A speech recognition method, speech recognition device, electronic device and storage medium

By using multiple voice collectors to collect voice data of multiple objects in speech recognition scenarios, the target voice data is generated to improve speech recognition performance, and the problems of large computing volume and low recognition performance in the prior art are solved, and more accurate and Efficient voice recognition.

CN113963694BActive Publication Date: 2024-11-15CHINA MOBILE (SUZHOU) SOFTWARE TECH CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202010700307.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-07-20
Publication Date
2024-11-15
Estimated Expiration
2040-07-20

AI Technical Summary

Technical Problem

In the existing speech recognition technology, the human analysis method has a large amount of calculation and inaccurate results. The multi-speech engine recognition method has a low recognition performance when the speech segment is too long.

Method used

By setting up multiple voice collectors in the speech recognition scene, voice data of multiple objects are collected, and target voice data associated with each object is generated based on these data, thereby generating voice recognition results, reducing artificial analysis and improving recognition performance.

Benefits of technology

There is no need to artificially analyze voice data, reducing the amount of calculations and improving the recognition performance and accuracy of results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113963694B_ABST
    Figure CN113963694B_ABST
Patent Text Reader

Abstract

The present application discloses a speech recognition method, a speech recognition device, an electronic device and a storage medium, the method comprising: acquiring a plurality of speech data in a current speech recognition scene; the plurality of speech data comprising speech data of a plurality of objects in the current speech recognition scene collected by a plurality of speech collectors; the plurality of speech collectors being located at different positions in the current speech recognition scene; generating target speech data associated with each of the plurality of objects based on the plurality of speech data; the target speech data associated with each object being derived from at least two of the plurality of speech data; generating speech recognition results based on the plurality of target speech data, and outputting the speech recognition results; in this way, there is no need to manually analyze the speech data, the amount of computation of the speech data is reduced, and the accuracy of the speech analysis results is ensured.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of speech processing technology, and in particular to a speech recognition method, a speech recognition device, an electronic device and a storage medium. Background Art

[0002] With the rapid development of speech recognition technology, speech recognition has become an important way of human-computer interaction. Common speech recognition methods in related technologies include manual analysis and multi-speech engine recognition.

[0003] When a manual analysis method is used, in some occasions such as meetings, after the voice information of the meeting participants is collected, the voice information corresponding to each person is manually analyzed. This method has the problem of large data analysis calculation or inaccurate analysis results. When a multi-speech engine recognition method is used, in some occasions such as meetings, after the voice information of the meeting participants is collected, the voice information is input into multiple speech recognition engines, and the confidence of the recognition result corresponding to each speech recognition engine is obtained. Finally, the recognition result with the highest confidence is determined as the final speech recognition result. This method has the problem of low recognition performance when the speech segment is too long. It can be seen that the manual analysis method in the related art has the problem of large data analysis calculation or inaccurate analysis results, and the multi-speech engine recognition method has the problem of low recognition performance. Summary of the invention

[0004] The present application aims to provide a speech recognition method, a speech recognition device, an electronic device and a storage medium, which solve the problems in the related art that the manual analysis method has a large amount of data analysis calculation or inaccurate analysis results, and the multi-speech engine recognition method has a low recognition performance.

[0005] The technical solution of this application is implemented as follows:

[0006] The present application provides a speech recognition method, the method comprising:

[0007] Acquire multiple voice data in a current voice recognition scene; the multiple voice data include voice data of multiple objects in the current voice recognition scene collected by multiple voice collectors; the multiple voice collectors are located at different positions in the current voice recognition scene;

[0008] Based on the plurality of voice data, generating target voice data associated with each of the plurality of objects; the target voice data associated with each object is derived from at least two of the plurality of voice data;

[0009] A speech recognition result is generated based on the plurality of target speech data, and the speech recognition result is output.

[0010] Optionally, the generating target voice data associated with each of the multiple objects based on the multiple voice data includes:

[0011] Segment each of the plurality of voice data to obtain each sub-voice data set after segmentation of each voice data; each sub-voice data set includes multiple segments of voice data;

[0012] Acquire a plurality of voiceprint features associated with each sub-voice data set;

[0013] Based on each sub-voice data set and a plurality of voiceprint features associated with each sub-voice data set, target voice data associated with each object is generated.

[0014] Optionally, the generating the target voice data associated with each object based on each sub-voice data set and a plurality of voiceprint features associated with each sub-voice data set includes:

[0015] Determining multiple sub-voice data having the same voiceprint feature and the same timestamp in the multiple sub-voice data sets;

[0016] Determine target sub-voice data from the multiple sub-voice data to obtain multiple target sub-voice data associated with the same voiceprint feature;

[0017] The target voice data associated with each object is generated based on the multiple target sub-voice data and the timestamps corresponding to the multiple target sub-voice data.

[0018] Optionally, the target sub-voice data is the voice data with the largest amplitude among the multiple sub-voice data.

[0019] Optionally, before generating a speech recognition result based on the target speech data, the method further includes:

[0020] Acquire first position information of the plurality of voice collectors in the current voice recognition scene;

[0021] Correspondingly, the generating of speech recognition results based on the plurality of target speech data further includes:

[0022] Determine second position information of each object of the plurality of objects in the current speech recognition scene based on the first position information and the plurality of target speech data;

[0023] A speech recognition result is generated based on the plurality of second position information and the plurality of target speech data.

[0024] Optionally, the generating a speech recognition result based on the plurality of second position information and the plurality of target speech data includes:

[0025] Performing speech emotion recognition on the plurality of target speech data to obtain a plurality of first recognition results;

[0026] Acquire a plurality of target text data associated with a plurality of the target voice data;

[0027] Performing semantic recognition on the plurality of target text data to obtain a plurality of second recognition results;

[0028] The speech recognition result is generated based on the plurality of the first recognition results, the plurality of the second recognition results, the plurality of the target text data and the plurality of the second position information.

[0029] Optionally, generating the speech recognition result based on the plurality of first recognition results, the plurality of second recognition results, the plurality of target text data and the plurality of second position information includes:

[0030] Extracting feature information of the target text data associated with each object based on the first recognition result associated with each object and the second recognition result associated with each object;

[0031] Based on the feature information associated with each object and the second position information associated with each object, a directional speech map is generated; and the speech recognition result includes the directional speech map.

