Signal noise reduction method and device and electronic equipment

Abstract

The invention discloses a signal noise reduction method and device and electronic equipment. The method comprises the following steps: receiving an analog audio signal subjected to channel decoding; collecting sound wave information generated by the analog audio signal to obtain a to-be-identified audio signal; identifying the noise type in the audio signal to be identified; noise reduction is carried out on environment noise in the to-be-recognized audio signal by adopting a first signal noise reduction strategy; performing noise reduction on channel noise in the to-be-identified audio signal by adopting a second signal noise reduction strategy; and taking the to-be-recognized audio signal after noise reduction as a target audio signal. The technical problem that the noise reduction effect is limited due to the fact that the correlation technique focuses on the suppression of the environmental noise and neglects the processing of the channel noise is solved.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a method, device and electronic device for signal noise reduction. Background Art

[0002] Direct satellite terminals generate channel noise due to long-distance signal transmission, and the antenna of the direct satellite terminal is affected by the polarity of the antenna and the interference of the human body on the signal, which leads to the current terminal intelligent use of external speaker for calls, which is easily affected by environmental noise. The relevant technology only suppresses or eliminates environmental noise, but the noise reduction effect is poor and the call quality is poor.

[0003] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0004] The embodiments of the present application provide a method, device and electronic device for signal noise reduction, so as to at least solve the technical problem that the related art focuses on suppressing environmental noise, ignores the processing of channel noise, and has limited noise reduction effect.

[0005] According to one aspect of an embodiment of the present application, a method for signal noise reduction is provided, comprising: receiving an analog audio signal after channel decoding, wherein the analog audio signal is used for an audio output device to generate sound waves; collecting sound wave information generated by the analog audio signal to obtain an audio signal to be identified, wherein the audio signal to be identified includes voice information after channel decoding and external environmental sound information; identifying the type of noise in the audio signal to be identified, wherein the noise type includes environmental noise and channel noise; using a first signal noise reduction strategy to reduce the environmental noise in the audio signal to be identified, wherein the first signal noise reduction strategy is used to eliminate the environmental noise; using a second signal noise reduction strategy to reduce the channel noise in the audio signal to be identified, wherein the second signal noise reduction strategy is used to repair the signal distortion caused by the channel noise; and using the denoised audio signal to be identified as the target audio signal.

[0006] In some embodiments of the present application, identifying the type of noise in an audio signal to be identified includes: constructing a first feature vector of the audio signal, wherein the first feature vector includes at least one of the following: spectral features, time domain features, and channel state information features; matching the first feature vector with a noise feature representation to obtain a matching result, wherein the noise feature representation is generated by a noise feature extraction model; and determining the noise type corresponding to the matching result.

[0007] In some embodiments of the present application, a first feature vector is matched with a noise feature representation to obtain a matching result, including: calculating a first similarity between the first feature vector and the first noise feature representation, wherein the first noise feature representation includes any one noise feature representation in a set of environmental noise feature representations; determining a first target similarity in a first similarity set where the first similarity is located, wherein the first target similarity includes a maximum value in the first similarity set; calculating a second similarity between the first feature vector and a second noise feature representation, wherein the second noise feature representation includes any one noise feature representation in a set of channel noise feature representations; determining a second target similarity in a second similarity set where the second similarity is located, wherein the second target similarity includes a maximum value in the second similarity set; and determining a matching result using the first target similarity and the second target similarity.

[0008] In some embodiments of the present application, a matching result is determined using a first target similarity and a second target similarity, including: comparing the first target similarity with a first preset threshold, and determining that environmental noise exists in the audio signal when the first similarity is greater than the first preset threshold; comparing the second target similarity with a second preset threshold, and determining that channel noise exists in the audio signal when the second similarity is greater than the second preset threshold.

[0009] In some embodiments of the present application, a first signal noise reduction strategy is used to reduce the ambient noise in the audio signal to be recognized, including: cutting the audio signal to be recognized into a speech segment and a non-speech segment; for the speech segment in the audio signal to be recognized, using a third preset threshold to adjust the noise threshold of the filter, wherein the third preset threshold is greater than the noise threshold; for the non-speech segment in the audio signal to be recognized, using a fourth preset threshold to adjust the noise threshold of the filter, wherein the fourth preset threshold is less than the noise threshold; and using the filter after adjusting the noise threshold to perform a filtering operation on the audio signal to be recognized.

[0010] In some embodiments of the present application, a second signal denoising strategy is used to reduce channel noise in an audio signal to be identified, including: identifying the sparsity of the audio signal to be identified in a mathematical transform domain, wherein the mathematical transform domain includes mathematical transform domains other than the time domain; when the audio signal to be identified is sparse in the mathematical transform domain, constructing a sparse representation of the audio signal to be identified in the mathematical transform domain; extracting target observation values ​​from the sparse representation to obtain observation vectors, wherein the target observation values ​​are used to reconstruct the audio signal to be identified; based on the observation vectors, estimating a channel model using a compressed sensing algorithm, wherein the channel model is used to describe the distortion and attenuation of the audio signal to be identified; and reconstructing the audio signal to be identified using the channel model.

[0011] In some embodiments of the present application, a channel model is used to reconstruct an audio signal to be identified, including: converting the audio signal to be identified into a sparse representation in a mathematical transform domain; using an observation vector and a channel model to restore the original audio signal corresponding to the channel noise in the audio to be identified to obtain a first audio signal; performing inverse channel compensation on the first audio signal to obtain a second audio signal, wherein the inverse channel compensation includes adjusting the spectrum amplitude; converting the second audio signal into a time domain representation to obtain a reconstructed audio signal, wherein the reconstructed audio signal is used for an audio output device of a cell device to generate sound waves.

[0012] According to another aspect of an embodiment of the present application, a signal noise reduction device is also provided, including: a receiving module, used to receive an analog audio signal after channel decoding, wherein the analog audio signal is used for an audio output device to generate sound waves; an acquisition module, used to acquire sound wave information generated by the analog audio signal to obtain an audio signal to be identified, wherein the audio signal to be identified includes voice information after channel decoding and external environmental sound information; an identification module, used to identify the type of noise in the audio signal to be identified, wherein the noise type includes environmental noise and channel noise; an execution module, used to adopt a first signal noise reduction strategy to reduce the environmental noise in the audio signal to be identified, wherein the first signal noise reduction strategy is used to eliminate the environmental noise; adopt a second signal noise reduction strategy to reduce the channel noise in the audio signal to be identified, wherein the second signal noise reduction strategy is used to repair the signal distortion caused by the channel noise; a determination module, used to use the denoised audio signal to be identified as the target audio signal.

[0013] According to another aspect of the embodiments of the present application, an electronic device is provided, including: a memory and a processor, the memory being used to store program instructions; the processor being connected to the memory and being used to execute the method for realizing the above-mentioned signal noise reduction.

[0014] According to another aspect of the embodiments of the present application, a non-volatile storage medium is provided, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the above-mentioned signal noise reduction method by running the computer program.

