A speech recognition method, apparatus, storage medium and device
By splitting the speech stream into multiple paths and using default and recommended scene parameters, the accuracy problem of the speech recognition engine when the scene changes is solved, achieving automatic adjustment and efficient recognition, and improving the user experience.
Patent Information
- Application Number
- CN202211725390.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-12-28
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2042-12-28
AI Technical Summary
Existing speech recognition engines struggle to automatically adjust scene parameters when faced with changes in the scene, especially when the speaker changes, leading to decreased recognition accuracy and an inability to meet business needs.
The user-input voice stream is divided into multiple streams. One stream is recognized using default scene parameters, while the other stream undergoes scene detection to obtain recommended scene parameters. These recommended scene parameters are then used to recognize the voice stream, ensuring high accuracy even when the scene changes.
It automatically adjusts scene parameters without requiring manual configuration from the user, improving the accuracy of speech recognition, reducing the user's understanding cost and usage threshold, and enhancing the user experience.
Smart Images

Figure CN116052650B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of speech recognition technology, and in particular to a speech recognition method, apparatus, storage medium and computing device. Background Technology
[0002] Speech recognition technology, also known as automatic speech recognition (ASR), computer speech recognition, or speech to text (STT), aims to automatically convert human speech into corresponding text using computers.
[0003] Speech recognition technology, with its advantages of intelligence, efficiency, and user-friendliness, has been widely applied in various fields of production and daily life, providing convenience for users. Correspondingly, the widespread application of speech recognition has also generated diverse demands. For example, speech recognition engines need to support multiple languages, regional dialects, and different fields (such as science and technology, healthcare, and education), but a single, universal engine often struggles to automatically handle the effects of various scenarios.
[0004] Currently, some speech recognition engines offer various scene parameters. Users can configure scene parameters in speech recognition products such as voice recorders to improve recognition performance in specific scenarios. For example, users can configure the audio mode of a meeting to a dialect mode, such as Cantonese. However, if unexpected situations arise, such as a change in speakers during a meeting, pre-set scene parameters may lead to a decrease in speech recognition accuracy, making it difficult to meet business needs. Summary of the Invention
[0005] The main purpose of this application is to provide a speech recognition method, device, storage medium, and computing device that can adaptively adjust scene parameters as the scene changes, ensuring the accuracy of speech recognition and meeting business needs.
[0006] In a first aspect, this application provides a speech recognition method, including:
[0007] Receives a voice stream input by the user, which is divided into multiple streams;
[0008] The first speech segment of the first speech stream in the multi-speech stream is identified using default scene parameters to obtain a first recognition result, and scene detection is performed based on the second speech stream in the multi-speech stream to obtain recommended scene parameters.
[0009] The recommended scene parameters are used to identify the second speech segment of the first speech stream to obtain the second recognition result.
[0010] In some possible implementations, the step of performing scene detection based on the second speech stream among the multiple speech streams to obtain recommended scene parameters includes:
[0011] Obtain the third speech segment corresponding to the time window from the second speech stream, perform scene detection based on the third speech segment, and obtain recommended scene parameters; or...
[0012] When the speaker information changes, scene detection is performed based on the second voice stream in the multi-channel voice stream to obtain recommended scene parameters.
[0013] In some possible implementations, the step of using recommended scene parameters to identify a second speech segment of the first speech stream to obtain a second recognition result includes:
[0014] When the recommended scene parameters determined based on the third speech segments corresponding to multiple time windows in the second speech stream are the same, the recommended scene parameters are used to identify the second speech segment of the first speech stream to obtain a second recognition result.
[0015] In some possible implementations, the step of using recommended scene parameters to identify a second speech segment of the first speech stream to obtain a second recognition result includes:
[0016] When the confidence level of the recommended scene parameters meets the first condition, the recommended scene parameters are automatically used to identify the second speech segment of the first speech stream to obtain a second recognition result.
[0017] In some possible implementations, the step of using recommended scene parameters to identify a second speech segment of the first speech stream to obtain a second recognition result includes:
[0018] When the confidence level of the recommended scenario parameters meets the second condition, the recommended scenario parameters are displayed to the user.
[0019] When a user triggers a parameter update operation, the recommended scenario parameters are used to identify the second speech segment of the first speech stream to obtain a second recognition result.
[0020] In some possible implementations, the scene parameters include one or more of the following: language, dialect, domain, near and far field information, or speaker information.