[0032] The present application provides a speech recognition device, the speech recognition device comprising:

[0033] An acquisition unit, configured to acquire a plurality of speech data in a current speech recognition scene; the plurality of speech data comprises speech data of a plurality of objects in the current speech recognition scene acquired by a plurality of speech collectors; the plurality of speech collectors are located at different positions in the current speech recognition scene;

[0034] A first processing unit, configured to generate target voice data associated with each of the plurality of objects based on the plurality of voice data; the target voice data associated with each object is derived from at least two of the plurality of voice data;

[0035] The second processing unit is used to generate a speech recognition result based on the multiple target speech data and output the speech recognition result.

[0036] The present application provides an electronic device, the electronic device comprising:

[0037] A memory for storing executable instructions;

[0038] The processor is used to execute the executable instructions stored in the memory to implement the speech recognition method as described above.

[0039] The present application provides a computer storage medium, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the speech recognition method as described above.

[0040] The present application provides a speech recognition method, a speech recognition device, an electronic device and a storage medium, which obtain multiple speech data in a current speech recognition scenario; the multiple speech data include speech data of multiple objects in the current speech recognition scenario collected by multiple speech collectors; the multiple speech collectors are located at different positions in the current speech recognition scenario; based on the multiple speech data, target speech data associated with each of the multiple objects are generated; the target speech data associated with each object comes from at least two of the multiple speech data; speech recognition results are generated based on the multiple target speech data, and the speech recognition results are output; that is, the present application generates target speech data associated with each of the multiple objects based on the multiple speech data including multiple objects collected by multiple speech collectors at different positions in the current speech recognition scenario, and then generates and outputs speech recognition results based on the target speech data; in this way, there is no need to manually analyze the speech data, intelligent analysis of the speech data is achieved, the calculation amount of the speech data is reduced, the recognition performance is improved, and the accuracy of the speech analysis results is guaranteed. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A flowchart of a speech recognition method provided in an embodiment of the present application;

[0042] Figure 2 A flowchart of another speech recognition method provided in an embodiment of the present application;

[0043] Figure 3 (a) to (c) are schematic diagrams of a process for segmenting multiple voice data provided in an embodiment of the present application;

[0044] Figure 4 A flowchart of another speech recognition method provided in an embodiment of the present application;

[0045] Figure 5 A schematic diagram of the structure of a speech recognition device provided in an embodiment of the present application;

[0046] Figure 6 A schematic diagram of the structure of a speech recognition processing device provided in an embodiment of the present application;

[0047] Figure 7A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0048] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0049] It should be understood that the "embodiments of the present application" or "the aforementioned embodiments" mentioned throughout the specification mean that specific features, structures or characteristics related to the embodiments are included in at least one embodiment of the present application. Therefore, "in the embodiments of the present application" or "in the aforementioned embodiments" appearing throughout the specification do not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. In the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. The above-mentioned serial numbers of the embodiments of the present application are for description only and do not represent the advantages and disadvantages of the embodiments.

[0050] The present application embodiment provides a speech recognition method applied to an electronic device, referring to Figure 1 As shown, the method comprises the following steps:

[0051] Step 101: Acquire multiple voice data in the current voice recognition scenario.

[0052] The multiple voice data include voice data of multiple objects in the current voice recognition scene collected by multiple voice collectors; the multiple voice collectors are located at different positions in the current voice recognition scene.

[0053] In the embodiments of the present application, the current voice recognition scenario can be understood as a scenario in which voice data can be acquired; illustratively, the current voice recognition scenario can be a scenario in which real-time subtitles are displayed on the screen during an input method, meeting, and / or court trial; the current voice recognition scenario can also be a recorded audio / video subtitle configuration and / or a customer service voice quality inspection and user-generated content (UGC) voice content review scenario with low real-time requirements, which is not specifically limited in the present application.

[0054] In the embodiment of the present application, the electronic device can collect voice data of multiple objects in the current voice recognition scene through multiple voice collectors. It should be noted that the multiple voice collectors are located at different positions in the current voice recognition scene, and the same voice collector can collect voice data of multiple objects in the current voice recognition scene, and different voice collectors can collect voice data of the same object in the current voice recognition scene.

[0055] In practical applications, the position of the voice collector can be understood as the position set in advance by the user in the current voice recognition scenario to collect high-quality voice data. At the same time, the distance between every two voice collectors in the multiple voice collectors meets the preset distance. Among them, the preset distance can be the distance corresponding to the model of the voice collector before the electronic device obtains multiple voice data; the preset distance can also be the distance set in advance by the user before the electronic device obtains multiple voice data. Whether the preset distance is the distance corresponding to the model of the voice collector or the distance set in advance by the user, it is based on the voice collector collecting a better voice signal. Exemplarily, if the user sets four voice collectors of the same model, one voice collector can be set in each of the four directions, namely, the southeast, northwest, and the distance between every two voice collectors in the four voice collectors is two meters. Further, the position of each voice collector is marked and stored in the electronic device.

[0056] For example, four voice collectors of the same model are installed on the conference table of a certain conference room, namely A, B, C, and D, and the participants of the meeting are A, B, C, and D; wherein, voice collector A is located in the east direction of the conference table, voice collector B is located in the south direction of the conference table, voice collector C is located in the west direction of the conference table, and voice collector D is located in the north direction of the conference table, and the distance between each two voice collectors of voice collectors A, B, C, and D is two meters. When the meeting participants hold a meeting in the conference room, voice collectors A and voice collectors B can collect voice data of object A at the same time, and voice collector A can also collect voice data of object A and object B at the same time, and the voice data collected by each voice collector will be stored in the electronic device.

[0057] Step 102: Generate target voice data associated with each of the multiple objects based on the multiple voice data.

[0058] The target voice data associated with each object is derived from at least two voice data among the multiple voice data.

[0059] In the embodiment of the present application, the target voice data may be understood as voice data associated with the same object among multiple objects.

[0060] In an embodiment of the present application, after the electronic device obtains multiple voice data in the current voice recognition scenario, it obtains voice data associated with the same object among multiple objects based on at least two of the multiple voice data, and then generates target voice data associated with each of the multiple objects.

[0061] Step 103: Generate a speech recognition result based on the multiple target speech data, and output the speech recognition result.