[0015] In the embodiment of the present application, by collecting the audio signal after the conversion of the satellite communication signal, identifying the noise type in the audio signal, and processing the corresponding noise respectively, the audio signal after noise reduction is obtained, thereby achieving the purpose of significantly reducing noise interference, thereby achieving the technical effect of improving voice clarity and improving communication quality, and further solving the technical problem that the related technology focuses on suppressing environmental noise, ignores the processing of channel noise, and has limited noise reduction effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0017] Figure 1 It is a hardware structure block diagram of a computer terminal according to a signal noise reduction method of an embodiment of the present application;

[0018] Figure 2 is a flow chart of a signal noise reduction method according to an embodiment of the present application;

[0019] Figure 3 It is a schematic diagram of the principle of audio signal feature extraction in a signal noise reduction method according to an embodiment of the present application;

[0020] Figure 4 is an overall framework diagram of a signal noise reduction method according to an embodiment of the present application;

[0021] Figure 5 is a schematic diagram of the principle of environmental noise reduction of a signal noise reduction method according to an embodiment of the present application;

[0022] Figure 6 It is a schematic diagram of the principle of reducing environmental noise and channel noise according to a signal noise reduction method in an embodiment of the present application;

[0023] Figure 7 It is a structural schematic diagram of a signal noise reduction device according to an embodiment of the present application. DETAILED DESCRIPTION

[0024] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution 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. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0025] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] In order to better understand the embodiments of the present application, the technical terms involved in the embodiments of the present application are explained as follows:

[0027] Channel Decoding: refers to restoring the received coded signal to its original data form, including error detection and correction to ensure the accuracy and integrity of the transmitted information.

[0028] Pulse Code Modulation Decoding (PCM for short) refers to converting digital signals encoded by PCM technology into original analog audio signals, thereby achieving accurate reproduction and playback of audio information.

[0029] Channel Noise: refers to the fact that during the signal transmission process, due to the inherent characteristics of the communication channel or the influence of external environmental factors (such as electromagnetic interference, atmospheric disturbances, etc.), the signal will be subjected to different degrees of noise superposition, resulting in a decrease in signal quality and reduced accuracy of information transmission.

[0030] Ambient Noise: refers to the sound generated by surrounding non-communication sources in any given environment. Such sound is usually not the target component of the communication signal, but it will inevitably interfere with the signal when receiving the audio signal, reducing the clarity of the communication and the quality of user perception.

[0031] Channel Estimation: refers to the process of predicting and determining the characteristics of the communication channel, including the transmission characteristics of the channel, noise statistics, etc. Through channel estimation, the system can obtain the current channel state information, which is then used for signal preprocessing, adjustment of encoding and decoding parameters, and effective channel compensation technology to overcome signal distortion caused by the channel and ensure high-quality data transmission.

[0032] With the development of science and technology, direct satellite terminals are increasingly used in the field of communications. However, due to the huge difference between satellite communications and ordinary terrestrial network communications, direct satellite terminals face many challenges in actual use. The most important problem is the long distance between the satellite and the earth. Even low-Earth orbit satellites are at least several hundred kilometers away from the earth. During such long-distance signal transmission, external environmental interference may cause signal transmission data loss or distortion.

[0033] In addition, the antenna of the satellite terminal is affected by the polarity of the antenna and the interference of the human body on the signal, so the current terminal can only use the external speaker to make calls. However, the external speaker will greatly interfere with the sound of the terminal, which inevitably increases the difficulty of external noise reduction.

[0034] The voice noise reduction method adopted by the related technology is mainly to perceive the peripheral environmental noise and noise through the main and secondary microphones of the terminal, and suppress or eliminate the noise or noise of the external environment through relevant algorithms. It focuses on suppressing environmental noise and ignores the management of channel noise, resulting in limited noise reduction efficiency and poor call clarity. In order to solve the above technical problems, the embodiments of the present application provide corresponding solutions, which are described in detail below.

[0035] The signal noise reduction method embodiment provided in the embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computer terminal for implementing a method for signal noise reduction. Figure 1 As shown, the computer terminal 10 may include one or more (102a, 102b, ..., 102n are used to illustrate) processors (the processor may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission module 106 for communication functions connected via a wired and / or wireless network. In addition, it may also include: a display, a keyboard, a cursor control device, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, and a BUS bus. Those skilled in the art can understand that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0036] It should be noted that the one or more processors and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10. As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0037] The memory 104 can be used to store software programs and modules of application software, such as program instructions / data storage devices corresponding to the signal noise reduction method in the embodiment of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, the above-mentioned signal noise reduction method is realized. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0038] The transmission module 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission module 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission module 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0039] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 .

[0040] It should be noted that, in some optional embodiments, the above Figure 1 The computer terminal shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of hardware elements and software elements. It should be noted that Figure 1 This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the computer terminal described above.

[0041] In the above-mentioned operating environment, an embodiment of the present application provides an embodiment of a method for signal noise reduction. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0042] Figure 2 is a flow chart of a signal noise reduction method according to an embodiment of the present application. Figure 2 As shown, the method comprises the following steps:

[0043] Step S202: receiving an analog audio signal after channel decoding, wherein the analog audio signal is used by an audio output device to generate sound waves.

[0044] In the above step S202, the analog audio signal is a continuously changing electrical signal used to represent audio information. In some embodiments, after receiving the signal transmitted via the satellite, the receiving end processing unit first restores the coded signal through a channel decoder to obtain an analog audio signal, and then sends the analog audio signal to an audio output device, such as a receiver, to generate audible sound waves.

[0045] Step S204 , collecting sound wave information generated by the analog audio signal to obtain an audio signal to be identified, wherein the audio signal to be identified includes voice information after channel decoding and external environmental sound information.

[0046] In the above step S204, the sound wave information refers to the sound waves generated and propagated by the audio output device, including the information carried by the audio signal. The audio signal to be identified refers to an audio signal formed by mixing the sound waves generated by the analog audio signal with the collected ambient sound. In some embodiments, the sound waves generated by the earpiece and the external ambient sound are simultaneously collected by the microphone of the terminal, and the collected information is converted into an electrical signal, which is pre-processed to form an audio signal to be identified.

[0047] Step S206: identifying the noise type in the audio signal to be identified, wherein the noise type includes environmental noise and channel noise.

[0048] In the above step S206, in the satellite terminal call scenario, noise can be divided into two categories, namely, environmental noise and channel noise. Environmental noise comes from the surrounding environment, such as wind noise, vehicle noise, etc.; channel noise is noise caused by interference and distortion in signal transmission. In some embodiments, an intelligent analysis algorithm or model can be used to analyze the collected audio signal to be identified to identify the type of noise therein.

[0049] In order to accurately identify the type of noise so as to take targeted noise reduction measures and improve the efficiency and accuracy of noise reduction, the following steps can be used to identify the type of noise: construct a first feature vector of the audio signal, wherein the first feature vector includes at least one of the following: spectral features, time domain features, and channel state information features; match the first feature vector with a noise feature representation to obtain a matching result, wherein the noise feature representation is generated by a noise feature extraction model; and determine the noise type corresponding to the matching result.

[0050] (1) Constructing the first eigenvector of the audio signal: The first eigenvector is a mathematical vector that contains the key attributes obtained after feature extraction of the audio signal, such as spectral features, time domain features, or channel state information features. Spectral features are used to describe the distribution and intensity of audio signals at different frequency components. Time domain features are used to describe the characteristics of audio signals that change over time, such as the amplitude, time interval, and trend of the signal. Channel state information features are used to reflect indicators of the current characteristics of the channel, including channel gain, multipath effect, delay, etc.

