Voice training data acquisition, model training method, device and electronic equipment

Abstract

The embodiment of the present invention discloses a method, device and electronic device for acquiring speech training data and training a model, including: inputting speech data in a speech data storage device to a target device in the form of an electrical signal; processing the speech data through a channel; and acquiring the through-channel speech data after the through-channel processing. The embodiment of the present invention can reduce the acquisition cost of speech training data, improve the quality and versatility of speech training data, and thus improve the training efficiency and recognition effect of a speech recognition model.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of speech data processing technology, and in particular to a speech training data acquisition, model training method, device, electronic device and storage medium. Background Art

[0002] In the process of speech recognition training, a large amount of audio training data is required. Usually, training a highly robust speech recognition model often requires thousands or even tens of thousands of hours of audio training data, and the cost of audio training data is relatively high.

[0003] When the target device to which the speech recognition model is applied changes, the hardware circuit for collecting audio data in the target device may also change accordingly, resulting in changes in the characteristics of the audio signal ultimately sent to the speech recognizer of the target device, such as changes in the spectrum and accuracy, which in turn leads to a decrease in the recognition rate. Therefore, the current process of collecting speech recognition training data generally requires collecting a large amount of speech training data through the target device, also known as "cross-channeling", to ensure that the data set used for speech recognition model training is consistent with the hardware processing process of the target device, thereby ensuring the robustness and accuracy of the speech recognition model, and that the performance of the trained speech recognition model on the target device is optimal.

[0004] At present, there are two main methods for existing voice training data transmission channels. One is to directly collect the original voice training data under different working conditions through the target device. This method will cost a lot of manpower and time costs, and each collection requires an independent recording space and a dedicated target device, which is costly and inefficient, and is easily affected by the stability of the target device, resulting in quality problems in the acquired voice training data; the other is to record and play the voice training data under different working conditions through the target device. This method is to collect the voice training data in advance, then play the voice training data, and transmit the audio signal through the air. The target device records and plays the voice training data under different working conditions; this method has high requirements for the playback device and the playback environment. Each time recording is performed, it also requires an independent recording space and a dedicated target device, and it is easy to introduce external environmental interference during recording, and the quality of the voice training data cannot be guaranteed, and the cost is also high. Summary of the invention

[0005] The embodiments of the present invention provide a method, device, electronic device and storage medium for acquiring speech training data and training a model, so as to reduce the cost of collecting speech training data, improve the quality and versatility of speech training data, and thus improve the training efficiency and recognition effect of the speech recognition model.

[0006] In a first aspect, an embodiment of the present invention provides a method for acquiring speech training data, comprising:

[0007] Inputting the voice data in the voice data storage device to the target device in the form of electrical signal transmission;

[0008] Processing the voice data through a channel;

[0009] Obtain the cross-channel voice data after cross-channel processing.

[0010] In a second aspect, an embodiment of the present invention further provides a model training method, comprising:

[0011] Acquire speech training data; wherein the speech training data is cross-channel speech data acquired by the speech training data acquisition method provided by any embodiment of the present invention;

[0012] A speech recognition model is trained according to the speech training data.

[0013] In a third aspect, an embodiment of the present invention further provides a speech training data acquisition device, comprising:

[0014] A voice data acquisition module, used to input the voice data in the voice data storage device to the target device in the form of electrical signal transmission;

[0015] A cross-channel processing module, used for processing the voice data across channels;

[0016] The cross-channel voice data acquisition module is used to acquire the cross-channel voice data after cross-channel processing.

[0017] In a fourth aspect, an embodiment of the present invention further provides a model training device, comprising:

[0018] A speech training data acquisition module, used to acquire speech training data; wherein the speech training data is cross-channel speech data acquired by the speech training data acquisition method provided by any embodiment of the present invention;

[0019] The speech recognition model training module is used to train the speech recognition model according to the speech training data.

[0020] In a fifth aspect, an embodiment of the present invention further provides an electronic device, the electronic device comprising:

[0021] one or more processors;

[0022] A storage device for storing one or more programs;

[0023] When the one or more programs are executed by the one or more processors, the one or more processors implement the method for acquiring speech training data provided by any embodiment of the present invention.

[0024] In a sixth aspect, an embodiment of the present invention further provides an electronic device, the electronic device comprising:

[0025] one or more processors;

[0026] A storage device for storing one or more programs;

[0027] When the one or more programs are executed by the one or more processors, the one or more processors implement the model training method provided by any embodiment of the present invention.

[0028] In a seventh aspect, an embodiment of the present invention further provides a computer storage medium on which a computer program is stored, and when the program is executed by a processor, the method for acquiring speech training data provided by any embodiment of the present invention is implemented.

[0029] In an eighth aspect, an embodiment of the present invention further provides a computer storage medium on which a computer program is stored, and when the program is executed by a processor, the model training method provided by any embodiment of the present invention is implemented.