[0062] In the embodiment of the present application, the speech recognition result can be understood as the result obtained by using the speech recognition method to process the target speech data.

[0063] In practical applications, after the electronic device generates target voice data associated with each of the multiple objects, it generates voice recognition results from the multiple target voice data through a voice recognition method, and outputs the voice recognition results. Among them, the voice recognition method includes methods based on linguistics and acoustics, random model methods, and probabilistic grammar analysis, etc.; the voice recognition method also includes a random model method. In the embodiment of the present application, the voice recognition method preferably uses a random model method to generate voice recognition results from the multiple target voice data.

[0064] A speech recognition method provided by the present application obtains multiple speech data in a current speech recognition scenario; the multiple speech data include speech data of multiple objects in the current speech recognition scenario collected by multiple speech collectors; the multiple speech collectors are located at different positions in the current speech recognition scenario; based on the multiple speech data, target speech data associated with each of the multiple objects is generated; the target speech data associated with each object comes from at least two of the multiple speech data; speech recognition results are generated based on the multiple target speech data, and the speech recognition results are output; that is, in an embodiment of the present application, based on the multiple speech data including multiple objects collected by multiple speech collectors at different positions in the current speech recognition scenario, target speech data associated with each of the multiple objects is generated, and then speech recognition results are generated and output based on the target speech data; in this way, there is no need to manually analyze the speech data, the amount of calculation of the speech data is reduced, and the accuracy of the speech analysis results is guaranteed.

[0065] Based on the above embodiments, the present application provides a speech recognition method, which is applied to electronic equipment, referring to Figure 2 As shown, the method comprises the following steps:

[0066] Step 201: Acquire multiple speech data in the current speech recognition scenario.

[0067] The multiple voice data include voice data of multiple objects in the current voice recognition scene collected by multiple voice collectors; the multiple voice collectors are located at different positions in the current voice recognition scene.

[0068] Step 202: Segment each voice data among the plurality of voice data to obtain each sub-voice data set after segmentation of each voice data.

[0069] Each sub-voice data set includes multiple segments of voice data.

[0070] In the embodiment of the present application, the electronic device segments each voice data among the multiple voice data to obtain each sub-voice data set including multiple segments of voice data after each voice data is segmented.

[0071] In practical applications, step 202 divides each voice data in the plurality of voice data to obtain each sub-voice data set after each voice data is divided, which can be implemented by the following steps:

[0072] Step S1, obtaining the interval time between sentences in each voice data among multiple voice data, if the interval time is greater than the preset interval time, segmenting each voice data among the multiple voice data to obtain the first sub-voice data set after each voice data segmentation.

[0073] The first sub-voice data set includes multiple segments of voice data.

[0074] In the embodiment of the present application, the first sub-speech data set can be understood as a speech data set obtained after each speech data is segmented.

[0075] In the embodiment of the present application, the preset break time may be a break time corresponding to the voice data segmentation mode and pre-stored in the electronic device before segmenting each voice data in the plurality of voice data; the preset break time may also be a break time set in real time by the user after the electronic device determines that it has entered the voice data segmentation state. It is understandable that whether it is a pre-stored break time or a real-time set break time, both are based on obtaining better voice data.

[0076] In an embodiment of the present application, after the electronic device obtains multiple voice data in the current voice recognition scenario, it obtains the interval time between sentences in each voice data in the multiple voice data, compares the interval time between sentences in each voice data with the preset interval time, and if the interval time between sentences is greater than the preset interval time, then each voice data in the multiple voice data is segmented at the interval between sentences to obtain a first sub-voice data set including multiple segments of voice data after each voice data is segmented.

[0077] In practical applications, such as Figure 3 As shown in (a) and (b), if the electronic device obtains multiple voice data in the current voice recognition scenario, such as Figure 3 As shown in (a), the interval time between sentences in each of the multiple voice data is obtained, and the interval time between sentences in each voice data is compared with the preset interval time. For example, the preset interval time can be set to 2 seconds. If the interval time between sentences exceeds 2 seconds, each of the multiple voice data is segmented at the interval between sentences, such as Figure 3As shown in (b), segmentation is performed at the solid line to obtain a first sub-speech data set including multiple segments of speech data after each speech data segmentation.

[0078] Step S2: segment each voice data in the first sub-voice data set after segmentation of each voice data at a preset time interval to obtain each sub-voice data set after segmentation of each voice data.

[0079] The first sub-speech data set includes each sub-speech data set.

[0080] In the embodiment of the present application, the preset time interval may be a time interval corresponding to the segmentation mode of the first sub-speech data set, which is pre-stored in the electronic device before the first sub-speech data set is segmented; the preset time interval may also be a time interval set in real time by the user after the electronic device determines that it has entered a state of segmenting the first sub-speech data set. It can be understood that whether it is a pre-stored time interval or a time interval set in real time, both are based on obtaining better voice data.

[0081] In an embodiment of the present application, after the electronic device obtains the first sub-voice data set, it re-divides each voice data in the first sub-voice data set after each voice data segmentation according to a preset time interval to obtain each sub-voice data set after each voice data segmentation.

[0082] In practical applications, such as Figure 3 As shown in (b) and (c), the electronic device obtains a first sub-speech data set, such as Figure 3 As shown in (b), each voice data segment in the first sub-voice data set after each voice data segment is segmented is segmented again according to a preset time interval. For example, the preset time interval can be set to 10-30ms, and each voice data segment in the first sub-voice data set is segmented into voice data of smaller time periods, such as Figure 3 The segmentation is performed at the dotted line shown in (c) to obtain each sub-speech data set after each speech data segmentation of the multiple speech data segments. It should be noted that each speech data segment in the first sub-speech data set is segmented again to obtain speech data of a smaller time period, so that the speech recognition result of each object can be obtained more accurately.

[0083] Step 203: Obtain multiple voiceprint features associated with each sub-speech data set.

[0084] In the embodiment of the present application, the voiceprint feature can be understood as a feature possessed by different objects, and the voiceprint can also be understood as a feature for identity recognition. The voiceprint feature has specificity and stability.

[0085] In the embodiment of the present application, the electronic device obtains multiple voiceprint features associated with each sub-voice data set, and determines the number of objects in the current voice recognition scene based on the multiple voiceprint features associated with each sub-voice data set. It should be noted that the same object has the same voiceprint feature, and the electronic device can determine the number of objects in the current voice recognition scene based on the number of voiceprint features.

