CN110600008A - Voice wake-up optimization method and system - Google Patents

Abstract

The embodiment of the present invention provides an optimization method for voice wake-up. The method includes: constructing a second-level wake-up acoustic model, the second-level wake-up acoustic model includes a phoneme acoustic model and a word-level acoustic model; performing feature extraction on the received speech audio, and inputting the mentioned acoustic features into the second-level wake-up acoustic model The phoneme-level acoustic model in the phoneme-level acoustic model is used to extract the output features of the phoneme-level acoustic model; based on the output features of the phoneme-level acoustic model, it is used as the input of the word-level acoustic model in the secondary wake-up acoustic model to determine the confidence level of the wake-up word; when the confidence level When the preset wake-up threshold is exceeded, the voice audio is determined as the wake-up word for voice wake-up. The embodiment of the present invention also provides an optimization system for voice wake-up. The embodiment of the present invention directly reduces the dependence of the final classification effect on the accuracy of the phoneme modeling unit, and in the case of inaccurate phoneme classification, the wake word can still be correctly discriminated.

Description

Optimization method and system for voice wake-up

technical field

The invention relates to the field of intelligent voice dialogue, in particular to an optimization method and system for voice wake-up.

Background technique

Speech arousal usually uses deep neural networks to acoustically model basic acoustic units, which generally select phonemes.

In the above-described voice wake-up technology, the modeling unit is phoneme. First, the phoneme is predicted, classified, and processed; then the similarity between the processed sequence and the wake-up word sequence is calculated. If the similarity is greater than a certain threshold, wake up; otherwise Do not wake up.

In the process of realizing the present invention, the inventor found that there are at least the following problems in the related art:

This technique relies heavily on the accuracy of the acoustic model to classify the speech signal on the modeling unit. In the case of low signal-to-noise ratio, the classification accuracy of the acoustic model for phonemes is not high, so that it affects the arousal rate of scenes with low signal-to-noise ratio.

SUMMARY OF THE INVENTION

In order to at least solve the problem of low wake-up rate in low signal-to-noise ratio scenarios in the prior art.

In a first aspect, an embodiment of the present invention provides an optimization method for voice wake-up, including:

constructing a second-level wake-up acoustic model, where the second-level wake-up acoustic model includes a phoneme acoustic model and a word-level acoustic model;

Perform feature extraction on the received speech audio, input the mentioned acoustic features into the phoneme acoustic model in the secondary wake-up acoustic model, and extract the output features of the phoneme acoustic model;

Based on the output feature of the phoneme acoustic model, as the input of the word-level acoustic model in the secondary wake-up acoustic model, determine the confidence level of the wake-up word;

When the confidence level exceeds a preset wake-up threshold, the voice audio is determined as a wake-up word, and voice wake-up is performed.

In a second aspect, an embodiment of the present invention provides an optimization system for voice wake-up, including:

a model building program module for constructing a secondary arousal acoustic model, the secondary arousal acoustic model includes a phoneme acoustic model and a word-level acoustic model;

A feature extraction program module is used to perform feature extraction on the received speech audio, input the mentioned acoustic features into the phoneme acoustic model in the secondary wake-up acoustic model, and extract the output features of the phoneme acoustic model;

A confidence level determination program module, used for determining the confidence level of the wake-up word based on the output feature of the phoneme acoustic model as the input of the word-level acoustic model in the secondary wake-up acoustic model;

A wake-up program module, configured to determine the voice audio as a wake-up word and perform voice wake-up when the confidence level exceeds a preset wake-up threshold.

In a third aspect, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, The instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the method for optimizing wake-up by voice according to any embodiment of the present invention.

In a fourth aspect, an embodiment of the present invention provides a storage medium on which a computer program is stored, characterized in that, when the program is executed by a processor, the steps of the method for optimizing voice wake-up according to any embodiment of the present invention are implemented.

The beneficial effects of the embodiments of the present invention are: on the basis of one acoustic model, deep acoustic features extracted from speech signals of a certain length are used to input them into another classification model for direct classification, which directly reduces the impact of the final classification effect on phonemes. Depends on the accuracy of the modeling unit, in the case of inaccurate phoneme classification, the wake word can still be correctly discriminated.

Description of drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative efforts.

1 is a flowchart of a method for optimizing voice wake-up provided by an embodiment of the present invention;

FIG. 2 is a schematic structural diagram of an optimization system for voice wake-up provided by an embodiment of the present invention.