[0021] In some possible implementations, the speech includes conference speech, and the step of performing scene detection based on a second speech stream from the multiple speech streams to obtain recommended scene parameters includes:
[0022] Scene detection is performed based on the second audio stream in the multi-channel audio stream and the conference information to obtain recommended scene parameters.
[0023] Secondly, this application provides a voice recognition device, comprising:
[0024] A voice acquisition module is used to receive a voice stream input by the user, which is divided into multiple channels;
[0025] The recognition module is used to recognize the first speech segment of the first speech stream in the multi-speech stream using default scene parameters, and obtain the first recognition result;
[0026] The scene detection module is used to perform scene detection based on the second voice stream in the multi-channel voice stream to obtain recommended scene parameters;
[0027] The recognition module is further configured to use recommended scene parameters to recognize the second speech segment of the first speech stream and obtain a second recognition result.
[0028] In some possible implementations, the scene detection module is specifically used for:
[0029] Obtain the third speech segment corresponding to the time window from the second speech stream, perform scene detection based on the third speech segment, and obtain recommended scene parameters; or...
[0030] When the speaker information changes, scene detection is performed based on the second voice stream in the multi-channel voice stream to obtain recommended scene parameters.
[0031] In some possible implementations, the identification module is specifically used for:
[0032] When the recommended scene parameters determined based on the third speech segments corresponding to multiple time windows in the second speech stream are the same, the recommended scene parameters are used to identify the second speech segment of the first speech stream to obtain a second recognition result.
[0033] In some possible implementations, the identification module is specifically used for:
[0034] When the confidence level of the recommended scene parameters meets the first condition, the recommended scene parameters are automatically used to identify the second speech segment of the first speech stream to obtain a second recognition result.
[0035] In some possible implementations, the device further includes:
[0036] The display module is used to display the recommended scenario parameters to the user when the confidence level of the recommended scenario parameters meets the second condition.
[0037] The identification module is specifically used for:
[0038] When a user triggers a parameter update operation, the recommended scenario parameters are used to identify the second speech segment of the first speech stream to obtain a second recognition result.
[0039] In some possible implementations, the scene parameters include one or more of the following: language, dialect, domain, near and far field information, or speaker information.
[0040] In some possible implementations, the voice includes conference audio, and the scene detection module is specifically used for:
[0041] Scene detection is performed based on the second audio stream in the multi-channel audio stream and the conference information to obtain recommended scene parameters.
[0042] Thirdly, this application provides a computing device, including: a processor, a memory, and a system bus;
[0043] The processor and the memory are connected via the system bus;
[0044] The memory is used to store one or more programs, the one or more programs including instructions, which, when executed by the processor, cause the processor to perform any of the above-described implementations of the speech recognition method.
[0045] Fourthly, this application provides a computer-readable storage medium storing instructions that, when executed on a computing device, cause the computing device to perform any of the above-described implementations of the speech recognition method.
[0046] Fifthly, this application also provides a computer program product that, when run on a computing device, causes the computing device to execute any of the above-described implementations of the speech recognition method.
[0047] As can be seen from the above technical solution, this application has at least the following beneficial effects:
[0048] This application provides a speech recognition method. Specifically, it first receives a speech stream input by a user, which is divided into multiple streams. Then, it uses default scene parameters to recognize a first speech segment of a first speech stream in the multiple speech streams to obtain a first recognition result. Next, it performs scene detection based on a second speech stream in the multiple speech streams to obtain recommended scene parameters. Then, it uses the recommended scene parameters to recognize a second speech segment of the first speech stream to obtain a second recognition result.
[0049] The method described above divides the audio stream into multiple paths. One path uses default scene parameters for speech recognition, ensuring basic functionality. The other path is used for scene detection to obtain recommended scene parameters. This ensures that even if unexpected situations arise, such as speaker changes during a meeting, the recommended scene parameters can still be used for speech recognition, guaranteeing accuracy and meeting business needs. This method eliminates the need for manual configuration, reducing the user's understanding of various optimization modes, lowering the barrier to entry, increasing usage frequency, and enhancing the user experience. Attached Figure Description
[0050] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0051] Figure 1 A flowchart illustrating a speech recognition method provided in an embodiment of this application;
[0052] Figure 2 A schematic diagram of an information prompt interface provided in an embodiment of this application;
[0053] Figure 3 A schematic diagram of another information prompt interface provided in an embodiment of this application;
[0054] Figure 4 This is a schematic diagram of the structure of a speech recognition device provided in an embodiment of this application. Detailed Implementation
[0055] The terms "first" and "second" used in the embodiments of this application are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature.