[0086] Step 204: Generate target voice data associated with each object based on each sub-voice data set and a plurality of voiceprint features associated with each sub-voice data set.

[0087] In the embodiment of the present application, step 204 generates target voice data associated with each object based on each sub-voice data set and multiple voiceprint features associated with each sub-voice data set, which can be achieved by the following steps:

[0088] Step A1: Determine multiple sub-voice data sets having the same voiceprint feature and the same timestamp.

[0089] In the embodiment of the present application, the same voiceprint feature can be understood as the same object having the same voiceprint feature; multiple sub-voice data can be understood as voice data with the same voiceprint feature and the same timestamp selected from multiple sub-voice data sets; that is, the multiple sub-voice data include voice data obtained from the same object at the same time point or within the same time period.

[0090] In the embodiment of the present application, after obtaining multiple voiceprint features associated with each sub-voice data set, the electronic device determines multiple sub-voice data with the same voiceprint feature and the same timestamp in the multiple sub-voice data sets.

[0091] Step A2: determining target sub-voice data from the multiple sub-voice data, and obtaining multiple target sub-voice data associated with the same voiceprint feature.

[0092] The target sub-speech data is the speech data with the largest amplitude among the multiple sub-speech data.

[0093] In the embodiment of the present application, the amplitude is the maximum value of the difference between the sound pressure and the static pressure. The sound pressure is the pressure increment formed by the alternating changes of compression and rarefaction when the sound wave propagates in the air. In practical applications, the principle of the voice collector to collect voice data is to convert the pressure change wave in the air into the change of the electrical signal.

[0094] In the embodiment of the present application, the electronic device determines the target sub-voice data with the largest amplitude from the multiple sub-voice data, and obtains multiple target sub-voice data associated with the same voiceprint feature, which can be achieved in the following manner:

[0095] The electronic device obtains each amplitude of multiple sub-voice data with the same voiceprint feature and the same timestamp, and compares each amplitude of the multiple sub-voice data, obtains the sub-voice data with the maximum amplitude from the multiple sub-voice data as the target sub-voice data, and then obtains multiple target sub-voice data associated with the same voiceprint feature based on the target sub-voice data. It should be noted that if there are at least two sub-voice data with the same amplitude among the multiple sub-voice data, the sub-voice data of any amplitude is obtained from the at least two sub-voice data with the same amplitude as the target sub-voice data, and then multiple target sub-voice data associated with the same voiceprint feature are obtained based on the target sub-voice data.

[0096] Exemplarily, the electronic device obtains sub-voice data T1, sub-voice data T2 and sub-voice data T3 with the same voiceprint feature and the same timestamp, and obtains the amplitude of sub-voice data T1 as A1, the amplitude of sub-voice data T2 as A2 and the amplitude of sub-voice data T3 as A3, compares the amplitude A1 of sub-voice data T1, the amplitude A2 of sub-voice data T2 and the amplitude A3 of sub-voice data T3, and obtains the comparison results of the following two cases: Case 1: Amplitude A3>Amplitude A2 >Amplitude A1, it means that the amplitude A3 of the sub-voice data T3 is the maximum amplitude, and the sub-voice data T3 is taken as the target sub-voice data; Case 2: Amplitude A3 = Amplitude A2 > Amplitude A1, it means that the amplitude A3 of the sub-voice data T3 and the amplitude A2 of the sub-voice data T2 are both the maximum amplitudes, and the sub-voice data of any amplitude is obtained from the amplitude A3 of the sub-voice data T3 and the amplitude A2 of the sub-voice data T2 as the target sub-voice data, such as obtaining the sub-voice data T2 of amplitude A2 as the target sub-voice data. It should be noted that the situations listed in the embodiments of the present application are only for the convenience of users to understand the present solution, and do not fully represent the specific implementation methods to be protected by the present application.

[0097] Step A3: Generate target voice data associated with each object based on the multiple target sub-voice data and the timestamps corresponding to the multiple target sub-voice data.

[0098] In an embodiment of the present application, the electronic device obtains multiple target sub-voice data associated with the same voiceprint feature, and timestamps corresponding to the multiple target sub-voice data, and generates target voice data associated with each object.

[0099] Step 205: Generate a speech recognition result based on the multiple target speech data, and output the speech recognition result.

[0100] In the embodiment of the present application, step 205 generates a speech recognition result based on multiple target speech data and outputs the speech recognition result, which can be achieved by the following steps:

[0101] Step B1: Obtain first position information of multiple voice collectors in the current voice recognition scene.

[0102] In the embodiment of the present application, the first position information can be understood as the information of the position of the voice collector in the current voice recognition scene. Among them, the first position information can be the position information of multiple voice collectors pre-stored in the electronic device in the current voice recognition scene; the first position information can also be the position information of multiple voice collectors in the current voice recognition scene acquired by the electronic device in real time. It should be noted that whether the first position information is the position information pre-stored in the electronic device or the position information acquired in real time, the position information of multiple voice collectors shall prevail.

[0103] Step B2: Determine second position information of each of the multiple objects in the current speech recognition scene based on the first position information and the multiple target speech data.

[0104] The second position information may be understood as the first position information relative to the multiple voice collectors in the current voice recognition scenario.

[0105] In an embodiment of the present application, after the electronic device obtains the first position information of multiple voice collectors in the current voice recognition scene, it determines the second position information of each of the multiple objects relative to the first position information in the current voice recognition scene based on the first position information and multiple target voice data.

[0106] Step B3: perform speech emotion recognition on multiple target speech data to obtain multiple first recognition results.

[0107] In the embodiment of the present application, the first recognition result can be understood as the result obtained after performing speech emotion recognition on the target speech data.

[0108] In an embodiment of the present application, the electronic device extracts emotional speech features of each voice data from a plurality of target voice data, performs emotion recognition on each voice data based on the emotional speech features, and obtains a first recognition result.