[0051] In some embodiments, the spectrum features can be extracted by short-time Fourier transform (STFT), the time domain features can be extracted by autocorrelation function, and the channel state information features can be obtained by channel measurement technology. According to system requirements and noise types, one or more features are selected for fusion to form a first feature vector that fully reflects the characteristics of the audio signal.

[0052] Figure 3 is a schematic diagram of the principle of audio signal feature extraction according to a signal noise reduction method of an embodiment of the present application. In some embodiments, the first feature vector is constructed using the following steps: Figure 3 As shown:

[0053] S302: The microphone converts sound waves into analog electrical signals.

[0054] The microphone is an acoustic sensor. Its diaphragm is affected by the pressure fluctuations of sound waves and produces corresponding mechanical vibrations. These vibrations are converted into electrical signals, and the energy of the sound waves is converted into analog electrical signals through the converter inside the microphone.

[0055] S304: Silence removal and voice activity detection (Voice Activity Detection, VAD for short).

[0056] Silence removal or VAD is the process of identifying segments of an audio stream that actually contain speech signals, while removing silence or non-speech portions. In some embodiments, the VAD algorithm can determine whether there is speech activity within a certain period of time based on indicators such as the energy, zero-crossing rate, and spectral characteristics of the signal. If no speech activity (i.e., silence) is detected, this portion of the signal will be marked and may be ignored in subsequent processing, thereby improving system efficiency and reducing unnecessary calculations.

[0057] S306: Frame division and windowing.

[0058] The audio signal is framed, which means that the continuous audio signal is divided into a series of short periods of fixed length or overlap, each of which is called a frame. In some embodiments, the frame length can be set to 25 milliseconds, and the frame shift can be set to 10 milliseconds (i.e., the frames are partially overlapped) to ensure the stability of the speech while maintaining the feasibility of the calculation. After framing, each frame of the signal will be processed by windowing. In some embodiments, a Hamming window, a Hamming window or other window functions can be used. The purpose of windowing is to reduce signal mutations between frames and avoid spectrum leakage, thereby improving the accuracy of subsequent feature extraction.

[0059] S308: Mel Frequency Cepstral Coefficients (MFCC) feature extraction.

[0060] 1) Pre-emphasis: Pre-process each frame of the signal. For example, a pre-emphasis filter can be used to reduce the redundancy of the speech signal and enhance the high-frequency components in the signal to better reflect the characteristics of the speech.

[0061] 2) Discrete Fourier Transform (DFT): Each frame of the signal is converted to the frequency domain, for example, by performing a fast Fourier transform to convert the signal from the time domain to the frequency domain, which is used to analyze the spectral characteristics of the signal.

[0062] 3) Mel filter bank: The spectral data is passed through a set of filters evenly distributed on the Mel frequency scale to capture the important frequency band information in human auditory perception.

[0063] 4) Logarithmic energy spectrum: The energy spectrum of the filter output is converted into a logarithmic energy spectrum.

[0064] 5) Discrete Cosine Transform (DCT): The logarithmic energy spectrum is then subjected to discrete cosine transform to obtain MFCC coefficients, which are a set of cepstrum coefficients that reflect the spectrum envelope.

[0065] 6) Retain low-frequency cepstral coefficients: The first few (e.g., the first 12) of the MFCC coefficients are retained because they contain the main information of the speech signal, while high-frequency cepstral coefficients are usually associated with noise and are therefore ignored.

[0066] Through the above steps, each frame of audio signal is finally converted into a 12-dimensional MFCC feature vector, which constitutes the feature representation of the speech signal (ie, the first feature vector).

[0067] (2) Matching the first feature vector with the noise feature representation to obtain a matching result: the noise feature representation refers to a vector or data structure generated by a noise feature extraction model, for example, a deep learning-based model or a traditional signal processing algorithm, used to represent the characteristics of a specific type of noise.

[0068] In some embodiments, the noise feature extraction model is constructed in the following manner:

[0069] 1) Data preparation: including environmental speech model library and channel speech model library. The environmental speech model library is to collect and organize a large number of environmental sound samples, including but not limited to recording data in different scenarios such as offices, shopping malls, and outdoors; the channel speech model library is to accumulate signal data in satellite communication environments, including signal distortion, interference, and various channel status information.

[0070] 2) Feature Engineering: Preprocess the environmental sound samples and channel signal data, including but not limited to noise removal, standardization, segmentation and other steps, to prepare the data set for model training; then extract features such as spectral features, time domain features and channel state information features, which will be used to train the model to identify noise types.

[0071] 3) Model training: Use a deep learning framework, such as the BERT model (Bidirectional Encoder Representations from Transformers), to train the environmental noise and channel noise separately. For example, the data in the environmental speech model library is used to train the model to identify the characteristics of environmental noise, while the data in the channel speech model library is used to learn the characteristics of channel noise. In some embodiments, supervised learning techniques can be used, that is, noise samples are used as input and the corresponding noise type is used as the output label to train the model to accurately distinguish and identify different types of noise.

[0072] In some embodiments, when a channel speech signal is recognized using a BERT model, the BERT model may be trained in the following manner:

[0073] ① Data preprocessing and input representation construction: A microphone (MIC) can be used to collect a large number of speech signal samples in various channel environments, and the collected speech signals can be pre-filtered, amplified and sampled to ensure the clarity and quality of the signals, and converted into digital signal formats at the same time; Short-Time Fourier Transform (STFT) or Mel-Frequency Cepstral Coefficient (MFCC) and other technologies can be used to extract time-frequency features from digital speech signals and convert them into feature vector representations that can be understood and processed by the model; the extracted feature vectors are combined with position codes and other possible input vectors (such as pitch and timbre codes) to generate a comprehensive input representation for each time step.

[0074] ② Pre-training of the BERT model: It can be trained through the Masked Language Model (MLM), that is, on the unlabeled text data, some words are randomly masked, and the BERT model is trained to predict the correct form of the masked words; or the Next Sentence Prediction (NSP) training is used, that is, two sentences are provided as input, and the training model predicts whether the second sentence is the continuation of the first sentence.

[0075] ③ Speech signal input layer processing: receiving and digitizing speech signals, that is, using MIC to receive, and converting analog signals into digital signals through preprocessing technology; feature extraction, that is, using algorithms such as short-time Fourier transform or MFCC to extract the time-frequency features of speech signals and construct the representation of the input layer.

[0076] ④ Hidden layer multi-head self-attention mechanism training: linear transformation, that is, the feature representation of the input layer is linearly transformed to obtain the query (Q), key (K) and value (V) matrices; self-attention head calculation, that is, each self-attention head calculates the attention weight of each position in the input sequence respectively, and the weight reflects the degree of correlation between words in the sequence; matrix multiplication and weight matrix generation, that is, each self-attention head performs the dot product operation of the query matrix and the key matrix to obtain the attention weight matrix, and then multiplies the weight matrix with the value matrix to generate feature representation; result splicing, that is, splicing the feature representations generated by all self-attention heads to form a higher-level comprehensive feature representation. This step is repeated in subsequent layers to generate more abstract and semantically rich representations.

[0077] ⑤ Output layer and speech recognition: prediction mechanism, that is, using methods such as maximum likelihood estimation or cross entropy loss to predict the most likely word at each time step based on the feature representation generated by the hidden layer; speech recognition, that is, converting the prediction results into text form and outputting the recognized words or phrases.