Detailed ways

In order to make the purposes, 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 accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments These are some embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

FIG. 1 is a flowchart of an optimization method for voice wake-up provided by an embodiment of the present invention, including the following steps:

S11: Construct a second-level wake-up acoustic model, where the second-level wake-up acoustic model includes a phoneme acoustic model and a word-level acoustic model;

S12: Perform feature extraction on the received speech audio, input the mentioned acoustic features into the phoneme acoustic model in the secondary wake-up acoustic model, and extract the output features of the phoneme acoustic model;

S13: Based on the output feature of the phoneme acoustic model, as the input of the word-level acoustic model in the secondary wake-up acoustic model, determine the confidence level of the wake-up word;

S14: When the confidence level exceeds a preset wake-up threshold, determine the voice audio as a wake-up word, and perform voice wake-up.

In this embodiment, it is different from a modeled acoustic model, and is also different from the comparison of the results of two general acoustic models. The output results of the two models are not used for comparison, because in the case of a low signal-to-noise ratio Therefore, the selection of multiple acoustic models will not significantly improve the classification accuracy of phonemes.

For step S11, different from a modeling phoneme acoustic model, on this basis, a second-level wake-up acoustic model is constructed, and the second-level wake-up acoustic model includes a phoneme acoustic model and a word-level acoustic model, wherein the task of the acoustic model is to Calculate P(O|W), the probability of generating a speech waveform for the model. Acoustic model is an important part of speech recognition system, it occupies most of the computational overhead of speech recognition and determines the performance of speech recognition system. Traditional speech recognition systems generally use an acoustic model based on GMM-HMM, where GMM is used to model the distribution of speech acoustic features, and HMM is used to model the timing of speech signals. After the rise of deep learning in 2006, Deep Neural Networks (DNN) have been applied to speech acoustic models. The phoneme acoustic model determines the probability of each phoneme in the speech waveform, and the word-level acoustic model determines the probability of each word in the speech waveform.

For step S12, in order to receive the real-time voice wake-up, the smart device needs to collect the voice and audio in the environment in real time, perform feature extraction on the voice and audio received after the collection, and input the extracted acoustic features into the secondary wake-up acoustics The phoneme acoustic model in the model, and the output features of the phoneme acoustic model, such as the phoneme sequence of the speech audio, are extracted.

For step S13, based on the output features of the phoneme acoustic model, such as the audio sequence output in step S12, as the input of the word-level acoustic model in the secondary wake-up acoustic model, classify by another acoustic model, With a clear classification, the confidence of the user audio as a wake-up word can be more accurately determined.

For step S14, when the confidence level exceeds a preset wake-up threshold, the voice audio is determined as a wake-up word, and voice wake-up is performed. .

It can be seen from this embodiment that on the basis of one acoustic model, deep acoustic features extracted from speech signals of a certain length are used to input them into another classification model for direct classification, which directly reduces the final classification effect on phoneme construction. Dependence on the accuracy of the modular unit, in the case of inaccurate phoneme classification, the wake word can still be correctly discriminated.

As an implementation manner, in this embodiment, one of the two-level wake-up acoustic models is a phoneme acoustic model, and the other acoustic model is a word-level acoustic model.

In this embodiment, one of the acoustic models is a phoneme acoustic model, and the other acoustic model is a word-level acoustic model. After repeated experiments, the inventor found that using the phoneme acoustic model for wake-up word recognition, in the case of low signal-to-noise ratio, the wake-up performance is low, and the recognition performance is heavily dependent on the accuracy of phoneme classification by the phoneme acoustic model. On the basis of the phoneme acoustic model, a word-level acoustic model is connected to directly classify the wake-up words. Even if the phoneme classification is inaccurate, the recognition effect of the wake-up words can be improved by direct classification and make up for the single phoneme. Inadequate acoustic models.

As an implementation manner, in this embodiment, after extracting the output features of the phoneme acoustic model, the method further includes:

Send the output features of each frame to the feature accumulator;

When the number of frames of speech and audio in the feature accumulator is accumulated to reach a preset threshold, the output features in the feature accumulator are spliced into a one-dimensional feature;

The one-dimensional features are input to the word-level acoustic model to complete the coupling of the two models.

In this embodiment, after the features output by the phoneme acoustic model are extracted, the output features of each frame are sent to a feature accumulator for accumulation. When a certain number of frames are accumulated, these features are spliced into one-dimensional complete features, and the one-dimensional features are input into the word-level acoustic model, so that the two acoustic models can be coupled to ensure the use of the two models.