[0109] In practical applications, before extracting emotional speech features from each speech data, each speech data needs to be preprocessed. The preprocessing of speech data includes pre-emphasis, short-time analysis, framing, windowing, and endpoint detection. Among them, pre-emphasis: the higher the frequency in the speech data, the smaller the corresponding component. Pre-emphasis can increase the spectrum of the high-frequency part and make the spectrum of the signal flat, which is convenient for spectrum analysis or vocal tract parameter analysis. Short-time analysis: speech data changes with time as a whole, which is a non-stationary process, and speech data has time-varying characteristics, but within a short time range such as 10-30ms, its characteristics remain relatively stable, that is, speech has short-term stability. Framing: In order to perform short-time analysis, the speech data is segmented according to a preset time interval such as 10-30ms, and each segment is called a frame; in order to make a smooth transition between frames and maintain continuity, an overlapping segmentation method can also be used. Windowing: Use a certain window function to w(n) multiply s(n) to form a windowed speech data s w (n) = s(n) * w(n), the window length in the window function, that is, the number of sample points, corresponds to one frame. Endpoint detection: Accurately find the starting point and ending point of a piece of speech data to ensure that valid speech data and useless noise signals are separated.

[0110] In the embodiment of the present application, the extraction of emotional speech features is performed by converting the speech waveform into a parameter representation at a relatively minimum data rate. Among them, the algorithm for extracting emotional speech features includes Mel Frequency Cepstrum Coefficient (MFCC), Linear Prediction Cepstrum Coefficient (LPCC), Linear Spectrum Frequency (LSF), Discrete Wavelet Transform (DWT), Perceptual Linear Predictive (PLP), etc. Of course, the algorithm for extracting emotional speech features also includes others such as Linear Predictive Coding (LPC), etc. This application does not specifically limit the speech feature extraction algorithm.

[0111] Speech emotion recognition is essentially the classification and pattern recognition of speech feature parameters of speech emotion.

[0112] In the embodiment of the present application, the electronic device extracts the emotional speech features in each of the multiple target speech data based on the LPC algorithm, and uses the emotion classification algorithm based on the training library to perform emotion classification on each of the multiple target speech data to obtain multiple first recognition results. Among them, the emotion classification algorithm includes artificial neural network (ANN, Artificial Neural Network), hidden Markov model (HMM, Hidden Markov Model), support vector machine (SVM, Support Vector Machine), etc. Of course, the emotion recognition algorithm also includes others such as decision tree (DT, DecisionTree), and the present application does not specifically limit the emotion classification algorithm.

[0113] Speech emotion datasets are an important basis for studying speech emotion recognition. Speech emotion datasets include Belfast English emotion database, Berlin Emo-DB emotion database, CASIA Chinese emotion database, ACCorpus series Chinese emotion database, and of course, speech emotion datasets also include others such as FAU AIBO children's German emotion database. This application does not specifically limit speech emotion datasets.

[0114] In an embodiment of the present application, the electronic device uses the LPC algorithm to extract the emotional speech features of each speech data in multiple target speech data, uses the HMM algorithm for speech emotion classification, and uses the CASIA Chinese emotion database for training library to obtain multiple first recognition results.

[0115] Step B4: Acquire multiple target text data associated with multiple target voice data.

[0116] In the embodiment of the present application, after acquiring multiple target voice data, the electronic device converts the multiple target voice data into multiple target text data.

[0117] Step B5: perform semantic recognition on multiple target text data to obtain multiple second recognition results.

[0118] In the embodiment of the present application, semantic recognition can be understood as automatic word segmentation of the target text data, further organizing the structure of the target text data and even understanding the meaning of the target text data; the second recognition result can be understood as the result obtained after semantic recognition of the target text data.

[0119] In the embodiment of the present application, the electronic device performs semantic recognition on multiple target text data to obtain multiple second recognition results. Among them, the semantic recognition method includes a word segmentation method based on string matching and a maximum matching method. The present application does not specifically limit the semantic recognition method.

[0120] Step B6: Generate a speech recognition result based on the multiple first recognition results, the multiple second recognition results, the multiple target text data and the multiple second position information.

[0121] In the embodiment of the present application, step B6 generates a speech recognition result based on multiple first recognition results, multiple second recognition results, multiple target text data and multiple second position information by the following steps:

[0122] Step C1: extracting feature information of target text data associated with each object based on the first recognition result associated with each object and the second recognition result associated with each object.

[0123] In the embodiments of the present application, feature information can be understood as information of target text data associated with each object in the current speech recognition scenario. For example, feature information can be information of a certain word with a high frequency of occurrence in the target text data associated with each object; feature information can also be information of high topic relevance in the current speech recognition scenario, which is not specifically limited in the present application.

[0124] In actual application scenarios, such as in a meeting, the electronic device calculates the frequency of occurrence of words contained in the target text data associated with each object, compares the frequency of occurrence of each word, and obtains words with higher frequency as feature information.

[0125] In the embodiment of the present application, the electronic device extracts feature information of target text data associated with each object based on the first recognition result associated with each object and the second recognition result associated with each object.

[0126] Step C2: Generate a directional speech map based on the feature information associated with each object and the second position information associated with each object; the speech recognition result includes the directional speech map.

[0127] In the embodiment of the present application, the directional speech graph includes multiple objects, feature information associated with each object, and second position information associated with each object.

[0128] In an embodiment of the present application, the electronic device extracts feature information of target text data associated with each object, and generates a speech recognition result including a directional speech graph based on the feature information associated with each object and the second position information associated with each object.

[0129] In actual application scenarios, such as in a meeting, the electronic device extracts the feature information of the target text data associated with each object, and integrates the feature information of the target text data associated with each of the multiple objects as the feature information of this meeting, that is, the theme of this meeting. And based on the second position information associated with each object, a directional voice map of this meeting is generated.

[0130] In other embodiments of the present application, the electronic device can output a direction display map and display a direction voice map through a display module.

[0131] It should be noted that, for the description of the same steps and the same contents in this embodiment as those in other embodiments, reference can be made to the description in other embodiments and will not be repeated here.

[0132] Based on the above embodiments, the present application provides a speech recognition method, which is applied to electronic equipment, referring to Figure 4 As shown, the method comprises the following steps:

[0133] Step 301: Obtain four speech data in the current speech recognition scenario.

[0134] The four voice data include voice data of multiple objects in the current voice recognition scene collected by four voice collectors; the four voice collectors are located at different positions in the current voice recognition scene.