[0078] ⑥Model evaluation and optimization: Through performance evaluation on the validation set, adjust model parameters, and optimize algorithm configuration to improve the accuracy and robustness of the model in channel speech detection and recognition tasks.

[0079] 4) Model fusion and optimization: The environmental noise model and the channel noise model are integrated to build a comprehensive noise feature extraction model that can simultaneously process and identify both types of noise. In some embodiments, the model can be optimized by adjusting hyperparameters, increasing the amount of training data, or using more advanced deep learning techniques to improve the accuracy and efficiency of the model in identifying noise.

[0080] Noise representation is generated using a noise feature extraction model. In some embodiments, the extracted first feature vector can be input into a comprehensive noise feature extraction model. The model will analyze the input features and match them with feature representations in an environmental speech model library and a channel speech model library.

[0081] By matching the similarity between the first feature vector and the noise feature representation, the source of the noise can be accurately determined, specifically: calculating the first similarity between the first feature vector and the first noise feature representation, wherein the first noise feature representation includes any one noise feature representation in the environmental noise feature representation set; determining the first target similarity in the first similarity set where the first similarity is located, wherein the first target similarity includes the maximum value in the first similarity set; calculating the second similarity between the first feature vector and the second noise feature representation, wherein the second noise feature representation includes any one noise feature representation in the channel noise feature representation set; determining the second target similarity in the second similarity set where the second similarity is located, wherein the second target similarity includes the maximum value in the second similarity set; and determining the matching result using the first target similarity and the second target similarity.

[0082] 1) Calculate the first similarity: The environmental noise feature representation refers to the mathematical representation of a specific environmental noise pattern. For example, the noise feature representation obtained in the environmental speech model library through data preprocessing and feature extraction technology may include the performance of the noise in the spectral domain, time domain, and statistical characteristics.

[0083] In some embodiments, a pre-trained environmental noise recognition model can be used to compare the first feature vector with each noise feature representation in the environmental noise feature representation set. For example, the similarity between the first feature vector and each environmental noise feature representation can be calculated using metrics such as cosine similarity and Pearson correlation coefficient to obtain a first similarity set.

[0084] 2) Determine the first target similarity: In some embodiments, the maximum value can be found in the first similarity set, and the maximum value is determined as the first target similarity. The first target similarity can be used to clarify the main characteristics of the environmental noise and provide a basis for subsequent environmental noise reduction strategies.

[0085] 3) Calculating the second similarity: The channel noise feature representation refers to the noise feature representation obtained by analyzing the characteristics of the satellite communication channel. For example, the noise feature representation obtained by data preprocessing and feature extraction technology in the channel speech model library may include the spectrum, delay and multipath effect of the channel noise.

[0086] In some embodiments, a channel noise recognition model may be used to compare the first feature vector with each noise feature representation in the channel noise feature representation set. For example, similarity between the first feature vector and each channel noise feature representation is calculated by similarity measurement, such as Euclidean distance, correlation analysis, etc., to obtain a second similarity set.

[0087] 4) Determine the second target similarity: In some embodiments, a maximum value may be found in the second similarity set, and the maximum value may be determined as the second target similarity. The second target similarity may be used to indicate the intensity and type of noise, thereby providing accurate guidance for signal repair and optimization.

[0088] 5) Determine the matching result using the first target similarity and the second target similarity: Analyze the numerical values ​​and differences between the first target similarity and the second target similarity to determine the relative importance or type of environmental noise and channel noise in the signal. For example, if the first target similarity is significantly greater than the second target similarity, it can be inferred that the main noise is environmental noise; otherwise, it is mainly channel noise; when the two noises exist at the same time and the values ​​are close, it can be considered that the signal contains both environmental noise and channel noise, which need to be processed separately.

[0089] In real-time audio signal processing, in order to improve the response speed of noise reduction processing, the following steps can be used to determine the matching result, specifically: comparing the first target similarity with the first preset threshold, and when the first similarity is greater than the first preset threshold, determining that the audio signal exists environmental noise; comparing the second target similarity with the second preset threshold, and when the second similarity is greater than the second preset threshold, determining that the audio signal exists channel noise.

[0090] In the above steps, the first preset threshold is used to determine whether the first target similarity reaches a threshold value sufficient to determine the presence of environmental noise, and can be set according to the standard level of environmental noise or the prevalence of background noise. The second preset threshold is used to determine whether the second target similarity reaches a standard level sufficient to confirm the presence of channel noise, and can be set according to the common range of channel noise or the possible degree of signal distortion.

[0091] (3) Determine the noise type corresponding to the matching result: In some embodiments, if the first target similarity and the second target similarity are both greater than their respective preset thresholds, it means that the audio signal is affected by both environmental noise and channel noise, and the system needs to adopt environmental noise reduction and channel repair strategies separately or simultaneously; if only one target similarity is greater than the threshold, the corresponding noise type processing is performed according to this condition; if both are lower than the threshold, it is preliminarily determined that the signal quality is good and no noise reduction processing is required.

[0092] Step S208: using a first signal noise reduction strategy to reduce the environmental noise in the audio signal to be identified, wherein the first signal noise reduction strategy is used to eliminate the environmental noise.

[0093] In the above step S208, the first signal noise reduction strategy includes a noise reduction strategy specifically for environmental noise, for example, through a microphone array and digital signal processing technology (such as low-pass filtering, wavelet decomposition, singular value decomposition) to isolate and reduce the impact of environmental noise. The first signal noise reduction strategy is executed at the front end of signal processing, for example, before PCM decoding, to ensure that the voice received by the user is as clear as possible.

[0094] In order to effectively distinguish between speech signals and noise signals and ensure the stability and reliability of the system in a noisy environment, the environmental noise can be reduced by dynamically adjusting the noise threshold. Specifically: the audio signal to be recognized is divided into speech segments and non-speech segments; for the speech segments in the audio signal to be recognized, the noise threshold of the filter is adjusted by a third preset threshold, wherein the third preset threshold is greater than the noise threshold; for the non-speech segments in the audio signal to be recognized, the noise threshold of the filter is adjusted by a fourth preset threshold, wherein the fourth preset threshold is less than the noise threshold; and the filter after adjusting the noise threshold is used to filter the audio signal to be recognized.

[0095] (1) Cutting of the audio signal to be identified: A speech segment refers to a segment in the audio signal that contains human speech sounds, and a non-speech segment refers to a segment in the audio signal that does not contain human voices, which may include background noise, silence, or short non-human sound sources such as traffic noise, wind noise, etc. In some embodiments, a voice activity detection algorithm may be used to cut the audio signal to be identified into continuous speech segments and non-speech segments based on the spectrum and time domain characteristics of the audio signal.

[0096] (2) Speech segment noise threshold adjustment: The third preset threshold refers to the value used to adjust the filter noise threshold in the speech segment, the purpose of which is to effectively suppress background noise while ensuring the quality of the human voice. In some embodiments, for the detected speech segment, the system uses the third preset threshold to adjust the noise threshold of the filter. For example, the noise threshold can be dynamically adjusted according to the third preset threshold and the characteristics of the speech segment through a predefined function or algorithm.

[0097] (3) Non-speech segment noise threshold adjustment: The fourth preset threshold refers to the value used to adjust the noise threshold of the filter in the non-speech segment, with the purpose of adopting a stricter noise reduction strategy in the silent or background noise segment. In some embodiments, for the non-speech segment, the system uses the fourth preset threshold to adjust the noise threshold of the filter. For example, the noise threshold can be dynamically adjusted according to the fourth preset threshold and the characteristics of the non-speech segment through a predefined function or algorithm.