As an implementation manner, in this embodiment, before the feature extraction is performed on the received speech audio, the method further includes:

Receive the audio signal in real time according to the acoustic sensor, and determine whether the audio signal is voice audio through the voice endpoint detection model;

When the audio signal is speech audio, the acoustic feature extraction is performed on the received dialogue speech.

Since voice wake-up needs to detect the received audio in real time, it is very wasteful of resources to detect if the audio is received. Before extracting the features of the received voice and audio, receive the audio signal in real time according to the acoustic sensor in the smart device, detect whether the audio signal is the voice audio of the user speaking, ensure that there is a user speaking, and then go to the detection to avoid receiving the audio The signal goes to voice wake-up detection to improve the efficiency of audio wake-up detection.

FIG. 2 is a schematic structural diagram of a voice wake-up optimization system according to an embodiment of the present invention. The system can execute the voice wake-up optimization method described in any of the foregoing embodiments, and is configured in a terminal.

A voice wake-up optimization system provided in this embodiment includes: a model building program module 11 , a feature extraction program module 12 , a confidence level determination program module 13 and a wake-up program module 14 .

Wherein, the model building program module 11 is used to build a secondary wake-up acoustic model, which includes a phoneme acoustic model and a word-level acoustic model; the feature extraction program module 12 is used to perform feature extraction on the received speech audio , the mentioned acoustic feature is input to the phoneme acoustic model in the described secondary wake-up acoustic model, and the output feature of the described phoneme acoustic model is extracted; the confidence level determination program module 13 is used for the output feature based on the described phoneme acoustic model, As the input of the word-level acoustic model in the second-level wake-up acoustic model, the confidence of the wake-up word is determined; the wake-up program module 14 is configured to determine the voice audio as wake-up when the confidence exceeds a preset wake-up threshold word for voice wake-up.

Further, the one of the acoustic models is a phoneme acoustic model, and the one of the acoustic models is a word-level acoustic model.

Further, after the feature extraction program module, the system further includes: a feature accumulation program module for:

Send the output features of each frame to the feature accumulator;

When the number of frames of speech and audio in the feature accumulator is accumulated to reach a preset threshold, the output features in the feature accumulator are spliced into a one-dimensional feature;

The one-dimensional features are input to the other acoustic model to complete the coupling of the two models.

Further, the feature extraction program module is also used for:

Receive the audio signal in real time according to the acoustic sensor, and determine whether the audio signal is voice audio through the voice endpoint detection model;

When the audio signal is speech audio, the acoustic feature extraction is performed on the received dialogue speech.

Embodiments of the present invention further provide a non-volatile computer storage medium, where the computer storage medium stores computer-executable instructions, and the computer-executable instructions can execute the optimization method for voice wake-up in any of the above method embodiments;

As an embodiment, the non-volatile computer storage medium of the present invention stores computer-executable instructions, and the computer-executable instructions are set to:

constructing a second-level wake-up acoustic model, where the second-level wake-up acoustic model includes a phoneme acoustic model and a word-level acoustic model;

Perform feature extraction on the received speech audio, input the mentioned acoustic features into the phoneme acoustic model in the secondary wake-up acoustic model, and extract the output features of the phoneme acoustic model;

Based on the output feature of the phoneme acoustic model, as the input of the word-level acoustic model in the secondary wake-up acoustic model, determine the confidence level of the wake-up word;

When the confidence level exceeds a preset wake-up threshold, the voice audio is determined as a wake-up word, and voice wake-up is performed.

As a non-volatile computer-readable storage medium, it can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as program instructions/modules corresponding to the methods in the embodiments of the present invention. One or more program instructions are stored in a non-volatile computer-readable storage medium, and when executed by a processor, perform the optimization method for voice wake-up in any of the above method embodiments.

The non-volatile computer-readable storage medium may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created according to the use of the device. data etc. In addition, the non-volatile computer-readable storage medium may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the non-volatile computer-readable storage medium may optionally include memory located remotely from the processor, which may be connected to the device through a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

An embodiment of the present invention further provides an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor , the instructions are executed by the at least one processor, so that the at least one processor can execute the steps of the method for optimizing wake-up by voice according to any embodiment of the present invention.

The clients in the embodiments of the present application exist in various forms, including but not limited to:

(1) Mobile communication equipment: This type of equipment is characterized by having mobile communication functions, and its main goal is to provide voice and data communication. Such terminals include: smart phones, multimedia phones, feature phones, and low-end phones.

