Satellite communication voice compression method and device and computer equipment

Abstract

The application discloses a satellite communication voice compression method and device and a computer device. The method comprises the following steps: sampling a voice signal according to a preset sampling frequency to obtain sample voice data, wherein the preset sampling frequency is determined according to a channel state of the voice signal; and compressing the to-be-compressed voice data by using a target neural network model to obtain target voice data, wherein the target neural network model is obtained by training in the following manner: taking the sample voice data and the channel state when the sample voice data is sampled as input, taking compressed voice data obtained by compressing the sample voice data by using a compression algorithm as output, and training the target neural network model. The application solves the technical problem of low bandwidth resource utilization caused by occupying too much bandwidth resource to ensure voice quality in the prior art.

Description

Technical Field

[0001] The present application relates to the field of satellite communications, and in particular to a method, apparatus and computer equipment for compressing satellite communication voice. Background Art

[0002] The existing satellite communication voice compression algorithms, in order to ensure that the bandwidth required for voice transmission is reduced, need to sacrifice voice quality when compressing the voice to be transmitted by satellite communication. If you want to ensure higher voice quality, you need to occupy more bandwidth resources. Summary of the Invention

[0003] The embodiments of the present application provide a satellite communication voice compression method, apparatus, and computer device to at least solve the technical problem in the related art of reduced bandwidth resource utilization due to excessive bandwidth resources occupied to ensure voice quality.

[0004] According to one aspect of an embodiment of the present application, a method for compressing satellite communication voice is provided, comprising: sampling a voice signal at a preset sampling frequency to obtain sample voice data, wherein the preset sampling frequency is determined based on the channel state for transmitting the voice signal; compressing the voice data to be compressed using a target neural network model to obtain target voice data, wherein the target neural network model is trained in the following manner: taking sample voice data and the channel state when the sample voice data is sampled as input, and taking compressed voice data obtained by compressing the sample voice data using a compression algorithm as output, and training the target neural network model.

[0005] Optionally, the voice signal is sampled according to a preset sampling frequency to obtain sample voice data, including: obtaining the channel state for transmitting the voice signal, the channel state including: the bandwidth and signal-to-noise ratio of the channel; determining the channel state for transmitting the voice signal based on the bandwidth and the signal-to-noise ratio of the channel; when the channel state meets the preset conditions, the voice signal is sampled using a first preset frequency, and when the channel state does not meet the preset conditions, the voice signal is sampled using a second preset frequency, the first preset frequency being higher than the second preset frequency.

[0006] Optionally, the sample voice data is compressed using a compression algorithm, including: pre-emphasis processing on the sample voice data; dividing the pre-emphasized sample voice data into multiple frames of fixed length, wherein the number of iterations required for the target neural network model is equal to the number of voice data samples contained in each fixed-length frame; and compressing the multiple fixed-length frames to obtain the compressed voice data.

[0007] Optionally, the target neural network is obtained in the following manner, including: determining the sample speech data, the channel state when sampling the sample speech data, and the compressed speech data as a training data set, and randomly selecting a portion of data from the training data set to pre-train the initial neural network model to obtain initial model parameters; after setting the initial neural network model according to the initial model parameters, inputting multiple fixed-length frames corresponding to the sample speech data into the initial neural network model respectively, performing forward propagation, and obtaining a model output value; determining a loss function value based on a comparison between the model output value and the compressed speech data, and adjusting the model parameters of the initial neural network model based on the loss function value until the loss function converges to obtain the target neural network model.

[0008] Optionally, the model parameters of the initial neural network model are adjusted according to the loss function value, including: obtaining the gradient of each model parameter in the initial neural network model with respect to the loss function; and completing the update of each model parameter according to the gradient of each model parameter with respect to the loss function.

[0009] Optionally, performing pre-emphasis processing on the sample voice data includes: obtaining a voice signal at each time point in the sample voice data; and performing first-order high-pass filtering on the voice signal at each time point to complete the pre-emphasis processing.