[0056] First, some technical terms involved in the embodiments of this application will be introduced.
[0057] Automatic speech recognition (ASR) refers to the process of automatically converting human speech into corresponding text. Speech recognition is widely used in fields such as science and technology, medicine, and education. For example, in the medical field, voice robots can recognize patients' questions and provide answers based on the recognition results. In education, voice robots can assist students in practicing their spoken language.
[0058] Speech recognition is based on a speech recognition engine. General-purpose speech recognition engines struggle to automatically balance recognition performance (such as accuracy) across various scenarios. To address this, many speech recognition products or services (such as voice recorders) offer various scenario parameters to allow users to select appropriate parameters, thereby improving recognition performance in these scenarios.
[0059] Scene parameters include, but are not limited to, one or more of the following: language, dialect, domain, near and far field information, or speaker information. Languages can include one or more of Chinese, English, Japanese, Korean, French, and German. Different dialects can be included for different languages. For example, Chinese dialects can include Cantonese, Henan dialect, Sichuan dialect, and Beijing dialect (also known as Standard Mandarin). Similarly, English dialects can include British English and American English. British English can be further divided into London accent and slang; similarly, American English can be further subdivided according to region. When the phonemic information of some regional dialects differs significantly from Standard Mandarin or Standard Mandarin, it is difficult to achieve good recognition results using a general-purpose speech recognition engine. Therefore, a customized dialect acoustic model is required.
[0060] Specialized terminology in key industries such as healthcare and education is prone to errors in general-purpose speech recognition engines. Therefore, multiple domain-specific models can be trained for these key areas and used as a supplement to the main model during speech recognition engine decoding. In this way, after the user sets the domain, the corresponding domain model can be used for decoding.
[0061] Far-field mode is suitable for scenarios requiring a wider sound pickup range, such as large conference rooms, but it is more prone to false noise absorption; near-field mode is suitable for scenarios such as personal presentations, avoiding the false absorption of distant noise. Therefore, users can set either far-field or near-field mode according to their business needs.
[0062] In multi-person conversation scenarios, speaker separation can also be performed to distinguish each person's speech in the final recognition result. During speech recognition, if speaker information, such as the number of speakers, is set—for example, by combining meeting information to extract the number of speakers—the speaker separation effect can be significantly improved.
[0063] While the above solution offers a rich set of functional modes, unexpected situations may arise. For example, if the meeting mode is set to Sichuan dialect because all attendees are speaking Sichuan dialect, but the leader gives a summary in Mandarin at the end of the meeting, the pre-set scenario parameters (pre-selected functional mode, such as Sichuan dialect mode) may lead to a decrease in speech recognition accuracy, making it difficult to meet business needs. In other words, pre-set modes may have the opposite effect.
[0064] In view of this, this application provides a speech recognition method. This method can be executed by a computing device. The computing device can be a terminal, including but not limited to smartphones, smart wearable devices (such as smartwatches), tablets, laptops, or desktop computers. In some examples, the computing device can also be a server. It should be noted that the speech recognition method of this application can be executed by a single computing device or by multiple computing devices (e.g., a cluster of multiple computing devices).
[0065] Specifically, the computing device receives a voice stream input by the user, which is divided into multiple streams. The computing device then uses default scene parameters to recognize a first voice segment of a first voice stream in the multiple voice streams to obtain a first recognition result. It also performs scene detection based on a second voice stream in the multiple voice streams to obtain recommended scene parameters. The computing device then uses the recommended scene parameters to recognize a second voice segment of the first voice stream to obtain a second recognition result.
[0066] This method divides the audio stream into multiple paths. One path uses default scene parameters for speech recognition to ensure basic functionality, while the other path is used for scene detection to obtain recommended scene parameters. This ensures that even if unexpected situations arise, such as a change in speaker during a meeting, the recommended scene parameters can still be used for speech recognition, guaranteeing accuracy and meeting business needs. This solves the problem that pre-set modes might have a counterproductive effect.
[0067] Furthermore, this method eliminates the need for manual configuration by users, addressing the issues of end-users wanting smarter products but being unwilling to explore features with high learning curves and understanding requirements, and the fact that most users consistently use default scene parameters (default modes), resulting in recognition results that often fail to meet user expectations. This method reduces the learning curve for users of various effect optimization modes, lowers the barrier to entry, increases usage frequency, and enhances the user experience.
