio processing method, apparatus and computer readable storage medium | |
CN111079705B (en) | Vibration signal classification method | |
CN114927128A (en) | Voice keyword detection method and device, electronic equipment and readable storage medium | |
CN114937449A (en) | Voice keyword recognition method and system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210115 |