(2) Ultra-mobile personal computer equipment: This type of equipment belongs to the category of personal computers, has computing and processing functions, and generally has the characteristics of mobile Internet access. Such terminals include: PDAs, MIDs, and UMPC devices, such as tablet computers.

(3) Portable entertainment equipment: This type of equipment can display and play multimedia content. Such devices include: audio and video players, handheld game consoles, e-books, as well as smart toys and portable car navigation devices.

(4) Other electronic devices with data processing functions.

In this document, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such existence between these entities or operations. The actual relationship or sequence. Furthermore, the terms "comprising" and "comprising" include not only those elements, but also other elements not expressly listed, or elements inherent to such a process, method, article or apparatus. Without further limitation, an element defined by the phrase "comprises" does not preclude the presence of additional identical elements in a process, method, article, or device that includes the element.

The device embodiments described above are only illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in One place, or it can be distributed over multiple network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution in this embodiment. Those of ordinary skill in the art can understand and implement it without creative effort.

From the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and certainly can also be implemented by hardware. Based on this understanding, the above-mentioned technical solutions can be embodied in the form of software products in essence or the parts that make contributions to the prior art, and the computer software products can be stored in computer-readable storage media, such as ROM/RAM, magnetic A disc, an optical disc, etc., includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform the methods described in various embodiments or some parts of the embodiments.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, but not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that it can still be Modifications are made to the technical solutions described in the foregoing embodiments, or some technical features thereof are equivalently replaced; and these modifications or replacements do not make the essence of the corresponding technical solutions depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (8)

1. An optimization method for voice wake-up, comprising: constructing a second-level wake-up acoustic model, where the second-level wake-up acoustic model includes a phoneme acoustic model and a word-level acoustic model; Perform feature extraction on the received speech audio, input the mentioned acoustic features into the phoneme acoustic model in the secondary wake-up acoustic model, and extract the output features of the phoneme acoustic model; Based on the output feature of the phoneme acoustic model, as the input of the word-level acoustic model in the secondary wake-up acoustic model, determine the confidence level of the wake-up word; When the confidence level exceeds a preset wake-up threshold, the voice audio is determined as a wake-up word, and voice wake-up is performed. 2. The method according to claim 1, wherein after the extracting the output features of the phoneme acoustic model, the method further comprises: Send the output features of each frame to the feature accumulator; When the number of frames of speech and audio in the feature accumulator is accumulated to reach a preset threshold, the output features in the feature accumulator are spliced into a one-dimensional feature; The one-dimensional features are input to the word-level acoustic model to complete the coupling of the two models. 3. The method according to claim 1, wherein, before the feature extraction is performed on the received speech audio, the method further comprises: Receive the audio signal in real time according to the acoustic sensor, and determine whether the audio signal is voice audio through the voice endpoint detection model; When the audio signal is speech audio, the acoustic feature extraction is performed on the received dialogue speech. 4. An optimized system for voice wake-up, comprising: a model building program module for constructing a secondary arousal acoustic model, the secondary arousal acoustic model includes a phoneme acoustic model and a word-level acoustic model; A feature extraction program module is used to perform feature extraction on the received speech audio, input the mentioned acoustic features into the phoneme acoustic model in the secondary wake-up acoustic model, and extract the output features of the phoneme acoustic model; A confidence level determination program module, used for determining the confidence level of the wake-up word based on the output feature of the phoneme acoustic model as the input of the word-level acoustic model in the secondary wake-up acoustic model; A wake-up program module, configured to determine the voice audio as a wake-up word when the confidence level exceeds a preset wake-up threshold, and perform voice wake-up. 5. The system according to claim 4, wherein, after the feature extraction program module, the system further comprises: a feature accumulation program module for: Send the output features of each frame to the feature accumulator; When the number of frames of speech and audio in the feature accumulator reaches a preset threshold, the output features in the feature accumulator are spliced into a one-dimensional feature; The one-dimensional features are input to the other acoustic model to complete the coupling of the two models. 6. The system of claim 4, wherein the feature extraction program module is further used to: Receive the audio signal in real time according to the acoustic sensor, and determine whether the audio signal is voice audio through the voice endpoint detection model; When the audio signal is voice audio, perform acoustic feature extraction on the received dialogue voice. 7. An electronic device comprising: at least one processor, and a memory communicatively coupled to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions Executed by the at least one processor to enable the at least one processor to perform the steps of the method of any of claims 1-3. 8. A storage medium on which a computer program is stored, characterized in that, when the program is executed by a processor, the steps of the method according to any one of claims 1-3 are implemented.