[0068] To make the technical solution of this application clearer and easier to understand, the method of the embodiment of this application will be described below with reference to the accompanying drawings.
[0069] See Figure 1 The flowchart shown illustrates a speech recognition method, which includes the following steps:
[0070] S102: The computing device receives a voice stream input by the user.
[0071] A voice stream refers to a stream of audio data. This voice stream can be a real-time audio stream or a recorded audio stream. For example, in a meeting scenario, users (such as attendees) can join the meeting via computing devices (such as smartphones or laptops), and the computing devices can capture the user's voice stream in real time via a microphone. As another example, in a speech contest, computing devices can record the voice streams of each speaker using a voice recorder.
[0072] S104: The computing device splits the voice stream into multiple streams.
[0073] Specifically, a computing device can divide an audio stream into multiple streams by copying it. Considering storage and transmission overhead, the computing device can also divide the audio stream into multiple streams by sampling. Specifically, the computing device can save the original audio stream and sample the audio stream according to a configured sampling rate to generate a new audio stream.
[0074] It should be noted that S104 is an optional step in the embodiments of this application. The method of the embodiments of this application may also be performed without executing S104. For example, a computing device may also acquire multiple voice streams through multiple microphones.
[0075] S106: The computing device uses default scene parameters to identify the first speech segment of the first speech stream in the multi-speech stream and obtains the first recognition result.
[0076] Scene parameters include one or more of the following: language, dialect, domain, near / far field information, or speaker information. Default scene parameters can be the default values for the aforementioned scene parameters. For example, the default value for the language parameter can be Chinese. It should be noted that some scene parameters can have default values. For example, the default value for the domain parameter can be a default value.
[0077] Specifically, the computing device, based on default scene parameters (general parameter settings), calls the corresponding model of the speech recognition engine, such as the main model, to perform speech recognition on the first speech stream among multiple speech streams. The first speech stream can be a lossless speech stream, thus ensuring the integrity of the speech recognition result. When performing speech recognition on the first speech stream, the computing device typically performs speech recognition on a segment-by-speech basis. For example, the computing device can first call the main model of the speech recognition engine based on the default scene parameters to perform speech recognition on the first speech segment of the first speech stream, thereby obtaining the first recognition result.
[0078] S108: The computing device performs scene detection based on the second voice stream in the multi-channel voice stream to obtain recommended scene parameters.
[0079] When none of the multiple speech streams are lossless, the computing device can identify any one of the speech streams as the second speech stream for scene detection. When the speech streams other than the original speech streams are sampled speech streams, the computing device can identify the sampled speech streams as the second speech stream.
[0080] The computing device can perform speech recognition on the second speech stream in a multi-channel speech stream, and perform pattern matching based on the recognition results to obtain recommended scene parameters. These recommended scene parameters refer to the recommended parameter values for the scene parameters.
[0081] For specific domains, computing devices can convert the second-channel speech stream into text using speech recognition. Through text analysis, when the text matches domain-related words a set number of times, the recommended domain can be determined. For example, if the text matches "metaverse" or "Great Hall of the People" a set number of times, the "domain" in the recommended scenario parameters can be determined to be either science and technology or politics.
[0082] In a meeting scenario, when the audio is conference audio, the computing device can also perform scene detection based on the second audio stream from multiple audio streams and meeting information to determine the recommended domain. The meeting information can include meeting-related presentation materials, such as PowerPoint presentations or Word document projections. If the word count of these materials exceeds a set threshold, a text clustering algorithm can be used to determine the similarity between the meeting-related presentation materials and existing domain models. Based on this similarity, the recommended domain can be determined.
[0083] If the text does not match any domain-related words, or the number of matches does not reach the set number, or the similarity between the conference-related presentation materials and the existing domain model does not reach the set similarity, then the recommended domain can be determined by combining historical conference results. Specifically, the computing device obtains the clustering scores of the conference results from M conferences prior to the current conference for the current account, as well as the clustering scores of the speech recognition results for the current conference within a set time period (e.g., 10 minutes). The above clustering scores are weighted and summed to obtain the similarity with the existing domain model. Based on this similarity, the recommended domain can be determined.