[0098] (4) Filtering operation: After adjusting the noise threshold, the filter performs a filtering operation on the entire audio signal to be recognized, for example, applying corresponding noise reduction processing to the speech segment and the non-speech segment. The filtering may include adaptive filtering, spectral subtraction or wavelet noise reduction, etc.

[0099] Step S210: using a second signal noise reduction strategy to reduce the channel noise in the audio signal to be identified, wherein the second signal noise reduction strategy is used to repair the signal distortion caused by the channel noise.

[0100] In the above step S210, the second signal noise reduction strategy includes a noise reduction strategy for channel noise, for example, using a deep learning model (such as BERT) to analyze and reconstruct the signal after channel decoding to repair the signal distortion caused by channel noise.

[0101] Through the combination of compressed sensing algorithm and channel model, the signal distortion and attenuation caused by channel noise can be effectively repaired, specifically: the sparsity of the audio signal to be identified in the mathematical transform domain is identified, wherein the mathematical transform domain includes mathematical transform domains other than the time domain; when the audio signal to be identified is sparse in the mathematical transform domain, a sparse representation of the audio signal to be identified in the mathematical transform domain is constructed; the target observation value is extracted from the sparse representation to obtain the observation vector, wherein the target observation value is used to reconstruct the audio signal to be identified; based on the observation vector, the channel model is estimated by using the compressed sensing algorithm, wherein the channel model is used to describe the distortion and attenuation of the audio signal to be identified; the channel model is used to reconstruct the audio signal to be identified.

[0102] (1) Identify the sparsity of the audio signal to be identified in the mathematical transform domain: Sparsity means that in the mathematical transform domain, if the signal can be represented as a linear combination of a small number of non-zero coefficients and a basis, then the signal is said to be sparse. The mathematical transform domain refers to a space other than the time domain that can transform the signal from its original representation to another mathematical representation, such as the frequency domain, wavelet domain, MFCC domain, etc.

[0103] In some embodiments, the audio signal to be identified can be transformed from the time domain to other mathematical transform domains (such as the frequency domain or the wavelet domain), and its sparsity in these domains can be analyzed, for example, by calculating the coefficient distribution of the signal in the transform domain to observe whether there are a large number of coefficients with zero values ​​or close to zero values.

[0104] (2) Constructing a sparse representation of the audio signal to be identified in the mathematical transform domain: Sparse representation means that the signal can be represented as a vector or matrix containing mostly zero values, while a small number of non-zero values ​​carry the main information of the signal. In some embodiments, the signal can be represented as a sparse vector or matrix by thresholding, regularization compression (such as L1 regularization), and other methods.

[0105] (3) Extracting target observations from the sparse representation to obtain an observation vector: The observation vector refers to a vector consisting of target observations extracted from the sparse representation. The target observations include signal features measured in a mathematical transformation domain and are used in subsequent channel model estimation and signal reconstruction processes. In some embodiments, the target observations may be extracted by performing a linear transformation on the sparse representation through a measurement matrix.

[0106] (4) Use compressed sensing algorithm to estimate the channel model: The compressed sensing algorithm is a mathematical method for accurately reconstructing signals from a small number of observations. It can use the sparsity of the signal to solve the sparse representation of the signal through optimization problems to achieve accurate reconstruction of the signal. The channel model refers to a mathematical model that describes the distortion and attenuation characteristics of the audio signal to be identified during the transmission process. The characteristics may include amplitude attenuation, phase rotation, frequency offset, and delay.

[0107] Estimating the channel model using a compressed sensing algorithm involves solving an optimization problem, the goal of which is to recover a sparse representation of the signal from the observation vector under the constraint of channel noise. In some embodiments, the estimation of the channel model depends on signal characteristics, channel characteristics, and noise models, and can be achieved by regularized least squares, basis pursuit algorithm, and the like.

[0108] (5) Reconstructing the audio signal to be recognized using the channel model: In some embodiments, the amplitude, phase, and spectrum characteristics of the signal can be adjusted according to the channel model to compensate for the distortion and attenuation generated during the transmission process, thereby obtaining an audio signal that is closer to the original signal. For example, this can be achieved by using a reverse transmission algorithm, linear predictive coding (LPC), or other adaptive signal processing techniques.

[0109] By repairing signal distortion and compensating for signal attenuation through inverse channel compensation, the degree of signal restoration can be significantly improved. Specifically: the audio signal to be identified is converted into a sparse representation in a mathematical transform domain; the original audio signal corresponding to the channel noise in the audio to be identified is restored using an observation vector and a channel model to obtain a first audio signal; the first audio signal is inversely channel compensated to obtain a second audio signal, wherein the inverse channel compensation includes adjusting the spectrum amplitude; the second audio signal is converted into a time domain representation to obtain a reconstructed audio signal, wherein the reconstructed audio signal is used for an audio output device of a cell device to generate sound waves.

[0110] 1) Converting the audio signal to be identified into a sparse representation in a mathematical transformation domain: The mathematical transformation domain includes a mathematical transformation domain other than the time domain that has been identified as having sparsity during the channel model estimation process. In some embodiments, after receiving the audio signal to be identified, the audio signal to be identified can be converted from the time domain to the frequency domain or another mathematical transformation domain by a preset algorithm, for example, discrete Fourier transform (DFT), wavelet transform, MFCC transform, etc.

[0111] 2) Using the observation vector and the channel model to restore the original audio signal corresponding to the channel noise in the audio signal to be identified, and obtain a first audio signal: the first audio signal refers to the audio signal restored from the observation vector by estimating the channel noise. In some embodiments, the observation vector and the channel model can be used to restore the original audio signal corresponding to the channel noise by solving an optimization problem, such as minimizing the L1 norm of the sparse representation while satisfying constraints such as the observation vector.

[0112] 3) Performing inverse channel compensation on the first audio signal to obtain a second audio signal: The second audio signal refers to an audio signal obtained after performing inverse channel compensation on the first audio signal. In some embodiments, the spectrum amplitude of the first audio signal can be adjusted according to the frequency response characteristics described in the channel model to inversely compensate for the change in signal amplitude caused by channel noise, for example, by constructing a compensation filter whose frequency response is opposite to the frequency response of the channel model.

[0113] In an optional embodiment, the phase of the signal may also be compensated to eliminate the phase distortion caused by the channel noise. For example, this may be achieved by using a filter whose phase response is opposite to the channel phase response.

[0114] 4) Convert the second audio signal into a time domain representation to obtain a reconstructed audio signal: The reconstructed audio signal refers to an audio signal after inverse channel compensation and mathematical transformation domain return to the time domain, which is close to the original transmitted audio signal and is used for the audio output of the cell device. In some embodiments, the base station can restore the signal to its original time domain representation through methods such as inverse discrete Fourier transform (IDFT), inverse wavelet transform or inverse MFCC conversion, so that the audio output device (such as a speaker) of the cell device (such as a mobile phone) can produce clear sound waves.

[0115] In order to facilitate understanding of the above process of reducing channel noise, a detailed explanation is given below in conjunction with a specific embodiment:

[0116] 1. Identify sparsity: that is, determine the sparsity of the signal K in a certain mathematical transformation domain (such as frequency domain or wavelet domain).