[0030] The embodiment of the present invention inputs the voice data in the voice data storage device to the target device in the form of electrical signal propagation, processes the voice data through a channel, obtains the voice training data after the channel processing, and trains the voice recognition model according to the voice training data, thereby solving the problems existing in the existing voice training data acquisition method, such as the inability to guarantee the quality of the voice training data and the high labor and time costs, so as to reduce the cost of collecting voice training data, improve the quality and versatility of voice training data, and further improve the training efficiency and recognition effect of the voice recognition model. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1a is a flow chart of a method for acquiring speech training data provided by Embodiment 1 of the present invention;

[0032] Figure 1b It is a schematic diagram of the channel processing of speech training data in the prior art;

[0033] Figure 1c is a schematic diagram of a method for acquiring speech training data provided in Embodiment 1 of the present invention;

[0034] Figure 1d is a schematic diagram of a method for acquiring speech training data and noise training data provided in Embodiment 1 of the present invention;

[0035] Figure 2 is a flow chart of a model training method provided in Embodiment 2 of the present invention;

[0036] Figure 3is a schematic diagram of a speech training data acquisition device provided in Embodiment 3 of the present invention;

[0037] Figure 4 is a schematic diagram of a model training device provided by Embodiment 4 of the present invention;

[0038] Figure 5 This is a schematic diagram of the structure of an electronic device provided in Embodiment 5 of the present invention. DETAILED DESCRIPTION

[0039] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention.

[0040] It should also be noted that, for ease of description, only the part relevant to the present invention but not all content is shown in the accompanying drawings. It should be mentioned before discussing exemplary embodiments in more detail that some exemplary embodiments are described as processing or methods depicted as flow charts. Although the flow chart describes each operation (or step) as sequential processing, many operations therein can be implemented in parallel, concurrently or simultaneously. In addition, the order of each operation can be rearranged. When its operation is completed, the processing can be terminated, but it can also have additional steps not included in the accompanying drawings. The processing can correspond to methods, functions, procedures, subroutines, subprograms, etc.

[0041] The terms "first" and "second" and the like in the description, claims and drawings of the embodiments of the present invention are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but may include steps or units that are not listed.

[0042] Embodiment 1

[0043] Figure 1a is a flow chart of a method for acquiring speech training data provided by the first embodiment of the present invention. This embodiment is applicable to the case of acquiring high-quality general speech training data. The method can be executed by a speech training data acquisition device, which can be implemented by software and / or hardware and can generally be integrated in an electronic device. The electronic device can be a device that applies a speech recognition model. Figure 1a As shown, the method includes the following operations:

[0044] S110, inputting the voice data in the voice data storage device to the target device in the form of electrical signal transmission.

[0045] Among them, the voice data storage device can be a hardware device for acquiring and storing voice data. Exemplarily, the voice data storage device can be a collection device with data storage capabilities, which is used to collect and store voice data. For example, the voice data storage device can be a professional recording device, etc. Alternatively, the voice data storage device can also obtain voice data in other ways, for example: the voice data storage device can download voice data from the Internet or import voice data from an existing storage device. The embodiment of the present invention does not limit the specific device type of the voice data storage device and the method of acquiring voice data. The voice data can be original unprocessed audio data, or it can be audio data obtained after processing such as noise addition on the basis of the original voice data. The embodiment of the present invention also limits the type of different voice data. The target device can be a device that applies a voice recognition model. Exemplarily, the target device can be a vehicle device or an audio device, etc., and the number of target devices can be one or more.

[0046] In the embodiment of the present invention, before inputting voice data to the target device, the voice data can be first obtained by using the voice data storage device. It can be understood that the amount of voice data is usually large, so the voice data storage device can be used to obtain and save the voice data in real time. Accordingly, after obtaining all the voice data in the voice data storage device, all the voice data can be input to the target device in the form of electrical signal transmission.

[0047] It can be seen that inputting voice data to the target device in the form of electrical signal transmission can prevent voice data from being input to the target device through air transmission. For example, voice data can be prevented from being input to the microphone module built into the target device through air transmission. The advantage of this treatment is that it can effectively prevent interference or distortion, such as environmental noise, reverberation or attenuation, from being introduced during the process of voice data being input to the target device, so as to prevent voice data from being destroyed by various interference factors, thereby highly ensuring the quality of voice data.

[0048] In an optional embodiment of the present invention, the step of inputting the voice data in the voice data storage device to the target device in the form of an electrical signal may include: acquiring wake-up audio data and / or voice recognition sample audio data through a voice acquisition device; converting the wake-up audio data and / or voice recognition sample audio data into a voice analog signal through a voice analog signal conversion device; and inputting the voice analog signal to the target device in the form of an electrical signal.

[0049] Among them, the voice acquisition device can be a kind of voice data storage device, which can be used to acquire and store voice data by acquisition. Optionally, the voice data can be pure human voice audio data, which can include but is not limited to wake-up audio data and / or voice recognition sample audio data. Among them, the wake-up audio data can be audio data used to wake up the voice recognition model. For example, the wake-up audio data can be a wake-up sentence such as "Hello, Xiaobai". The voice recognition sample audio data can be a sentence or phrase used to identify the user's intention. For example, the voice recognition sample audio data can be an interactive sentence such as "Please pause the current music". The voice analog signal conversion device can be used to convert the input voice data into an analog signal, and then input the converted voice analog signal into the target device. Among them, the voice analog signal is audio data in the form of an analog signal. Optionally, the voice analog signal conversion device can be other hardware devices different from the target device, such as a computer and other hardware devices that can output audio data.