[0084] Regarding speaker information, computing devices can detect the number of speakers or their speaking speed. When a computing device detects the number of speakers, it can use this number as a parameter to connect with the speech recognition engine, thereby improving speaker separation. Specifically, the computing device can obtain the number of speakers based on its own product functions; for example, in a meeting scenario, it can obtain the number of speakers based on the number of online participants in the meeting information. The computing device can also obtain the number of speakers by performing semantic analysis on the text of the speech recognition results. For example, in customer service or interview scenarios, the computing device can determine the number of speakers by combining key information mentioned in the dialogue. Finally, the computing device can count the number of words in the text of the speech recognition results and then determine the speaker's speaking speed based on the number of words and the duration of the speech.
[0085] For near and far field information, computing devices can combine speaker information, signal-to-noise ratio (SNR) detection, or voice activity detection (VAD) to determine recommended near and far field information. VAD plays a crucial role in speech signal processing, such as speech enhancement and speech recognition. Its function is to identify speech segments and non-speech segments from a speech signal (clean or noisy). A VAD system typically consists of two parts: feature extraction and speech / non-speech decision. Commonly used feature extraction methods include energy-based feature extraction, frequency domain-based feature extraction, cepstral-based feature extraction, harmonic-based feature extraction, or long-term information-based feature extraction. Energy-based feature extraction is often implemented in hardware, while spectral (spectral and cepstral) features perform well at low SNRs. When the SNR reaches 0dB, harmonic and long-term information-based features are more robust. Decision criteria can be categorized into threshold-based criteria, statistical model-based criteria, or machine learning-based criteria.
[0086] Computing devices can remove noise from noisy speech signals to obtain clean speech signals, thus achieving speech enhancement. The common approach is to use the VAD function to obtain non-speech segments, which can be considered pure noise segments. Useful information can then be extracted from the pure noise signal, such as by performing a Fourier transform to obtain the noise spectrum, before further processing.
[0087] To facilitate understanding, the process of determining the recommended near-field and far-field information is illustrated below with an example. In this example, the computing device continuously monitors the signal-to-noise ratio (SNR) of the second speech stream. When it detects a persistent SNR below the SNR threshold (indicating high noise), and the speaker information indicates that most of the time the speaker is a single character with occasional fragmented segments of other characters, and the VAD features also show differences within the same time period, it predicts that the current scene is one where no single person is speaking and there is ambient noise. If the confidence level of the prediction result meets the requirements, the recommended near-field information can be determined, instructing the computing device to switch from far-field mode to near-field mode.
[0088] The above are some specific implementations of how a computing device performs scene detection and obtains recommended scene parameters based on the second voice stream in the multi-channel voice stream and conference information. In other possible implementations of this application, the computing device may also obtain recommended scene parameters through other methods.
[0089] It should be noted that the scene can typically change dynamically. To achieve adaptive updates of scene parameters, the computing device can continuously detect the scene on the second audio stream to obtain corresponding recommended scene parameters. This application also provides various scene detection triggering mechanisms, which are illustrated below.
[0090] In one possible implementation, the computing device can acquire a third speech segment corresponding to a time window from the second speech stream, perform scene detection based on the third speech segment, and obtain recommended scene parameters. The time window can be a fixed time window; considering periods of silence, the fixed time window can be set to 10 minutes. Thus, the computing device can periodically trigger scene detection according to the aforementioned time window.
[0091] In another possible implementation, the computing device can also perform scene detection based on the second speech stream in the multi-speech stream when the speaker information changes, and obtain recommended scene parameters.
[0092] The above S106 and S108 can be executed in parallel or sequentially according to a set execution order. This embodiment does not impose any restrictions on this.
[0093] S110: The computing device uses recommended scenario parameters to recognize the second speech segment of the first speech stream and obtains a second recognition result.
[0094] In this embodiment, to prevent accidental switching, the computing device can perform scene detection within multiple time windows. When the recommended scene parameters determined based on the third speech segments corresponding to multiple time windows in the second speech stream are the same, the computing device uses the recommended scene parameters to recognize the second speech segment of the first speech stream, obtaining a second recognition result. This can reduce the impact of mode switching (scene parameter updates) on the performance of speech recognition, especially the impact of switching acoustic models on the performance of speech recognition.
[0095] Furthermore, when determining the recommended scenario parameters, the computing device can also determine the confidence level of those recommended scenario parameters.