[0010] Optionally, the method also includes: deploying an initial neural network model in a communication satellite; sending model parameters corresponding to the target neural network model to the communication satellite to adjust the initial neural network model according to the model parameters to obtain the target neural network model; receiving the compression result of the communication satellite using the target neural network model on the voice data to be compressed, and updating the target neural network model according to the compression result.

[0011] According to another aspect of an embodiment of the present application, a satellite communication voice compression device is also provided, including: an acquisition module, used to sample a voice signal according to a preset sampling frequency to obtain sample voice data, wherein the preset sampling frequency is determined according to the channel state of transmitting the voice signal; a compression module, used to compress the voice data to be compressed using a target neural network model to obtain target voice data, wherein the target neural network model is trained in the following manner: taking sample voice data and the channel state when sampling the sample voice data as input, and taking compressed voice data obtained by compressing the sample voice data using a compression algorithm as output, and training the target neural network model.

[0012] According to another aspect of the embodiment of the present application, a computer device is also provided, including: 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 above-mentioned satellite communication voice compression method.

[0013] 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 satellite communication voice compression method by running the computer program.

[0014] According to another aspect of the embodiments of the present application, a computer program product is provided, comprising computer instructions, which implement the above-mentioned satellite communication voice compression method when executed by a processor.

[0015] In an embodiment of the present application, sample voice data is obtained by sampling a voice signal according to a preset sampling frequency, wherein the preset sampling frequency is determined according to the channel state of transmitting the voice signal; and a target neural network model is used to compress the voice data to obtain target voice data, wherein the target neural network model is trained in the following manner: taking the sample voice data and the channel state when the sample voice data is sampled as input, and taking the compressed voice data obtained by compressing the sample voice data using a compression algorithm as output, and training the target neural network model, thereby achieving the purpose of compressing the voice data using a neural network model, and then ensuring the voice quality while improving the compression rate, thereby achieving the technical effect of improving the bandwidth utilization during voice transmission, and thus solving the technical problem in the related art that the bandwidth resource utilization is reduced due to excessive bandwidth resources occupied in order to ensure voice quality. 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 This is a hardware structure block diagram of a computer terminal for implementing a satellite communication voice compression method according to an embodiment of the present application;

[0018] Figure 2 is a flow chart of a satellite communication voice compression method according to an embodiment of the present application;

[0019] Figure 3 is a schematic diagram of a target neural network architecture according to an embodiment of the present application;

[0020] Figure 4is a schematic diagram of a target neural network training process according to an embodiment of the present application;

[0021] Figure 5 is a flowchart of another satellite communication voice compression method according to an embodiment of the present application;

[0022] Figure 6 This is a structural diagram of a satellite communication voice compression device according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0024] 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 sequential order. 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 a sequence 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.

[0025] The information collected in the embodiments of the present application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or reject the automated decision results; if the user chooses to reject, the expert decision-making process will be entered.

[0026] Currently available satellite communication voice compression algorithms include but are not limited to the following, each with its own advantages and limitations:

[0027] G.729 / G.729a: This is a speech coding algorithm recommended by the ITU-T and widely used in voice communications. It uses a lossy compression algorithm with a low bit rate (8 kbit / s). However, G.729 has a trade-off between compression rate and voice quality, and some high-frequency signals may be lost.

[0028] G.723.1: This is a dual-rate speech coding algorithm developed by the ITU-T, supporting compression ratios of 5.3 kbit / s and 6.3 kbit / s. G.723.1 provides better speech quality at lower bit rates, but at the expense of speech quality.

[0029] AMR (Adaptive Multi-Rate): AMR is an adaptive multi-rate speech coding algorithm commonly used in mobile communication systems. It dynamically adjusts the coding rate, ranging from 4.75 kbit / s to 12.2 kbit / s, based on the quality of the communication link and available bandwidth. While providing good voice quality, AMR also achieves high bandwidth utilization, albeit at the expense of speech quality.