[0050] In an embodiment of the present invention, after the wake-up audio data and / or the voice recognition sample audio data are acquired through the voice acquisition device, the wake-up audio data and / or the voice recognition sample audio data can be input into the voice analog signal conversion device, so that the wake-up audio data and / or the voice recognition sample audio data can be converted into a voice analog signal through the voice analog signal conversion device. Then, the voice analog signal output by the voice analog signal conversion device can be input into the target device in the form of an electrical signal.

[0051] In a specific example, after the recording device collects and stores voice data as a voice data storage device, the voice data is stored in a computer device. The computer device can input voice data to the target device by playing voice data. It should be noted that the playback in the embodiment of the present invention does not mean that the voice data is played into the air, that is, in the process of the computer device inputting voice data to the target device by playing voice data, the voice data will not be transmitted into the air, and other devices cannot collect the voice data played by the computer device from the air. That is, the computer device inputs voice data to the target device by playing voice data, and the voice analog signal output by the computer device can be input into the target device through a direct signal transmission method.

[0052] In an optional embodiment of the present invention, the step of inputting the voice data in the voice data storage device to the target device in the form of an electrical signal may include: acquiring wake-up audio data and / or voice recognition sample audio data through a voice acquisition device; acquiring noise sample audio data through a noise acquisition device; converting the wake-up audio data and / or voice recognition sample audio data into a voice analog signal through a voice analog signal conversion device; converting the noise sample audio data into a noise analog signal through a noise analog signal conversion device; and simultaneously inputting the voice analog signal and the noise analog signal into the target device respectively in the form of an electrical signal.

[0053] Among them, the noise collection device can be used to collect noise sample audio data. The noise sample audio data can be the noise data of the target device under different working conditions. Similarly, the noise simulation signal conversion device can be used to convert the noise sample audio data into a noise simulation signal. Exemplarily, for audio equipment, the noise sample audio data can be echo data or human voice data, etc. For a vehicle, the noise sample audio data can be any one of the interior noise of a vehicle in a parking lot, the interior noise of a vehicle driving on a city road, and the interior noise of a vehicle driving on a freeway; and the noise sample audio data can include any one or more of tire noise, wind noise, engine noise, noisy human voices, and motor vehicle noise.

[0054] In the embodiment of the present invention, the voice data may only include pure human voice data. At the same time, in order to highly simulate the physical device data in the real environment, the voice data may also include pure human voice audio data and noise sample audio data, that is, the voice data may be noisy human voice audio data, which is used to train the voice recognition model.

[0055] Correspondingly, if the training data used to train the speech recognition model includes both pure human voice audio data and noise sample audio data, that is, noisy human voice audio data, it is necessary to use multiple analog signal conversion devices to convert the pure human voice audio data and the noise sample audio data into corresponding analog signals respectively, and at the same time input the speech analog signal and the noise analog signal into the target device in the form of electrical signal transmission.

[0056] In an optional embodiment of the present invention, the voice analog signal and the noise analog signal are simultaneously input to the target device in the form of electrical signals, respectively, and may be input to the same signal input point of the target device.

[0057] The same signal input point of the target device may be an audio signal input point in the target device, such as an input pin of a microphone.

[0058] Specifically, after obtaining the pure human voice audio data and the noise sample audio data, two analog signals, namely the voice analog signal and the noise analog signal, can be obtained through the voice analog signal conversion device and the noise analog signal conversion device respectively. Then the two analog signals are simultaneously input to the same signal input point of the target device, which is equivalent to directly inputting one channel of noisy human voice audio data to the target device.

[0059] It should be noted that, since the types of noise data under different working conditions are different, pure human voice audio data and matched noise sample audio data can be simultaneously input to the target device according to different working condition types to simulate noisy human voice audio data of the target device under different working conditions.

[0060] S120: Process the voice data across channels.

[0061] Among them, cross-channel processing is the process of collecting voice data through the target device.

[0062] Correspondingly, after the target device obtains the voice data in the voice data storage device, it can perform cross-channel processing on the voice data so that the voice data can perform cross-channel operations on the target device, thereby ensuring that the final acquired voice data is consistent with the hardware processing process of the target device, thereby ensuring the robustness and accuracy of the speech recognition model.

[0063] It should be noted that the cross-channel processing in the embodiment of the present invention is not the same as the cross-channel processing in the prior art. The cross-channel processing process in the prior art is generally: after collecting voice data through a microphone, the collected voice data is processed by the hardware of one or more cross-channel modules. In the embodiment of the present invention, since the voice data is input to the target device in the form of electrical signal transmission, the process of collecting voice data through a microphone or other collection device is omitted, that is, the process of the collection device obtaining voice data through air transmission can be omitted.

[0064] Exemplarily, cross-channel processing can use software or hardware devices with specific functions such as amplifiers, filters or analog-to-digital converters to process voice data, which can be set according to the cross-channel requirements of the target device, and the embodiments of the present invention are not limited to this.