[0096] When the confidence level meets the first condition, such as when the confidence level is greater than the first threshold, the computing device can automatically use the recommended scene parameters to recognize the second speech segment of the first speech stream and obtain a second recognition result. Furthermore, the computing device can also indicate that the scene parameter update is complete (mode switching complete). See also Figure 2 The diagram shown is a schematic of the information prompt interface. The information prompt interface 200 includes a pop-up window 202, which includes a prompt message 204, specifically: "Mode A has been intelligently matched for you. You will soon experience better voice recognition effects."
[0097] When the confidence level meets the second condition, such as when the confidence level is greater than the second threshold and less than or equal to the first threshold, the computing device can display the recommended scene parameters to the user. When the user triggers a parameter update operation, the recommended scene parameters are used to recognize the second speech segment of the first speech stream to obtain a second recognition result. See also Figure 3 The diagram shows an information prompt interface. Information prompt interface 300 includes a pop-up window 302, which contains a prompt message 304: "Mode A has been intelligently matched for you. Click 'Confirm' to experience better voice recognition." When the user clicks the confirmation control 306, the scene parameters are updated to the recommended scene parameters, completing the mode switch and achieving better voice recognition results.
[0098] Based on the above description, the speech recognition method provided in this application divides the speech stream into multiple paths. One path uses default scene parameters for speech recognition to ensure basic functionality, while the other path is used for scene detection to obtain recommended scene parameters. This ensures that even if unexpected situations arise, such as a change in the speaker during a meeting, the recommended scene parameters can still be used for speech recognition, guaranteeing accuracy and meeting business needs. This solves the problem that pre-set modes may have a counterproductive effect. Furthermore, this method eliminates the need for manual configuration by the user, reducing the understanding cost of various effect optimization modes, lowering the barrier to entry, increasing usage frequency, and improving user experience.
[0099] Based on the methods provided in the embodiments of this application, the embodiments of this application also provide a speech recognition device corresponding to the above methods. The units / modules described in the embodiments of this application can be implemented in software or hardware. The names of the units / modules do not, in some cases, constitute a limitation on the unit / module itself.
[0100] See Figure 4 This is a schematic diagram of the structure of a voice recognition device provided in an embodiment of this application. The voice recognition device 400 includes:
[0101] The voice acquisition module 402 is used to receive the voice stream input by the user, and the voice stream is divided into multiple channels;
[0102] The recognition module 404 is used to recognize the first speech segment of the first speech stream in the multi-speech stream using default scene parameters, and obtain the first recognition result;
[0103] Scene detection module 406 is used to perform scene detection based on the second voice stream in the multi-channel voice stream to obtain recommended scene parameters;
[0104] The recognition module 404 is further configured to use recommended scene parameters to recognize the second speech segment of the first speech stream and obtain a second recognition result.
[0105] In some possible implementations, the scene detection module 406 is specifically used for:
[0106] Obtain the third speech segment corresponding to the time window from the second speech stream, perform scene detection based on the third speech segment, and obtain recommended scene parameters; or...
[0107] When the speaker information changes, scene detection is performed based on the second voice stream in the multi-channel voice stream to obtain recommended scene parameters.
[0108] In some possible implementations, the identification module 404 is specifically used for:
[0109] When the recommended scene parameters determined based on the third speech segments corresponding to multiple time windows in the second speech stream are the same, the recommended scene parameters are used to identify the second speech segment of the first speech stream to obtain a second recognition result.
[0110] In some possible implementations, the identification module 404 is specifically used for:
[0111] When the confidence level of the recommended scene parameters meets the first condition, the recommended scene parameters are automatically used to identify the second speech segment of the first speech stream to obtain a second recognition result.
[0112] In some possible implementations, the device 400 further includes:
[0113] The display module 408 is used to display the recommended scene parameters to the user when the confidence level of the recommended scene parameters meets the second condition;
[0114] The identification module 404 is specifically used for:
[0115] When a user triggers a parameter update operation, the recommended scenario parameters are used to identify the second speech segment of the first speech stream to obtain a second recognition result.
[0116] In some possible implementations, the scene parameters include one or more of the following: language, dialect, domain, near and far field information, or speaker information.
[0117] In some possible implementations, the voice includes conference audio, and the scene detection module 406 is specifically used for:
[0118] Scene detection is performed based on the second audio stream in the multi-channel audio stream and the conference information to obtain recommended scene parameters.