[0135] In an embodiment of the present application, the electronic device includes four voice collectors, which are located in the southeast, northwest, and northeast directions of the current voice recognition scene, and every two of the four voice collectors are half a meter apart from each other. Each voice collector collects voice data of multiple objects in the four directions of the current voice recognition scene to obtain four voice data.

[0136] Step 302: Obtain the interval time between sentences in each of the four voice data. If the interval time is greater than the preset interval time, segment each of the four voice data to obtain the first sub-voice data set after segmentation of each voice data.

[0137] The first sub-voice data set includes multiple segments of voice data.

[0138] Step 303: segment each voice data in the first sub-voice data set after segmentation of each voice data according to a preset time interval, extract multiple voiceprint features in each sub-voice data, and obtain voice data of each object in the multiple objects.

[0139] In an embodiment of the present application, the electronic device uses GMM to cluster the extracted multiple voiceprint features, determines the number of objects in each sub-voice data based on the differences in the voiceprint features, and finally performs voice splicing to obtain the voice data of each of the multiple objects, that is, the voice data of multiple objects in the same voice data at the same time is divided into multiple sub-voice data of each object.

[0140] Step 304: determine multiple sub-voice data with the same voiceprint feature and the same timestamp from the multiple sub-voice data sets, obtain the voice data with the largest amplitude among the multiple sub-voice data, and determine it as the target sub-voice data.

[0141] Step 305: Based on the multiple sub-voice data, multiple target sub-voice data associated with the same voiceprint feature are obtained.

[0142] Step 306: Determine second position information of each of the multiple objects in the current speech recognition scene based on the first position information of the speech collector of the multiple target sub-speech data collected.

[0143] Step 307: Perform speech emotion recognition on the multiple sub-target speech data to obtain multiple first recognition results.

[0144] In an embodiment of the present application, the electronic device performs speech emotion recognition on multiple sub-target speech data. The preprocessing of the speech data includes pre-emphasis, framing, windowing, endpoint detection, and uses the LPC algorithm to extract speech features. The HMM algorithm is used to perform emotion classification on the multiple sub-target speech data. In addition, the training library uses the CASIA Chinese emotion database, and ultimately obtains multiple first recognition results.

[0145] Step 308: Acquire multiple target text data associated with the multiple target voice data, perform semantic recognition on the multiple target text data, and obtain multiple second recognition results.

[0146] Step 309: extract keyword information of target text data associated with each object based on the first recognition result associated with each object and the second recognition result associated with each object.

[0147] Step 310: Generate a directional speech graph based on the keyword information associated with each object and the second position information associated with each object.

[0148] Among them, the speech recognition result includes a directional speech map.

[0149] Step 311: Output and export the directional speech graph.

[0150] From the above, it can be seen that in the embodiment of the present application, the position information of each object among multiple objects, the target voice data of each object, and the keywords extracted based on the target voice data of each object are matched, without the need for manual analysis, which reduces the computational complexity of voice data analysis and ensures the accuracy of voice results.

[0151] It should be noted that, for the description of the same steps and the same contents in this embodiment as those in other embodiments, reference can be made to the description in other embodiments and will not be repeated here.

[0152] Based on the above embodiments, the present invention provides a speech recognition device, which can be applied to Figures 1-2 In a speech recognition method provided in a corresponding embodiment, referring to Figure 5 As shown, the speech recognition device 5 comprises:

[0153] The acquisition unit 51 is used to acquire multiple voice data in the current voice recognition scene; the multiple voice data include voice data of multiple objects in the current voice recognition scene collected by multiple voice collectors; the multiple voice collectors are located at different positions in the current voice recognition scene;

[0154] A first processing unit 52 is used to generate target voice data associated with each of the multiple objects based on the multiple voice data; the target voice data associated with each object is derived from at least two of the multiple voice data;

[0155] The second processing unit 53 is used to generate a speech recognition result based on the multiple target speech data and output the speech recognition result.

[0156] In other embodiments of the present application, the first processing unit 52 is also used to segment each voice data among the multiple voice data to obtain each sub-voice data set after each voice data segmentation; each sub-voice data set includes multiple segments of voice data; obtain multiple voiceprint features associated with each sub-voice data set; based on each sub-voice data set and the multiple voiceprint features associated with each sub-voice data set, generate target voice data associated with each object.

[0157] In other embodiments of the present application, the first processing unit 52 is also used to determine multiple sub-voice data with the same voiceprint feature and the same timestamp in multiple sub-voice data sets; determine the target sub-voice data from the multiple sub-voice data to obtain multiple target sub-voice data associated with the same voiceprint feature; and generate target voice data associated with each object based on the multiple target sub-voice data and the timestamps corresponding to the multiple target sub-voice data.

[0158] In other embodiments of the present application, the first processing unit 52 is further configured to select the target sub-voice data as the voice data having the largest amplitude among the multiple sub-voice data.

[0159] In other embodiments of the present application, the second processing unit 53 is also used to obtain first position information of multiple voice collectors in the current voice recognition scene; based on the first position information and multiple target voice data, determine the second position information of each of the multiple objects in the current voice recognition scene; based on the multiple second position information and multiple target voice data, generate voice recognition results.

[0160] In other embodiments of the present application, the second processing unit 53 is also used to perform speech emotion recognition on multiple target speech data to obtain multiple first recognition results; obtain multiple target text data associated with the multiple target speech data; perform semantic recognition on the multiple target text data to obtain multiple second recognition results; and generate speech recognition results based on the multiple first recognition results, the multiple second recognition results, the multiple target text data and the multiple second position information.

[0161] In other embodiments of the present application, the second processing unit 53 is also used to extract feature information of target text data associated with each object based on the first recognition result associated with each object and the second recognition result associated with each object; generate a directional speech map based on the feature information associated with each object and the second position information associated with each object; and the speech recognition result includes a directional speech map.

[0162] Based on the above embodiments, the present application provides another speech recognition device, which can be applied to Figure 3 In a speech recognition method provided in a corresponding embodiment, referring to Figure 6 As shown, Figure 6 The speech recognition processing device 6 corresponds to Figure 5 The speech recognition device 5 in the speech recognition device 5, wherein the acquisition unit 51 in the speech recognition device 5 includes the speech collection unit 61 in the speech recognition processing device 6, the first processing unit 52 in the speech recognition device 5 includes the speech data segmentation unit 62, the human voice separation unit 63 and the voiceprint recognition engine unit 64 in the speech recognition processing device 6; the second processing unit 53 in the speech recognition device 5 includes the speech data selection and object position determination unit 65, the speech emotion recognition unit 66, the speech recognition unit 67, the text conversion unit 68 and the generation output unit 69 in the speech recognition processing device 6.