[0065] In an optional embodiment of the present invention, the cross-channel processing of the voice data may include processing the voice data using any one or more of a microphone, a filter circuit, a signal amplification circuit, and an analog / digital converter of the target device.

[0066] Among them, the microphone can be used to receive the electrical signal of the voice data. Optionally, the input pin of the microphone can be used to receive the electrical signal of the voice data. The filter circuit can allow specific frequency components in the voice data to pass through, while greatly attenuating other frequency components, thereby filtering out interference noise in the voice data. Exemplarily, it can be a filter. The signal amplification circuit can be used to amplify the voice data for subsequent data processing. Exemplarily, it can be a power amplifier. The analog / digital converter can convert the analog signal into a digital signal.

[0067] In an embodiment of the present invention, after the microphone of the target device acquires the voice data, the voice data can be filtered by a filtering circuit to eliminate the interference of noise and acquire voice data of a specific frequency that meets the requirements. After filtering, the voice data can be amplified by a signal amplification circuit. Since the voice data acquired by the microphone of the target device is an analog signal, the amplified voice data can be converted from analog to digital by an analog / digital converter to convert the analog signal into a digital signal.

[0068] It should be noted that cross-channel processing may include processing voice data using any one or more of the microphone, filter circuit, signal amplification circuit, and analog / digital converter of the target device, and the embodiments of the present invention do not limit the specific processing method of cross-channel. It should also be noted that when the microphone of the target device is used for cross-channel processing, only the input pin of the microphone is used to obtain voice data. Since the input pin of the microphone is directly connected to other cross-channel modules, obtaining voice data through the input pin of the microphone can be understood as directly connecting the voice data in the form of an electrical signal to the cross-channel module connected to the microphone. For example, assuming that the input pin of the microphone is directly connected to the input end of the filter circuit, obtaining voice data using the input pin of the microphone can be understood as directly connecting the voice data in the form of an electrical signal to the input end of the filter circuit.

[0069] It is understandable that, in the embodiment of the present invention, the cross-channel module involved in the cross-channel processing of voice data can work independently of the target device. That is, the cross-channel processing flow of the target device on the voice data can be executed on the target device, or it can be executed independently from the target device. When the cross-channel processing flow is executed independently of the target device, it is necessary to ensure that each cross-channel module is consistent with the modules involved in the cross-channel processing flow in the target device, so as to ensure that when the cross-channel process is executed independently of the target device, the voice data obtained is consistent with the hardware processing process of the target device, thereby ensuring the accuracy of the voice training data.

[0070] It should also be noted that if the voice data input to the target device includes both pure human voice audio data and noise sample audio data, after the noise sample audio data is converted into a noise simulation signal through a noise simulation signal conversion device, the noise simulation signal also needs to refer to the above-mentioned voice data cross-channel processing method, and the voice simulation signal corresponding to the voice data is simultaneously subjected to cross-channel processing.

[0071] S130: Obtain cross-channel voice data after cross-channel processing.

[0072] The cross-channel voice data may be voice data output after the voice data has been processed through the cross-channel, and may be used to train a voice recognition model of a target device.

[0073] Correspondingly, after the speech data has been completely processed through the channels, the data after the processing through the channels, that is, the speech data through the channels, can be used as the final speech training data.

[0074] Optionally, if the voice data input to the target device includes both pure human voice audio data and noise sample audio data, the voice simulation signal corresponding to the voice data and the noise simulation signal corresponding to the noise sample audio data can be cross-channel processed at the same time, and the cross-channel voice data obtained by the cross-channel processing is used as the final voice training data. In this case, the voice training data obtained is noisy voice data, which can more realistically reflect the working environment of the target device.

[0075] Figure 1b FIG. 1 is a schematic diagram of the voice data channel processing in the prior art. Figure 1b As shown, in the prior art, the target device obtains voice data by directly recording the audio data of the speaker, and performs cross-channel processing on it. Specifically, the audio data recorded by the target device through the electret microphone passes through a signal amplifier, a filter, and an analog-to-digital converter, and is input into the CPU (Central Processing Unit / Processor) for storage to obtain the final voice training data. It can be seen that if the target device type is different, it is necessary to repeatedly collect voice training data for different target devices, which requires a huge waste of manpower and time costs. In addition, in the process of recording audio data with an electret microphone, it is very easy to introduce the influence of various external interference factors such as noise or distortion, resulting in a reduction in the quality of the voice training data.

[0076] In order to solve the above problem, in an embodiment of the present invention, voice training data is no longer obtained by directly recording audio data through a microphone. Instead, high-fidelity voice data is obtained in advance by professional equipment in a certain environment, and the voice data is directly input into the target device by propagation of electrical signals.