[0030] Opus: Opus is an open, royalty-free audio codec developed by the IETF as RFC 6716. Opus supports variable rate coding and can transmit at very low bit rates but at the expense of voice quality.

[0031] LPC (Linear Predictive Coding): A commonly used speech compression algorithm, also commonly used in satellite communications. It achieves compression through predictive modeling of speech signals and error coding. However, this algorithm has high computational complexity.

[0032] As mentioned above, in the related art, there are some compression algorithms to improve bandwidth utilization. However, these algorithms improve bandwidth utilization at the expense of voice quality.

[0033] In order to solve the problems existing in the related art, the embodiment of the present application provides a satellite communication voice compression method, which can be run on Figure 1 In the computer terminal shown, the computer terminal is explained below.

[0034] The satellite communication voice compression method provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 The hardware structure block diagram of a computer terminal for implementing a satellite communication voice compression method is shown in FIG. Figure 1As shown, the computer terminal 10 may include one or more (illustrated by 102a, 102b, ..., 102n in the figure) 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. It will be understood by those skilled in the art 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 than shown, or with Figure 1 Different configurations shown.

[0035] 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 circuitry." The data processing circuitry may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuitry may be a single, independent processing module, or may be incorporated in whole or in part into any of the other components of the computer terminal 10. As described in the embodiments of the present application, the data processing circuitry serves as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0036] Memory 104 can be used to store software programs and modules for application software, such as the program instructions / data storage device corresponding to the satellite communication voice compression method in the embodiments of the present application. The processor executes the software programs and modules stored in memory 104 to perform various functional applications and data processing, thereby implementing the aforementioned satellite communication voice compression method. Memory 104 can include high-speed random access memory (RAM) and non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, memory 104 can further include memory remotely located from the processor, which can be connected to computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0037] The transmission module 106 is configured to receive or transmit data via a network. A specific example of the aforementioned network may include a wireless network provided by the communications provider of the computer terminal 10. In one embodiment, the transmission module 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In another embodiment, the transmission module 106 may be a radio frequency (RF) module, which is configured to communicate with the Internet wirelessly.

[0038] 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 .

[0039] 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. 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.

[0040] In the above operating environment, an embodiment of the present application provides an embodiment of a method for compressing satellite communication voice. 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 a 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.

[0041] Figure 2 FIG. 1 is a flow chart of a satellite communication voice compression method according to an embodiment of the present application, such as Figure 2 As shown, the method includes the following steps:

[0042] Step S202: Sampling the voice signal at a preset sampling frequency to obtain sample voice data, wherein the preset sampling frequency is determined according to the state of the channel for transmitting the voice signal.

[0043] In step S202, a higher sampling frequency is used when the channel quality is good, and the sampling frequency is reduced to a certain extent based on the channel quality. This allows voice data to be successfully transmitted even when the channel quality is poor. A feature representing the sampling frequency of the currently input data is also input into the neural network.

[0044] Step S204: compress the speech data to be compressed using a target neural network model to obtain target speech data, wherein the target neural network model is trained in the following manner: taking sample speech data and the channel state when the sample speech data is sampled as input, and taking compressed speech data obtained by compressing the sample speech data using a compression algorithm as output, and training the target neural network model.

[0045] In the above step S204, the voice sample data is obtained by sampling the voice signal at a preset frequency.

[0046] In addition, the compression algorithm can adopt the LPC algorithm, in which the current sample is a linear combination of several past samples, and each frame of the speech signal is also correlated. The recurrent neural network can be set with two hidden layers, and the activation function selects the hyperbolic tangent function (tanh(x)).

[0047] In some embodiments of the present application, the method provided in this application can be applied to a satellite telephone communication system. When a user makes a voice call through a satellite terminal device, the voice compression algorithm of the present invention can effectively reduce the bandwidth resources required for communication, improve call quality and stability, and reduce communication costs. It can also be applied in the field of satellite broadcasting and advertising broadcasting. The method provided in this application can compress voice content and then transmit it through a satellite channel. This can effectively utilize limited satellite bandwidth resources to achieve high-quality audio broadcasting and advertising dissemination. With the development of satellite Internet access technology, more and more regions and users are accessing the Internet through satellite connections. The method provided in this application can also be used for voice communication in a satellite Internet access system to provide users with a high-quality voice call experience.