[0117] 2. Construct sparse representation: Use the basis matrix Ψ to sparsely represent the signal K, that is:

[0118] K=ΨΘ

[0119] Among them, K is the audio signal to be identified, Ψ is the basis matrix (corresponding to the orthogonal basis vectors in the mathematical transform domain), and Θ is the projection coefficient vector (corresponding to the sparse representation of the audio signal to be identified in the mathematical transform domain).

[0120] 3. Extract target observations: Extract M observations from the sparse representation through the measurement matrix Φ to obtain the observation vector Z, that is:

[0121] Z=ΦK=ΦΨΘ

[0122] Where Z is the observation vector (corresponding to the target observation extracted from the sparse representation) and Φ is the measurement matrix (used to extract the observation from the signal).

[0123] 4. Estimation of channel model: In some embodiments, when a compressed sensing algorithm is used for channel estimation, the observation vector may be a received signal Y of a base station, and the measurement matrix may be a signal X transmitted by a cell. The compressed sensing algorithm is used to estimate the channel model using the observation vector and the measurement matrix, that is:

[0124] Y=XH

[0125] Among them, Y is the base station receiving signal (corresponding to the observation vector), X is the cell sending signal (corresponding to the measurement matrix), and H is the channel model (used to describe the distortion and attenuation of the audio signal to be identified).

[0126] 5. Signal reconstruction: Use the estimated channel model H to reconstruct the signal K.

[0127] Step S212: taking the denoised audio signal to be recognized as a target audio signal.

[0128] In the above step S212, the target audio signal refers to the audio signal after the noise reduction processing. After the target audio signal is converted from analog to digital (converted from digital signal back to analog signal), it can be played through the audio output device of the terminal.

[0129] Through the above steps S202 to S212, by collecting the audio signal after the satellite communication signal is converted, identifying the noise type in the audio signal, and processing the corresponding noise respectively, the audio signal after noise reduction is obtained, thereby achieving the purpose of significantly reducing noise interference, thereby achieving the technical effect of improving voice clarity and improving communication quality, and further solving the technical problem that the related technology focuses on suppressing environmental noise, ignores the processing of channel noise, and the noise reduction effect is limited.

[0130] Figure 4 is an overall framework diagram of a signal noise reduction method according to an embodiment of the present application, such as Figure 4 As shown, in some embodiments, the process of signal noise reduction is as follows:

[0131] S402: Antenna signal reception, that is, the terminal antenna receives a wireless communication signal from a satellite. In some embodiments, the communication signal may be a modulated signal carrying voice information.

[0132] S404: Signal processing, that is, the received signal first needs to go through channel signal processing, which may include demodulation, amplification and filtering, to convert the received RF signal into a digital baseband signal for subsequent signal decoding and analysis. The purpose of signal processing is to optimize signal quality and provide the clearest possible signal input for the decoding stage.

[0133] S406: Channel decoding, that is, after the signal is initially processed, channel decoding is performed. For example, the decoder recovers the original data stream of the signal according to the channel coding scheme, such as forward error correction (FEC) or Turbo decoding, which can detect and correct errors introduced during the transmission process.

[0134] S408: Language decoding, that is, converting the digital signal into a voice signal, for example, using PCM pulse code modulation and other technologies. The purpose of language decoding is to convert the data stream into an audible audio signal.

[0135] S410: PCM decoding, that is, converting the digital voice signal into an analog voice signal. For example, the decoder interprets the binary data stream into an analog voltage signal (the basis for subsequent audio playback).

[0136] S412: Generate analog audio. After PCM decoding, the signal is converted into analog audio. The analog audio signal is sent to the audio output device of the terminal (such as a receiver) and converted into sound waves through the audio output device.

[0137] S414: An earpiece is used to convert analog audio signals into sound waves in the air.

[0138] S416: Collecting environmental sounds, that is, while the earpiece is playing sound waves, the microphone of the terminal device will capture the sound produced by the earpiece and the sound of the surrounding environment, such as noise or noise, which needs to be processed in subsequent steps.

[0139] S418: The MIC (microphone) collects sound wave information from the earpiece and the environment in real time, providing data input for subsequent noise identification and elimination. The collected sound wave information will be used for the identification and processing of environmental noise, as well as the comparison between the terminal and the standard voice model.

[0140] S420: Intelligent operation and analysis, i.e., identifying environmental noise and channel noise. After the noise is identified, a corresponding noise reduction algorithm is applied in the channel decoding or PCM decoding stage according to the noise type. For example, a speech model is used to repair the distorted signal of the channel noise, or low-pass filtering and other technologies are used to eliminate environmental noise.

[0141] In some embodiments, if channel noise is detected, channel noise compensation can be performed according to the BERT model, that is, the signal is processed in the S406 channel decoding stage to repair the distorted signal through the speech model; if environmental noise is detected, the audio signal is processed in the PCM decoding stage of S410 to eliminate the environmental noise through an intelligent algorithm.

[0142] Through hierarchical noise reduction technology, combined with intelligent speech recognition models and big data models, the impact of channel noise caused by long-distance signal transmission in satellite communications and environmental noise around the terminal on voice quality is overcome, and the signal transmission quality and call clarity of direct-connected satellite terminals in complex environments are improved, thereby achieving a clearer and more accurate voice communication experience.

[0143] Figure 5is a schematic diagram of the principle of environmental noise reduction according to a signal noise reduction method of an embodiment of the present application. In some embodiments, environmental noise can be reduced by the following steps, such as: Figure 5 As shown:

[0144] S502: Pick up feature segments through a microphone (MIC) and match them with data in the model.

[0145] The microphone on the terminal device collects audio information in the environment, converts sound waves into electrical signals and performs digital processing to form a digital audio signal. Subsequently, the audio signal is framed and feature extracted. The extracted features may include Mel Frequency Cepstral Coefficient (MFCC), Zero Crossing Rate (ZCR), energy, etc. These features are used to describe the characteristics of each frame signal. After feature extraction, the feature vector of each frame is matched with the data in the pre-trained acoustic model and language model. Through comparison and analysis, the model can determine whether there is environmental noise in the signal.

[0146] S510: Model training.

[0147] Model training includes two parts: acoustic model training and language model training. Acoustic models, such as BERT-based models, are trained with a large number of speech samples from a speech database to learn the time-frequency characteristics of speech signals and how to identify clear speech signals in a noisy environment. Language models are trained with a language database to learn the vocabulary, grammar, syntax, and semantic rules of a language so that they can understand the meaning when recognizing speech signals. The trained model can identify and distinguish environmental noise from channel noise, providing support for subsequent intelligent noise reduction.

[0148] S504: Endpoint detection.

[0149] The first step in front-end preprocessing is endpoint detection, which aims to identify the start and end points of speech in an audio signal. By analyzing features such as the short-term energy and zero-crossing rate of the signal, the endpoint detection algorithm can accurately determine which parts contain valid speech signals, thereby ignoring silent or non-speech segments and reducing the computational burden of subsequent processing.

[0150] S506: Noise reduction.

[0151] After identifying the valid speech segment, the speech signal is subjected to noise reduction processing. For example, by using algorithms such as low-pass filtering, wavelet transform, singular value decomposition, etc., the noise components in the signal are removed and clear speech is retained. The selection of noise reduction algorithms and parameter adjustment need to consider the noise type and intensity to ensure the best noise reduction effect.

[0152] S508: Feature extraction.