[0077] Figure 1c is a schematic diagram of a method for acquiring speech training data provided by the first embodiment of the present invention. In a specific example, Figure 1c As shown, by using professional recording equipment such as an electret microphone, audio data in a high-fidelity format can be recorded in a set environment, and the recorded audio data can be used as voice data. Optionally, the set environment can be an anechoic room or different working conditions of the target device, such as different driving environments of the target vehicle. At the same time, the audio format of the voice data can also be set according to the requirements of the speech recognition model, such as setting it to a 44K mono (mono) 16-bit format. After obtaining the voice data, it can be output through a voice analog signal conversion device such as a computer. For example, it is output to device A and device B (different target devices) through a computer device. It should be noted that since the input pin of the microphone of the target device is directly connected to the subsequent cross-channel module, in the process of voice data cross-channeling, it can be directly connected to the cross-channel module such as the filter without passing through the microphone. After the filter filters the voice data, the obtained voice data is input into the A / D (Analog / Digital) analog-to-digital conversion module to output a digital signal, and the final voice training data is obtained and stored in the CPU. It should be noted that Figure 1c The target devices only list device A and device B. In addition, more target devices such as device C or device D may be included. Meanwhile, the cross-channel modules of different target devices may be the same or different.

[0078] In addition, the speech data in the embodiment of the present invention may also include pure human voice audio data and noisy human voice audio data. Figure 1d is a schematic diagram of a method for acquiring speech training data and noise training data provided by Embodiment 1 of the present invention. In a specific example, Figure 1d As shown in FIG. 1 , by using professional recording equipment such as an electret microphone to record high-fidelity audio data, a high-fidelity voice dataset and an environmental noise dataset can be recorded separately, and the two types of audio data can be used as voice data at the same time. Specifically, the voice recognition model used for a vehicle can be used as an example. Figure 1d As shown, the specific voice data acquisition process may include:

[0079] Step 1: Collect high-fidelity vocal data.

[0080] High-fidelity voice audio can be recorded in a professional anechoic room or recording studio using professional recording equipment. The content can be wake-up audio data for wake-up training, such as "Hello, Xiaobai", or speech recognition sample audio data for speech recognition.

[0081] Step 2: Collect high-fidelity driving noise data in different driving scenarios, which can typically be divided into four driving scenarios.

[0082] Scene 1 can be: parking environment, idling, open windows, air conditioning off, and environmental noise decibel value between 50 and 55db; Scene 2 can be: urban roads, low-speed driving state: 40-60km / h, closed windows, air conditioning on, and environmental noise decibel value between 55 and 60db; Scene 3 can be: urban roads, low-speed driving state: 40-60km / h, open windows, air conditioning off, and environmental noise decibel value between 60 and 70db; Scene 4 can be: expressway, high-speed driving state: 80-120km / h, closed windows, air conditioning on, and environmental noise decibel value between 55 and 65db. In the above four scenes, each scene can collect 50 hours of environmental noise, which can include tire noise, wind noise, engine noise, noisy human voices or other motor vehicle noise and other different types of environmental noise. The recording equipment for collecting high-fidelity voice data and high-fidelity driving noise data can be a high-fidelity condenser analog microphone and a computer, and the recorded audio can be saved in a high-fidelity format with a sampling rate of 44.1k and mono 16-bit.

[0083] Step 3: The high-fidelity voice data recorded in step 1 is used as a sample audio data for speech recognition, and is played by a computer device. The analog signal of the voice data is provided to the cross-channel module of the target device through the computer, and the analog signal including only the voice signal is directly input to the target device through the propagation of electrical signals. After passing through the cross-channel modules such as filters and analog-to-digital converters, it is saved in the CPU as 16k mono 16-bit format audio data to obtain the cross-channel voice data in a quiet scene. The voice data is pure human voice data.

[0084] Step 4: Play the high-fidelity voice data and high-fidelity driving noise data recorded in step 1 using different computer devices respectively, provide voice simulation signals and noise simulation signals to the cross-channel module of the target device through the computer, input the two analog signals into the cross-channel module of the target device at the same time through the propagation of electrical signals, and save them into 16k mono 16-bit format audio data in the CPU after passing through the cross-channel modules such as filters and analog-to-digital converters to obtain cross-channel voice data in the noise scene. This voice data is noisy human voice audio data. Since the noise data sets in different driving scenes are different, the cross-channel noise addition process can be operated once for each driving scene to obtain noisy human voice audio data in different driving scenes.

[0085] The final cross-channel speech data obtained in the above process can be added to the training data set of the speech recognition model. After training, a recognition model with strong robustness and high accuracy can be obtained.

[0086] It can be seen that compared with the traditional data recording method, the speech training data acquisition method provided in the embodiment of the present invention no longer directly records human voice audio or environmental noise audio, but first uses a high-fidelity recording device to collect human voice audio and environmental noise audio as a high-fidelity audio data set, and then plays the high-fidelity audio data set file through a computer device, and directly inputs the analog signal played by the computer into the target device in the form of an electrical signal for subsequent filtering and analog-to-digital conversion. The above method can not only reduce the acquisition cost of speech data and improve the operability of generating speech data under different noise scenarios, but also avoid the problems of environmental reverberation, electromagnetic interference or signal attenuation introduced in the traditional method, thereby ensuring the quality of speech data. At the same time, the same batch of recorded high-fidelity audio data sets do not need to be collected separately on multiple target devices, and can be repeatedly applied to different types of target devices, so that high-quality speech data can be obtained at a low cost, which improves the versatility of speech data.