[0048] Through steps S202 to S204, the neural network model is used to compress voice data, thereby achieving the goal of maintaining voice quality while improving the compression rate. This results in the technical effect of improving bandwidth utilization during voice transmission, thereby resolving the technical problem in related technologies of reduced bandwidth utilization due to excessive bandwidth resources occupied to ensure voice quality. This is described in detail below.

[0049] In step S202 of the above-mentioned satellite communication voice compression method, the voice signal is sampled according to a preset sampling frequency to obtain sample voice data, including: obtaining a channel state for transmitting the voice signal, the channel state including: a bandwidth and a signal-to-noise ratio of the channel; determining the channel state for transmitting the voice signal based on the bandwidth and the signal-to-noise ratio of the channel; and sampling the voice signal at a first preset frequency if the channel state meets a preset condition, and sampling the voice signal at a second preset frequency if the channel state does not meet the preset condition, wherein the first preset frequency is higher than the second preset frequency.

[0050] Specifically, each sample of voice data is sampled at a first preset frequency, such as 8kHz, and further downsampled based on the current channel quality. When the channel quality is good, the original 8kHz sampling rate is used. Depending on the channel quality, the sampling frequency of the signal is reduced to a certain extent, such as using a second preset frequency. This allows voice data to be successfully transmitted even when the channel quality is poor.

[0051] It is understood that bandwidth and signal-to-noise ratio are directly proportional to channel quality.

[0052] In some embodiments of the present application, a compression algorithm is used to compress the sample voice data, including: pre-emphasis processing of the sample voice data; dividing the pre-emphasized sample voice data into multiple frames of fixed length, wherein the number of iterations required for the target neural network model is equal to the number of voice data samples contained in each fixed-length frame; and compressing the multiple fixed-length frames to obtain the compressed voice data.

[0053] The pre-emphasis processing is performed on the sample voice data, including: obtaining the voice signal at each time point in the sample voice data; and performing first-order high-pass filtering on the voice signal at each time point to complete the pre-emphasis processing.

[0054] Specifically, the process of preprocessing the sample speech data using the LPC algorithm is as follows:

[0055] Step 1: Preprocessing: Before performing LPC analysis, it is usually necessary to preprocess the speech signal, including pre-emphasis and framing operations. Pre-emphasis: In order to suppress the influence of low-frequency components on high-frequency components in the speech signal, the speech signal can be pre-emphasized, that is, the speech signal is subjected to a first-order high-pass filter. The specific formula is as follows:

[0056] x′(n)=x(n)-αx(n-1)

[0057] Where x′(n) represents the speech signal at time point n after pre-emphasis processing, x(n) is the speech signal in the sample speech data without pre-emphasis processing, and α is the pre-emphasis coefficient, which ranges from 0.9 to 1.0.

[0058] Frame processing divides the sample voice data after pre-emphasis processing into frames of fixed length. Different from the traditional time-based framing, the method provided by this application is that each frame of sample voice data contains multiple voice data samples, for example, 32 voice data samples.

[0059] Assume that each sample speech data is generated by an autoregressive process:

[0060]

[0061] In the formula, m(n) represents the current sample speech data, p represents the prediction order, and a i represents the autoregressive coefficient, and e(n) represents the prediction error, which represents the difference between the current sample and the predicted value. Once the prediction error is obtained, it can be compressed. This is typically done using techniques such as Pulse Code Modulation (PCM) or differential coding to quantize the prediction error into discrete values, which are then further compressed using entropy coding.