[0153] Feature extraction is to convert the denoised speech signal into a set of numerical features for subsequent model recognition and processing. For example, MFCC or other feature extraction methods are used to ensure that the extracted features can accurately reflect the spectral and time domain characteristics of the speech signal and provide an effective input representation for the model.

[0154] S512: Backend identification processing.

[0155] After front-end preprocessing and feature extraction, the denoised and feature-extracted audio data will be passed to the back-end recognition processing. At this stage, the model will use the acoustic model and language model trained in S510 to analyze and recognize the passed feature vectors to determine the speech content. For example, the feature vectors can be processed using a multi-head self-attention mechanism to capture the semantic information in the signal and ultimately generate speech recognition results in text form. In some embodiments, adaptive adjustments can also be used to optimize recognition performance, taking into account changes in different environments and noise conditions.

[0156] Finally, the result output is the speech text after recognition processing, for example, a clear speech signal after noise reduction. The entire process from audio signal acquisition, feature extraction and model matching to noise reduction, speech recognition and result output is to ensure that the terminal device can provide high-quality voice communication experience in various complex environments. By combining front-end preprocessing and back-end recognition processing, and using well-trained acoustic models and language models, effective recognition and suppression of environmental noise is achieved, significantly improving the clarity and comprehensibility of speech signals.

[0157] Figure 6 is a schematic diagram of the principle of reducing environmental noise and channel noise according to a signal noise reduction method of an embodiment of the present application. In some embodiments, environmental noise and channel noise can be reduced by the following steps, such as Figure 6 As shown:

[0158] Since steps S602, S604 and S606 are Figure 5 The steps S502, S504, S506, S508 and S510 correspond to each other, so they are not repeated here.

[0159] S608: Channel signal processing.

[0160] When the model identifies channel noise, the signal will enter the channel signal processing stage, which aims to repair the signal distortion introduced by the channel characteristics. The signal processing steps include:

[0161] 1) Demodulation: The received RF signal is converted into a baseband signal through demodulation technology, preparing for subsequent channel decoding.

[0162] 2) Channel decoding: Use coding and decoding techniques, such as forward error correction (FEC) or Turbo decoding, to restore the original data stream of the signal and correct errors that may be introduced during the transmission process.

[0163] 3) Source decoding: Convert the decoded digital signal into voice information in preparation for voice decoding.

[0164] S610: Backend identification processing.

[0165] After the front-end preprocessing or channel signal processing, the audio signal or feature will be passed to the back-end recognition processing, and the acoustic model and language model in S606 can be used to analyze and recognize the signal to restore clear speech. Specifically:

[0166] 1) Speech decoding: Based on the model training results, the denoised signal is decoded and converted into speech text or audio output, including feature analysis, speech reconstruction, and compensation of distorted signals.

[0167] 2) Adaptive adjustment: Depending on the specific conditions of the signal and the type of noise, the model can be adaptively adjusted to optimize the decoding process and ensure high-quality speech output under different conditions.

[0168] The above process starts with the MIC receiving the signal, identifying the noise type through the model, intelligently selecting the front-end preprocessing or channel signal processing, and finally decoding and outputting the voice through the model, achieving dual noise reduction of environmental noise and channel noise, and significantly improving the quality of voice communication.

[0169] Figure 7 is a structural diagram of a signal noise reduction device according to an embodiment of the present application, such as Figure 7 As shown, the device comprises:

[0170] The receiving module 702 is used to receive the analog audio signal after channel decoding, wherein the analog audio signal is used for the audio output device to generate sound waves;

[0171] The acquisition module 704 is used to acquire the sound wave information generated by the analog audio signal to obtain the audio signal to be identified, wherein the audio signal to be identified includes the voice information after channel decoding and the external environment sound information;

[0172] An identification module 706, configured to identify a noise type in the audio signal to be identified, wherein the noise type includes environmental noise and channel noise;

[0173] An execution module 708 is configured to reduce the ambient noise in the audio signal to be identified by adopting a first signal noise reduction strategy, wherein the first signal noise reduction strategy is used to eliminate the ambient noise; and reduce the channel noise in the audio signal to be identified by adopting a second signal noise reduction strategy, wherein the second signal noise reduction strategy is used to repair the signal distortion caused by the channel noise;

[0174] The determination module 710 is configured to use the denoised audio signal to be identified as a target audio signal.

[0175] In the above-mentioned signal noise reduction device, the recognition module is also used to construct a first feature vector of the audio signal, wherein the first feature vector includes at least one of the following: spectral features, time domain features, and channel state information features; matching the first feature vector with a noise feature representation to obtain a matching result, wherein the noise feature representation is generated by a noise feature extraction model; and determining the noise type corresponding to the matching result.

[0176] In the above-mentioned signal denoising device, the recognition module is also used to calculate a first similarity between a first feature vector and a first noise feature representation, wherein the first noise feature representation includes any one noise feature representation in a set of environmental noise feature representations; determine a first target similarity in a first similarity set where the first similarity is located, wherein the first target similarity includes a maximum value in the first similarity set; calculate a second similarity between the first feature vector and a second noise feature representation, wherein the second noise feature representation includes any one noise feature representation in a set of channel noise feature representations; determine a second target similarity in a second similarity set where the second similarity is located, wherein the second target similarity includes a maximum value in the second similarity set; and determine a matching result using the first target similarity and the second target similarity.

[0177] In the above-mentioned signal noise reduction device, the recognition module is also used to compare the first target similarity with the first preset threshold, and when the first similarity is greater than the first preset threshold, determine that there is environmental noise in the audio signal; compare the second target similarity with the second preset threshold, and when the second similarity is greater than the second preset threshold, determine that there is channel noise in the audio signal.

[0178] In the above-mentioned signal noise reduction device, the execution module is also used to cut the audio signal to be recognized into speech segments and non-speech segments; for the speech segments in the audio signal to be recognized, the noise threshold of the filter is adjusted using a third preset threshold, wherein the third preset threshold is greater than the noise threshold; for the non-speech segments in the audio signal to be recognized, the noise threshold of the filter is adjusted using a fourth preset threshold, wherein the fourth preset threshold is less than the noise threshold; and the filter after adjusting the noise threshold is used to perform filtering operations on the audio signal to be recognized.

[0179] In the above-mentioned signal denoising device, the execution module is also used to identify the sparsity of the audio signal to be identified in the mathematical transform domain, wherein the mathematical transform domain includes mathematical transform domains other than the time domain; when the audio signal to be identified is sparse in the mathematical transform domain, a sparse representation of the audio signal to be identified in the mathematical transform domain is constructed; the target observation value is extracted from the sparse representation to obtain an observation vector, wherein the target observation value is used to reconstruct the audio signal to be identified; based on the observation vector, a compressed sensing algorithm is used to estimate a channel model, wherein the channel model is used to describe the distortion and attenuation of the audio signal to be identified; and the channel model is used to reconstruct the audio signal to be identified.

[0180] In the above-mentioned signal denoising device, the execution module is also used to convert the audio signal to be identified into a sparse representation in a mathematical transform domain; use the observation vector and the channel model to restore the original audio signal corresponding to the channel noise in the audio to be identified to obtain a first audio signal; perform inverse channel compensation on the first audio signal to obtain a second audio signal, wherein the inverse channel compensation includes adjusting the spectrum amplitude; convert the second audio signal into a time domain representation to obtain a reconstructed audio signal, wherein the reconstructed audio signal is used for the audio output device of the cell device to generate sound waves.