[0087] The embodiment of the present invention inputs the voice data in the voice data storage device to the target device in the form of electrical signal propagation, processes the voice data through a channel, obtains the voice training data after the channel processing, and trains the voice recognition model according to the voice training data, thereby solving the problems existing in the existing voice training data acquisition method, such as the inability to guarantee the quality of the voice training data and the high labor and time costs, so as to reduce the cost of collecting voice training data, improve the quality and versatility of voice training data, and further improve the training efficiency and recognition effect of the voice recognition model.

[0088] Embodiment 2

[0089] Figure 2 is a flow chart of a model training method provided by the second embodiment of the present invention. This embodiment is applicable to the case of training a highly robust speech recognition model. The method can be executed by a model training device, which can be implemented by software and / or hardware and can generally be integrated in an electronic device. The electronic device can be a device that applies a speech recognition model. Figure 2 As shown, the method includes the following operations:

[0090] S210, acquiring speech training data; wherein the speech training data is cross-channel speech data acquired by the speech training data acquisition method described in any embodiment of the present invention.

[0091] In the embodiment of the present invention, specifically, the electronic device can use pre-stored cross-channel voice data as voice training data, wherein the pre-stored cross-channel voice data is the cross-channel voice data output by the channel module of the target device.

[0092] S220: Training a speech recognition model according to the speech training data.

[0093] Accordingly, after acquiring the speech training data, the speech recognition model can be trained according to the speech training data. Model training using the speech training data acquired by the speech training data acquisition method provided by the embodiment of the present invention can improve the robustness and accuracy of the speech recognition model, thereby improving the training efficiency and recognition effect of the speech recognition model.

[0094] The embodiment of the present invention inputs the voice data in the voice data storage device to the target device in the form of electrical signal propagation, and processes the voice data through a channel to obtain voice training data, so as to train the voice recognition model according to the voice training data, thereby solving the problems of the existing voice training data acquisition method that the quality of the voice training data cannot be guaranteed and the labor cost and time cost are high, thereby reducing the cost of voice training data collection, improving the quality and versatility of voice training data, and further improving the training efficiency and recognition effect of the voice recognition model.

[0095] Embodiment 3

[0096] Figure 3 is a schematic diagram of a speech training data acquisition device provided in Embodiment 3 of the present invention, such as Figure 3 As shown, the device includes: a voice data acquisition module 310, a cross-channel processing module 320 and a cross-channel voice data acquisition module 330, wherein:

[0097] The voice data acquisition module 310 is used to input the voice data in the voice data storage device to the target device in the form of electrical signal transmission;

[0098] A cross-channel processing module 320, configured to process the voice data across channels;

[0099] The cross-channel voice data acquisition module 330 is used to acquire the cross-channel voice data after cross-channel processing.

[0100] Optionally, the voice data acquisition module may include: a first audio data acquisition unit, a first audio data conversion unit and a voice simulation signal acquisition unit, wherein: the first audio data acquisition unit is used to acquire wake-up audio data and / or voice recognition sample audio data through a voice acquisition device; the first audio data conversion unit is used to convert the wake-up audio data and / or voice recognition sample audio data into a voice simulation signal through a voice simulation signal conversion device; the voice simulation signal acquisition unit is used to input the voice simulation signal to the target device in the form of an electrical signal.

[0101] Optionally, the voice data acquisition module may also include: a second audio data acquisition unit, a noise data acquisition unit, a second audio data conversion unit, a noise data conversion unit and a voice and noise simulation signal acquisition unit, wherein: the second audio data acquisition unit is used to acquire wake-up audio data and / or voice recognition sample audio data through a voice acquisition device; the noise data acquisition unit is used to acquire noise sample audio data through a noise acquisition device; the second audio data conversion unit is used to convert the wake-up audio data and / or voice recognition sample audio data into a voice simulation signal through a voice simulation signal conversion device; the noise data conversion unit is used to convert the noise sample audio data into a noise simulation signal through a noise simulation signal conversion device; the voice and noise simulation signal acquisition unit is used to simultaneously input the voice simulation signal and the noise simulation signal into the target device in the form of electrical signal propagation.

[0102] Optionally, the cross-channel processing module may include: any one or more of a microphone unit, a filtering unit, a signal amplifying unit, and an analog / digital conversion unit.

[0103] Optionally, the voice and noise simulation signal acquisition unit may also be used to simultaneously input the voice simulation signal and the noise simulation signal into the same signal input point of the target device in the form of electrical signal propagation.

[0104] Optionally, the noise sample audio data may be any one of the interior noise of a vehicle in a parking lot, the interior noise of a vehicle traveling on a city road, and the interior noise of a vehicle traveling on a freeway; and the noise sample audio data may include any one or more of tire noise, wind noise, engine noise, noisy human voices, and motor vehicle noise.

[0105] The above-mentioned speech training data acquisition device can execute the speech training data acquisition method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method. For technical details not described in detail in this embodiment, please refer to the speech training data acquisition method provided by any embodiment of the present invention.