[0163] Based on the above embodiments, the present application provides an electronic device which can be applied to Figures 1-2 In a speech recognition method provided in a corresponding embodiment, referring to Figure 7 As shown, the electronic device 7 ( Figure 7 Electronic devices 7 in the Figure 5 The speech recognition device 5) includes: a memory 71 and a processor 72; wherein the processor 72 is used to execute the speech recognition program stored in the memory 71, and the electronic device 7 implements the following steps through the processor 72:

[0164] Acquire multiple voice data in the current voice recognition scene; the multiple voice data include voice data of multiple objects in the current voice recognition scene collected by multiple voice collectors; the multiple voice collectors are located at different positions in the current voice recognition scene;

[0165] Based on the plurality of speech data, generating target speech data associated with each of the plurality of objects; the target speech data associated with each object is derived from at least two of the plurality of speech data;

[0166] A speech recognition result is generated based on a plurality of target speech data, and the speech recognition result is output.

[0167] In other embodiments of the present application, the processor 72 is used to execute the speech recognition program stored in the memory 71 to implement the following steps:

[0168] Segment each voice data in the plurality of voice data to obtain each sub-voice data set after segmentation of each voice data; each sub-voice data set includes multiple segments of voice data;

[0169] Obtaining multiple voiceprint features associated with each sub-speech data set;

[0170] Based on each sub-voice data set and a plurality of voiceprint features associated with each sub-voice data set, target voice data associated with each object is generated.

[0171] In other embodiments of the present application, the processor 72 is used to execute the speech recognition program stored in the memory 71 to implement the following steps:

[0172] Determining multiple sub-voice data having the same voiceprint feature and the same timestamp in multiple sub-voice data sets;

[0173] Determine target sub-voice data from the multiple sub-voice data to obtain multiple target sub-voice data associated with the same voiceprint feature;

[0174] Based on the multiple target sub-voice data and the timestamps corresponding to the multiple target sub-voice data, the target voice data associated with each object is generated.

[0175] In other embodiments of the present application, the processor 72 is used to execute the speech recognition program stored in the memory 71 to implement the following steps:

[0176] The target sub-voice data is the voice data having the largest amplitude among the plurality of sub-voice data.

[0177] In other embodiments of the present application, the processor 72 is used to execute the speech recognition program stored in the memory 71 to implement the following steps:

[0178] Obtaining first position information of multiple voice collectors in the current voice recognition scene;

[0179] Accordingly, generating a speech recognition result based on the plurality of target speech data further includes:

[0180] Determining second position information of each of the plurality of objects in the current speech recognition scene based on the first position information and the plurality of target speech data;

[0181] Based on the plurality of second position information and the plurality of target voice data, a voice recognition result is generated.

[0182] In other embodiments of the present application, the processor 72 is used to execute the speech recognition program stored in the memory 71 to implement the following steps:

[0183] Performing speech emotion recognition on a plurality of target speech data to obtain a plurality of first recognition results;

[0184] Acquire multiple target text data associated with multiple target voice data;

[0185] Performing semantic recognition on multiple target text data to obtain multiple second recognition results;

[0186] A speech recognition result is generated based on the plurality of first recognition results, the plurality of second recognition results, the plurality of target text data and the plurality of second position information.

[0187] In other embodiments of the present application, the processor 72 is used to execute the speech recognition program stored in the memory 71 to implement the following steps:

[0188] Extracting feature information of target text data associated with each object based on the first recognition result associated with each object and the second recognition result associated with each object;

[0189] Based on the feature information associated with each object and the second position information associated with each object, a directional speech map is generated; and the speech recognition result includes the directional speech map.

[0190] Based on the foregoing embodiments, an embodiment of the present invention provides a computer storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement the following steps:

[0191] Acquire multiple voice data in the current voice recognition scene; the multiple voice data include voice data of multiple objects in the current voice recognition scene collected by multiple voice collectors; the multiple voice collectors are located at different positions in the current voice recognition scene;

[0192] Based on the plurality of speech data, generating target speech data associated with each of the plurality of objects; the target speech data associated with each object is derived from at least two of the plurality of speech data;

[0193] A speech recognition result is generated based on a plurality of target speech data, and the speech recognition result is output.

[0194] In other embodiments of the present invention, the one or more programs may be executed by one or more processors to implement the following steps:

[0195] Segment each voice data in the plurality of voice data to obtain each sub-voice data set after segmentation of each voice data; each sub-voice data set includes multiple segments of voice data;

[0196] Obtaining multiple voiceprint features associated with each sub-speech data set;

[0197] Based on each sub-voice data set and a plurality of voiceprint features associated with each sub-voice data set, target voice data associated with each object is generated.

[0198] In other embodiments of the present invention, the one or more programs may be executed by one or more processors to implement the following steps:

[0199] Determining multiple sub-voice data having the same voiceprint feature and the same timestamp in multiple sub-voice data sets;

[0200] Determine target sub-voice data from the multiple sub-voice data to obtain multiple target sub-voice data associated with the same voiceprint feature;

[0201] Based on the multiple target sub-voice data and the timestamps corresponding to the multiple target sub-voice data, the target voice data associated with each object is generated.

[0202] In other embodiments of the present invention, the one or more programs may be executed by one or more processors to implement the following steps:

[0203] The target sub-voice data is the voice data having the largest amplitude among the plurality of sub-voice data.

[0204] In other embodiments of the present invention, the one or more programs may be executed by one or more processors to implement the following steps:

[0205] Obtaining first position information of multiple voice collectors in the current voice recognition scene;

[0206] Accordingly, generating a speech recognition result based on the plurality of target speech data further includes:

[0207] Determining second position information of each of the plurality of objects in the current speech recognition scene based on the first position information and the plurality of target speech data;

[0208] Based on the plurality of second position information and the plurality of target voice data, a voice recognition result is generated.