[0119] This application also provides a computing device, including: a processor, a memory, and a system bus;
[0120] The processor and memory are connected via the system bus;
[0121] The memory is used to store one or more programs, wherein the one or more programs include instructions that, when executed by the processor, cause the processor to perform any of the above-described implementations of the speech recognition method.
[0122] This application also provides a computer-readable storage medium storing instructions that, when executed on a computing device, cause the computing device to perform any of the above-described implementations of the speech recognition method.
[0123] This application also provides a computer program product that, when run on a computing device, causes the computing device to execute any of the above-described implementations of the speech recognition method.
[0124] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that all or part of the steps in the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network communication device such as a media gateway, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.
[0125] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.
[0126] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0127] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A speech recognition method, characterized in that, include: Receives a voice stream input by the user, which is divided into multiple streams; The first speech segment of the first speech stream in the multi-speech stream is identified using default scene parameters to obtain a first recognition result, and scene detection is performed based on the second speech stream in the multi-speech stream to obtain recommended scene parameters. The second speech segment of the first speech stream is identified using the recommended scene parameters to obtain the second recognition result; The step of performing scene detection based on the second speech stream in the multi-speech stream to obtain recommended scene parameters includes: Obtain the third speech segment corresponding to the time window from the second speech stream, perform scene detection based on the third speech segment, and obtain recommended scene parameters; or... When the speaker information changes, scene detection is performed based on the second voice stream in the multi-channel voice stream to obtain recommended scene parameters; The step of using recommended scene parameters to identify the second speech segment of the first speech stream to obtain a second recognition result includes: When the recommended scene parameters determined based on the third speech segments corresponding to multiple time windows in the second speech stream are the same, the recommended scene parameters are used to identify the second speech segment of the first speech stream to obtain a second recognition result.
2. The method according to claim 1, characterized in that, The step of using recommended scene parameters to identify the second speech segment of the first speech stream to obtain a second recognition result includes: When the confidence level of the recommended scene parameters meets the first condition, the recommended scene parameters are automatically used to identify the second speech segment of the first speech stream to obtain a second recognition result.
3. The method according to claim 1, characterized in that, The step of using recommended scene parameters to identify the second speech segment of the first speech stream to obtain a second recognition result includes: When the confidence level of the recommended scenario parameters meets the second condition, the recommended scenario parameters are displayed to the user. When a user triggers a parameter update operation, the recommended scenario parameters are used to identify the second speech segment of the first speech stream to obtain a second recognition result.
4. The method according to claim 1, characterized in that, The scene parameters include one or more of the following: language, dialect, domain, near and far field information, or speaker information.
5. The method according to claim 4, characterized in that, The audio includes conference audio, and the step of performing scene detection based on the second audio stream from the multiple audio streams to obtain recommended scene parameters includes: Scene detection is performed based on the second audio stream in the multi-channel audio stream and the conference information to obtain recommended scene parameters.
6. A voice recognition device, characterized in that, include: A voice acquisition module is used to receive a voice stream input by the user, which is divided into multiple channels; The recognition module is used to recognize the first speech segment of the first speech stream in the multi-speech stream using default scene parameters, and obtain the first recognition result; The scene detection module is used to perform scene detection based on the second voice stream in the multi-channel voice stream to obtain recommended scene parameters; The recognition module is further configured to use recommended scene parameters to recognize the second speech segment of the first speech stream and obtain a second recognition result; The scene detection module is specifically used for: Obtain the third speech segment corresponding to the time window from the second speech stream, perform scene detection based on the third speech segment, and obtain recommended scene parameters; or... When the speaker information changes, scene detection is performed based on the second voice stream in the multi-channel voice stream to obtain recommended scene parameters; The identification module is specifically used for: When the recommended scene parameters determined based on the third speech segments corresponding to multiple time windows in the second speech stream are the same, the recommended scene parameters are used to identify the second speech segment of the first speech stream to obtain a second recognition result.
7. A computing device, characterized in that, include: Processor, memory, system bus; The processor and the memory are connected via the system bus; The memory is used to store one or more programs, the one or more programs including instructions that, when executed by the processor, cause the processor to perform the method according to any one of claims 1-5.
8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a computing device, cause the computing device to perform the method according to any one of claims 1-5.
Citation Information
Patent Citations
Multi-scene voice recognition method and device and intelligent customer service system applying same
CN111049996A
Audio noise reduction method and device and computer readable storage medium
CN113823303A
Voice recognition method and device, computer readable storage medium and electronic equipment
CN113889091A