[0106] Since the speech training data acquisition device introduced above is a device that can execute the speech training data acquisition method in the embodiment of the present invention, based on the speech training data acquisition method introduced in the embodiment of the present invention, the technicians in this field can understand the specific implementation of the speech training data acquisition device in this embodiment and its various variations, so how the speech training data acquisition device implements the speech training data acquisition method in the embodiment of the present invention is not described in detail here. As long as the technicians in this field implement the device used by the speech training data acquisition method in the embodiment of the present invention, it belongs to the scope of protection of this application.

[0107] Embodiment 4

[0108] Figure 4 is a schematic diagram of a model training device provided by Embodiment 4 of the present invention, such as Figure 4 As shown, the device includes: a speech training data acquisition module 410 and a speech recognition model training module 420, wherein:

[0109] A speech training data acquisition module 410 is used to acquire speech training data; wherein the speech training data is cross-channel speech data acquired by the speech training data acquisition method provided by any embodiment of the present invention;

[0110] The speech recognition model training module 420 is used to train the speech recognition model according to the speech training data.

[0111] The above-mentioned model training device can execute the model training method provided by any embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not described in detail in this embodiment, please refer to the model training method provided by any embodiment of the present invention.

[0112] Since the model training device introduced above is a device that can execute the model training method in the embodiment of the present invention, based on the model training method introduced in the embodiment of the present invention, the technical personnel in the field can understand the specific implementation of the model training device of this embodiment and its various variations, so how the model training device implements the model training method in the embodiment of the present invention is not introduced in detail here. As long as the technical personnel in the field implement the device used in the model training method in the embodiment of the present invention, it belongs to the scope of protection of this application.

[0113] Embodiment 5

[0114] Figure 5 This is a schematic diagram of the structure of an electronic device provided in Embodiment 5 of the present invention. Figure 5 A block diagram of an electronic device 512 suitable for implementing embodiments of the present invention is shown. Figure 5 The electronic device 512 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0115] like Figure 5 As shown, the electronic device 512 is in the form of a general purpose computing device. The components of the electronic device 512 may include, but are not limited to: one or more processors 516, a storage device 528, and a bus 518 connecting different system components (including the storage device 528 and the processor 516).

[0116] Bus 518 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus architectures. For example, these architectures include but are not limited to Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MCA) bus, Enhanced ISA bus, Video Electronics Standards Association (VESA) local bus and Peripheral Component Interconnect (PCI) bus.

[0117] The electronic device 512 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the electronic device 512, including volatile and non-volatile media, removable and non-removable media.

[0118] The storage device 528 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 530 and / or cache memory 532. The electronic device 512 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 534 may be used to read and write non-removable, non-volatile magnetic media ( Figure 5 not shown, usually called a "hard drive"). Although Figure 5 Not shown in the figure, a disk drive for reading and writing a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing a removable non-volatile optical disk (e.g., a read-only optical disk (Compact Disc-Read Only Memory, CD-ROM), a digital video disk (Digital Video Disc-Read Only Memory, DVD-ROM) or other optical media) may be provided. In these cases, each drive may be connected to the bus 518 via one or more data medium interfaces. The storage device 528 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present invention.

[0119] A program 536 having a set (at least one) of program modules 526 may be stored, for example, in a storage device 528, such program modules 526 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or some combination thereof may include an implementation of a network environment. The program modules 526 generally perform the functions and / or methods of the embodiments described herein.

[0120] The electronic device 512 may also communicate with one or more external devices 514 (e.g., keyboard, pointing device, camera, display 524, etc.), may also communicate with one or more devices that enable a user to interact with the device 512, and / or communicate with any device that enables the electronic device 512 to communicate with one or more other computing devices (e.g., network card, modem, etc.). Such communication may be performed through an input / output (I / O) interface 522. In addition, the electronic device 512 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN) and / or a public network, such as the Internet) through a network adapter 520. As shown, the network adapter 520 communicates with other modules of the electronic device 512 through a bus 518. It should be understood that although not shown in the figure, other hardware and / or software modules can be used in conjunction with the electronic device 512, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, disk arrays (Redundant Arrays of Independent Disks, RAID) systems, tape drives, and data backup storage systems.

[0121] The processor 516 executes various functional applications and data processing by running the programs stored in the storage device 528, such as implementing the method for acquiring speech training data provided in the above embodiment of the present invention.

[0122] That is, when the processing unit executes the program, it realizes: inputting the voice data in the voice data storage device to the target device in the form of electrical signal transmission; processing the voice data across channels; and acquiring the across-channel voice data after the across-channel processing.

[0123] Alternatively, when the processing unit executes the program, the following steps are achieved: acquiring speech training data; wherein the speech training data is cross-channel speech data acquired by the speech training data acquisition method provided by any embodiment of the present invention; and training a speech recognition model according to the speech training data.

[0124] Embodiment 6

[0125] Embodiment 6 of the present invention further provides a computer storage medium storing a computer program, wherein the computer program, when executed by a computer processor, is used to execute the method for acquiring speech training data described in any one of the above embodiments of the present invention: inputting speech data in a speech data storage device to a target device in the form of an electrical signal; processing the speech data across channels; and acquiring cross-channel speech data after the cross-channel processing.