[0062] In some embodiments of the present application, the target neural network is obtained by the following method, including: determining the sample speech data and the channel state when sampling the sample speech data and the compressed speech data as a training data set, and randomly selecting a portion of data from the training data set to pre-train the initial neural network model to obtain initial model parameters; after setting the initial neural network model according to the initial model parameters, inputting multiple fixed-length frames corresponding to the sample speech data into the initial neural network model respectively, performing forward propagation, and obtaining a model output value; determining a loss function value based on a comparison between the model output value and the compressed speech data, and adjusting the model parameters of the initial neural network model based on the loss function value until the loss function converges to obtain the target neural network model.

[0063] Among them, the process of adjusting the model parameters of the initial neural network model according to the loss function value includes: obtaining the gradient of each model parameter in the initial neural network model with respect to the loss function; and completing the update of each model parameter according to the gradient of each model parameter with respect to the loss function.

[0064] Figure 3 A schematic diagram of a target neural network architecture is shown, such as Figure 3As shown in the figure, x is a 32-dimensional vector representing the 32 speech data samples in a frame of sample speech data. y is the probability distribution vector of the compressed speech data obtained by compressing x using the LPC algorithm, mapped by the activation function. h represents the hidden layer unit, with two layers in total. o is the unnormalized logarithmic probability output by the RNN using the activation function. L is the loss function. U is the input-to-hidden connection weight matrix. V is the hidden-to-output connection weight matrix. W is the hidden-to-hidden recurrent connection weight matrix.

[0065] The initial neural network starts with a specific initial state h(0) and propagates forward from t = 1 to each time step t = τ. We use the following equation to update the state of the initial neural network:

[0066] a t =b+Wh t-1 +Ux t

[0067] h t =tanh(a t )

[0068]

[0069] o t =c+Vh

[0070]

[0071] Where a t represents the output value of the neural network at time step t, h t Represents the hidden state at time step t, b and c are the bias vectors from input to hidden and hidden to input. The Softmax function is a mapping that maps a k-dimensional vector to a probability distribution. represents the predicted value of the neural network at time step t.

[0072] It's important to note that backpropagation updates the initial neural network model parameters using gradient descent. The nodes being calculated include U, V, W, b, and c. Gradients are calculated for each node sequentially, from back to front. For any node, the gradients of all subsequent nodes are calculated, recursively calculating the node's gradient, starting from the last loss node.

[0073] Figure 4 shows a training process of a target neural network, such as Figure 4 Shown, including:

[0074] Initializing parameters: The first step in model training is to initialize the neural network parameters. Before training begins, a small amount of data is extracted for pre-training. The parameters can be initialized to the pre-trained weights. These parameters include the weights and biases of each layer in the neural network.

[0075] Forward propagation: In each training round, the training data is forward propagated through the neural network to obtain the model output. For this speech compression task, the input is 32 samples from a frame of speech, and the output is the probability distribution vector of the compressed speech signal mapped by the Softmax function.

[0076] Calculate the loss function: Use the model output and the true label to calculate the value of the loss function. It is used to measure the difference between the model output and the actual label.

[0077] Backpropagation: The backpropagation algorithm updates the parameters of the neural network based on the gradient of the loss function. This step calculates the gradient of each parameter with respect to the loss function using the formula above, and then uses an optimization algorithm such as gradient descent to update the parameters to reduce the value of the loss function.

[0078] Parameter optimization: After backpropagation, the stochastic gradient descent algorithm is used to update the neural network parameters so that the loss function gradually decreases. During this process, the model training effect can be optimized by adjusting the learning rate, regularization, and other methods.

[0079] Iterative training: Repeat the above steps until the stopping condition is reached (such as reaching the maximum number of iterations, loss function convergence, etc.). During the training process, the loss function on the training set and validation set can be monitored to evaluate the performance of the model and make adjustments.

[0080] Model Saving: After training is complete, the trained model parameters are saved for subsequent testing on new data. Once the model passes the test, it can be sent to the communication terminal and satellite for use.

[0081] Throughout the model training process, care must be taken to avoid overfitting and underfitting. Appropriate regularization methods and data augmentation techniques can be used to improve the model's generalization capabilities. Furthermore, appropriate dataset partitioning and cross-validation are crucial factors in ensuring effective model training. The selection of various training methods should be flexibly adjusted based on the dataset to achieve optimal training results.