[0209] In other embodiments of the present invention, the one or more programs may be executed by one or more processors to implement the following steps:

[0210] Performing speech emotion recognition on a plurality of target speech data to obtain a plurality of first recognition results;

[0211] Acquire multiple target text data associated with multiple target voice data;

[0212] Performing semantic recognition on multiple target text data to obtain multiple second recognition results;

[0213] A speech recognition result is generated based on the plurality of first recognition results, the plurality of second recognition results, the plurality of target text data and the plurality of second position information.

[0214] In other embodiments of the present invention, the one or more programs may be executed by one or more processors to implement the following steps:

[0215] Extracting feature information of target text data associated with each object based on the first recognition result associated with each object and the second recognition result associated with each object;

[0216] Based on the feature information associated with each object and the second position information associated with each object, a directional speech map is generated; and the speech recognition result includes the directional speech map.

[0217] It should be noted that, for the description of the same steps and the same contents in this embodiment as those in other embodiments, reference can be made to the description in other embodiments and will not be repeated here.

[0218] It should be noted that the above-mentioned computer storage medium / memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory (Flash Memory), a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM) and other memories; it can also be various terminals including one or any combination of the above-mentioned memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc.

[0219] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0220] The units described above 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 distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0221] In addition, all functional units in the embodiments of the present application can be integrated into one processing module, or each unit can be a separate unit, or two or more units can be integrated into one unit; the above integrated unit can be implemented in the form of hardware or in the form of hardware plus software functional units. A person of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium, which, when executed, executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, read-only memory (ROM), random access memory (RAM), disks or optical disks, and other media that can store program codes.

[0222] The methods disclosed in several method embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments.

[0223] The features disclosed in several product embodiments provided in this application can be arbitrarily combined without conflict to obtain new product embodiments.

[0224] The features disclosed in several method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments or device embodiments.

[0225] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A speech recognition method, characterized in that: The method comprises: Acquire multiple voice data in a current voice recognition scene; the multiple voice data include voice data of multiple objects in the current voice recognition scene collected by multiple voice collectors; the multiple voice collectors are located at different positions in the current voice recognition scene; Based on the plurality of voice data, generating target voice data associated with each of the plurality of objects; the target voice data associated with each object is derived from at least two of the plurality of voice data; Generating a speech recognition result based on the plurality of target speech data, and outputting the speech recognition result; Before generating a speech recognition result based on the target speech data, the method further includes: Acquire first position information of the plurality of voice collectors in the current voice recognition scene; Correspondingly, the generating of speech recognition results based on the plurality of target speech data further includes: Determine second position information of each object of the plurality of objects in the current speech recognition scene based on the first position information and the plurality of target speech data; generating a speech recognition result based on the plurality of second position information and the plurality of target speech data; The generating a speech recognition result based on the plurality of second position information and the plurality of target speech data comprises: Performing speech emotion recognition on the plurality of target speech data to obtain a plurality of first recognition results; Acquire a plurality of target text data associated with a plurality of the target voice data; Performing semantic recognition on the plurality of target text data to obtain a plurality of second recognition results; The speech recognition result is generated based on the plurality of the first recognition results, the plurality of the second recognition results, the plurality of the target text data and the plurality of the second position information.

2. The method according to claim 1, characterized in that: The step of generating target voice data associated with each of the plurality of objects based on the plurality of voice data comprises: Segment each of the plurality of voice data to obtain each sub-voice data set after segmentation of each voice data; each sub-voice data set includes multiple segments of voice data; Acquire a plurality of voiceprint features associated with each sub-voice data set; Based on each sub-voice data set and a plurality of voiceprint features associated with each sub-voice data set, target voice data associated with each object is generated.

3. The method according to claim 2, characterized in that The generating the target voice data associated with each object based on each sub-voice data set and a plurality of voiceprint features associated with each sub-voice data set comprises: Determining multiple sub-voice data having the same voiceprint feature and the same timestamp in the multiple sub-voice data sets; Determine target sub-voice data from the multiple sub-voice data to obtain multiple target sub-voice data associated with the same voiceprint feature; The target voice data associated with each object is generated based on the multiple target sub-voice data and the timestamps corresponding to the multiple target sub-voice data.

4. The method according to claim 3, characterized in that The target sub-voice data is the voice data with the largest amplitude among the plurality of sub-voice data.

5. The method according to claim 1, characterized in that The generating the speech recognition result based on the plurality of the first recognition results, the plurality of the second recognition results, the plurality of the target text data and the plurality of the second position information comprises: Extracting feature information of the target text data associated with each object based on the first recognition result associated with each object and the second recognition result associated with each object; Based on the feature information associated with each object and the second position information associated with each object, a directional speech map is generated; and the speech recognition result includes the directional speech map.

6. A speech recognition device, characterized in that: The device comprises: An acquisition unit, configured to acquire a plurality of speech data in a current speech recognition scene; the plurality of speech data comprises speech data of a plurality of objects in the current speech recognition scene acquired by a plurality of speech collectors; the plurality of speech collectors are located at different positions in the current speech recognition scene; A first processing unit, configured to generate target voice data associated with each of the plurality of objects based on the plurality of voice data; the target voice data associated with each object is derived from at least two of the plurality of voice data; A second processing unit, configured to generate a speech recognition result based on the plurality of target speech data, and output the speech recognition result; The second processing unit is further used to obtain the first position information of the multiple voice collectors in the current voice recognition scene; based on the first position information and the multiple target voice data, determine the second position information of each of the multiple objects in the current voice recognition scene; based on the multiple second position information and the multiple target voice data, generate a voice recognition result; The second processing unit is also used to perform speech emotion recognition on the multiple target speech data to obtain multiple first recognition results; obtain multiple target text data associated with the multiple target speech data; perform semantic recognition on the multiple target text data to obtain multiple second recognition results; and generate the speech recognition results based on the multiple first recognition results, the multiple second recognition results, the multiple target text data and the multiple second position information.

7. An electronic device, characterized in that: The electronic device comprises: A memory for storing executable instructions; A processor, configured to execute the executable instructions stored in the memory to implement the speech recognition method according to any one of claims 1 to 5.

8. A computer storage medium, characterized in that: The computer storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the speech recognition method according to any one of claims 1 to 5.