[0126] Alternatively, the computer program, when executed by a computer processor, is used to perform the model training method described in any of the above embodiments of the present invention: obtaining speech training data; wherein the speech training data is cross-channel speech data obtained by the speech training data acquisition method described in any of the embodiments; and training a speech recognition model based on the speech training data.

[0127] The computer storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, - but not limited to - an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0128] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0129] The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wire, optical cable, Radio Frequency (RF), etc., or any suitable combination of the foregoing.

[0130] Computer program code for performing the operations of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0131] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A method for acquiring speech training data, characterized in that: include: Inputting the voice data in the voice data storage device to the target device in the form of electrical signal transmission; Processing the voice data through a channel; Obtaining cross-channel voice data after cross-channel processing; The cross-channel processing is a process of collecting the voice data through a target device, which is used to ensure that the voice data finally obtained is consistent with the hardware processing process of the target device; The cross-channel voice data after the cross-channel processing is used to train a voice recognition model of a target device.

2. The method according to claim 1, characterized in that The method of inputting the voice data in the voice data storage device to the target device in the form of an electrical signal transmission includes: Acquire wake-up audio data and / or speech recognition sample audio data through a voice acquisition device; Converting the wake-up audio data and / or the speech recognition sample audio data into a speech analog signal through a speech analog signal conversion device; The voice analog signal is input to the target device in the form of an electrical signal.

3. The method according to claim 1, characterized in that The method of inputting the voice data in the voice data storage device to the target device in the form of an electrical signal transmission includes: Acquire wake-up audio data and / or speech recognition sample audio data through a voice acquisition device; Acquire noise sample audio data through a noise collection device; Converting the wake-up audio data and / or the speech recognition sample audio data into a speech analog signal through a speech analog signal conversion device; Converting the noise sample audio data into a noise analog signal by a noise analog signal conversion device; At the same time, the voice simulation signal and the noise simulation signal are respectively input to the target device in the form of electrical signal propagation.

4. The method according to claim 1, characterized in that: The cross-channel processing of the voice data includes: The voice data is processed using any one or more of a microphone, a filter circuit, a signal amplification circuit, and an analog / digital converter of the target device.

5. The method according to claim 3, characterized in that: The simultaneous inputting of the voice simulation signal and the noise simulation signal into the target device in the form of electrical signal propagation is input into the same signal input point of the target device.

6. The method according to claim 3, characterized in that The noise sample audio data is any one of the vehicle interior noise when the vehicle is in a parking lot, the vehicle interior noise when the vehicle is traveling on a city road, and the vehicle interior noise when the vehicle is traveling on a freeway; The noise sample audio data includes any one or more of tire noise, wind noise, engine noise, noisy human voice, and motor vehicle noise.

7. A model training method, characterized in that: include: Acquire speech training data; wherein the speech training data is cross-channel speech data acquired by the speech training data acquisition method according to any one of claims 1 to 6; A speech recognition model is trained according to the speech training data.

8. A speech training data acquisition device, characterized in that: include: A voice data acquisition module, used to input the voice data in the voice data storage device to the target device in the form of electrical signal transmission; A cross-channel processing module, used for processing the voice data across channels; A cross-channel voice data acquisition module, used to acquire the cross-channel voice data after cross-channel processing; The cross-channel processing is a process of collecting the voice data through a target device, which is used to ensure that the voice data finally obtained is consistent with the hardware processing process of the target device; The cross-channel voice data after the cross-channel processing is used to train a voice recognition model of a target device.

9. The device according to claim 8, characterized in that The voice data acquisition module comprises: A first audio data acquisition unit, configured to acquire wake-up audio data and / or speech recognition sample audio data through a voice acquisition device; A first audio data conversion unit, configured to convert the wake-up audio data and / or the speech recognition sample audio data into a speech analog signal through a speech analog signal conversion device; The voice analog signal acquisition unit is used to input the voice analog signal into the target device in the form of an electrical signal.

10. The device according to claim 8, characterized in that The voice data acquisition module comprises: A second audio data acquisition unit, used to acquire wake-up audio data and / or speech recognition sample audio data through a voice acquisition device; A noise data acquisition unit, used to acquire noise sample audio data through a noise acquisition device; A second audio data conversion unit, configured to convert the wake-up audio data and / or the speech recognition sample audio data into a speech analog signal through a speech analog signal conversion device; A noise data conversion unit, configured to convert the noise sample audio data into a noise simulation signal through a noise simulation signal conversion device; The voice and noise simulation signal acquisition unit is used to simultaneously input the voice simulation signal and the noise simulation signal into the target device in the form of electrical signal propagation.

11. The device according to claim 8, characterized in that The cross-channel processing module includes: any one or more of a microphone unit, a filtering unit, a signal amplifying unit, and an analog / digital conversion unit.

12. A model training device, characterized in that: include: A speech training data acquisition module, used to acquire speech training data; wherein the speech training data is cross-channel speech data acquired by the speech training data acquisition method according to any one of claims 1 to 6; The speech recognition model training module is used to train the speech recognition model according to the speech training data.

13. An electronic device, characterized in that: The electronic device comprises: one or more processors; A storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the speech training data acquisition method as described in any one of claims 1 to 6, or implement the model training method as described in claim 7.

Images