[0181] It should be noted that Figure 7 The signal noise reduction device shown is used to perform Figure 2 The signal noise reduction method shown is therefore Figure 2 The explanations in the signal noise reduction method in also apply to Figure 7 The signal noise reduction device shown will not be described in detail here.

[0182] An embodiment of the present application further provides an electronic device, which includes a memory and a processor, wherein the memory is used to store program instructions; the processor is connected to the memory and is used to execute the steps of the signal noise reduction method in each embodiment of the present application.

[0183] For example, the processor performs the following functions by executing program instructions stored in the memory:

[0184] Receive an analog audio signal after channel decoding, wherein the analog audio signal is used for an audio output device to generate sound waves; collect sound wave information generated by the analog audio signal to obtain an audio signal to be identified, wherein the audio signal to be identified includes voice information after channel decoding and external environmental sound information; identify the type of noise in the audio signal to be identified, wherein the noise type includes environmental noise and channel noise; use a first signal noise reduction strategy to reduce the environmental noise in the audio signal to be identified, wherein the first signal noise reduction strategy is used to eliminate the environmental noise; use a second signal noise reduction strategy to reduce the channel noise in the audio signal to be identified, wherein the second signal noise reduction strategy is used to repair the signal distortion caused by the channel noise; and use the denoised audio signal to be identified as the target audio signal.

[0185] An embodiment of the present application further provides a non-volatile storage medium, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the steps of the signal noise reduction method in each embodiment of the present application by running the computer program.

[0186] The embodiment of the present application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the signal noise reduction method in each embodiment of the present application.

[0187] The embodiments of the present application also provide a computer program, which, when executed by a processor, implements the steps of the signal noise reduction method in each embodiment of the present application.

[0188] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0189] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0190] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0191] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0192] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0193] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0194] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for signal noise reduction, characterized in that: include: Receiving an analog audio signal after channel decoding, wherein the analog audio signal is used by an audio output device to generate sound waves; Acquiring sound wave information generated by the analog audio signal to obtain an audio signal to be identified, wherein the audio signal to be identified includes voice information after channel decoding and external environmental sound information; Identifying a noise type in the audio signal to be identified, wherein the noise type includes environmental noise and channel noise; Using a first signal noise reduction strategy to reduce the environmental noise in the audio signal to be identified, wherein the first signal noise reduction strategy is used to eliminate the environmental noise; Using a second signal noise reduction strategy to reduce the channel noise in the audio signal to be identified, wherein the second signal noise reduction strategy is used to repair signal distortion caused by the channel noise; The to-be-recognized audio signal after noise reduction is used as a target audio signal.

2. The method according to claim 1, characterized in that The identifying the noise type in the audio signal to be identified includes: Constructing a first feature vector of the audio signal, wherein the first feature vector includes at least one of the following: a spectrum feature, a time domain feature, and a channel state information feature; Matching the first feature vector with a noise feature representation to obtain a matching result, wherein the noise feature representation is generated by a noise feature extraction model; Determine the noise type corresponding to the matching result.

3. The method according to claim 2, characterized in that The matching of the first feature vector with the noise feature representation to obtain a matching result includes: Calculating a first similarity between the first feature vector and a first noise feature representation, wherein the first noise feature representation includes any one noise feature representation in a set of environmental noise feature representations; determining a first target similarity in a first similarity set where the first similarity is located, wherein the first target similarity includes a maximum value in the first similarity set; Calculating a second similarity between the first feature vector and a second noise feature representation, wherein the second noise feature representation includes any one noise feature representation in a channel noise feature representation set; determining a second target similarity in a second similarity set where the second similarity is located, wherein the second target similarity includes a maximum value in the second similarity set; A matching result is determined using the first target similarity and the second target similarity.

4. The method according to claim 3, characterized in that: The determining a matching result by using the first target similarity and the second target similarity includes: comparing the first target similarity with a first preset threshold, and determining that the audio signal has environmental noise when the first similarity is greater than the first preset threshold; The second target similarity is compared with a second preset threshold, and when the second similarity is greater than the second preset threshold, it is determined that channel noise exists in the audio signal.

5. The method according to claim 1, characterized in that The adopting a first signal noise reduction strategy to reduce the ambient noise in the audio signal to be identified includes: Cutting the audio signal to be recognized into speech segments and non-speech segments; For the speech segment in the audio signal to be recognized, a noise threshold of a filter is adjusted by using a third preset threshold, wherein the third preset threshold is greater than the noise threshold; For the non-speech segment in the audio signal to be recognized, a fourth preset threshold is used to adjust the noise threshold of the filter, wherein the fourth preset threshold is less than the noise threshold; The filter after adjusting the noise threshold is used to perform a filtering operation on the audio signal to be recognized.

6. The method according to claim 1, characterized in that The adopting a second signal noise reduction strategy to reduce the channel noise in the audio signal to be identified includes: Identify the sparsity of the audio signal to be identified in a mathematical transformation domain, wherein the mathematical transformation domain includes a mathematical transformation domain other than a time domain; When the audio signal to be recognized is sparse in the mathematical transformation domain, constructing a sparse representation of the audio signal to be recognized in the mathematical transformation domain; Extracting a target observation value from the sparse representation to obtain an observation vector, wherein the target observation value is used to reconstruct the audio signal to be recognized; estimating a channel model using a compressed sensing algorithm according to the observation vector, wherein the channel model is used to describe the distortion and attenuation of the audio signal to be identified; The channel model is used to reconstruct the audio signal to be recognized.

7. The method according to claim 6, characterized in that The using the channel model to reconstruct the audio signal to be recognized includes: Converting the audio signal to be recognized into a sparse representation in the mathematical transformation domain; Restoring an original audio signal corresponding to the channel noise in the audio to be recognized by using the observation vector and the channel model to obtain a first audio signal; Performing inverse channel compensation on the first audio signal to obtain a second audio signal, wherein the inverse channel compensation includes adjusting a spectrum amplitude; The second audio signal is converted into a time domain representation to obtain a reconstructed audio signal, wherein the reconstructed audio signal is used by an audio output device of a cell device to generate sound waves.

8. A signal noise reduction device, characterized in that: include: A receiving module, used for receiving an analog audio signal after channel decoding, wherein the analog audio signal is used for an audio output device to generate sound waves; A collection module, used for collecting the sound wave information generated by the analog audio signal to obtain the audio signal to be identified, wherein the audio signal to be identified includes the voice information after channel decoding and the external environment sound information; An identification module, used to identify the noise type in the audio signal to be identified, wherein the noise type includes environmental noise and channel noise; an execution module, configured to adopt a first signal noise reduction strategy to reduce the environmental noise in the audio signal to be identified, wherein the first signal noise reduction strategy is used to eliminate the environmental noise; and adopt a second signal noise reduction strategy to reduce the channel noise in the audio signal to be identified, wherein the second signal noise reduction strategy is used to repair the signal distortion caused by the channel noise; The determination module is used to use the to-be-recognized audio signal after noise reduction as a target audio signal.

9. An electronic device, characterized in that: include: A memory and a processor, wherein the memory is used to store program instructions; the processor is connected to the memory and is used to execute the signal noise reduction method described in any one of claims 1 to 7.

10. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the signal noise reduction method according to any one of claims 1 to 7 by running the computer program.

Images