[0082] The satellite communication voice compression method provided in this application applies a recurrent neural network (RNN)-based voice compression algorithm to satellite communications, replacing the traditional linear predictive coding (LPC) algorithm. This algorithm can better capture the temporal information and long-term dependencies of voice signals, thereby improving the efficiency and quality of voice compression, enabling satellite communication systems to transmit higher-quality voice signals under limited bandwidth and resources. It adopts an end-to-end training approach, learning directly from the original input of the voice signal to the compressed output. Specifically, the communication terminal directly collects sample voice data and uploads it to a cloud server for model training. This simplifies the system structure, avoids the multi-stage processing required in traditional codec systems, improves system efficiency and performance, and reduces system complexity and cost. It is also important to note that the target neural network model has a certain degree of adaptability and can adapt to different voice signal characteristics and channel conditions. This makes the voice compression algorithm more versatile and applicable, capable of adapting to different compression ratios, different signal-to-noise ratios, and different voice signal types, improving the system's flexibility and robustness, thereby meeting the needs of different application scenarios.

[0083] In some embodiments of the present application, an initial neural network model can also be deployed in a communication satellite; model parameters corresponding to the target neural network model are sent to the communication satellite to adjust the initial neural network model according to the model parameters to obtain the target neural network model; the compression result of the speech data to be compressed using the target neural network model by the communication satellite is received, and the target neural network model is updated according to the compression result.

[0084] Figure 5 Another method for compressing satellite communication voice is shown. Figure 5 As shown, it includes: the communication terminal collects sample voice data and transmits it to the cloud server for model training, sends the trained model parameters to the communication terminal and the communication satellite, and compresses the voice data during the voice transmission between the communication terminal and the communication satellite.

[0085] Figure 6 FIG. 1 is a structural diagram of a satellite communication voice compression device according to an embodiment of the present application. Figure 6 As shown, the device includes:

[0086] An acquisition module 60 is configured to sample the voice signal at a preset sampling frequency to obtain sample voice data, wherein the preset sampling frequency is determined according to a state of a channel for transmitting the voice signal;

[0087] The compression module 62 is used to compress the speech data to be compressed using a target neural network model to obtain target speech data, wherein the target neural network model is trained in the following manner: taking sample speech data and the channel state when the sample speech data is sampled as input, and taking the compressed speech data obtained by compressing the sample speech data using a compression algorithm as output, and training the target neural network model.

[0088] By means of the above-mentioned satellite communication voice compression device, the voice signal is sampled according to a preset sampling frequency to obtain sample voice data, wherein the preset sampling frequency is determined according to the channel state of the voice signal being transmitted; the target neural network model is used to compress the voice data to be compressed to obtain target voice data, wherein the target neural network model is trained in the following manner: taking the sample voice data and the channel state when the sample voice data is sampled as input, and taking the compressed voice data obtained by compressing the sample voice data using a compression algorithm as output, and training the target neural network model, thereby achieving the purpose of compressing the voice data using a neural network model, and then ensuring the voice quality while improving the compression rate, thereby achieving the technical effect of improving the bandwidth utilization rate in the voice transmission process, and thus solving the technical problem in the related art that the bandwidth resource utilization rate is reduced due to excessive bandwidth resources being occupied in order to ensure voice quality.

[0089] The acquisition module 60 in the above-mentioned satellite communication voice compression device includes: a sampling submodule, which is used to sample the voice signal according to a preset sampling frequency to obtain sample voice data, including: obtaining the channel state for transmitting the voice signal, the channel state including: the bandwidth and signal-to-noise ratio of the channel; determining the channel state for transmitting the voice signal based on the bandwidth and the signal-to-noise ratio of the channel; if the channel state meets the preset conditions, sampling the voice signal at a first preset frequency; if the channel state does not meet the preset conditions, sampling the voice signal at a second preset frequency, the first preset frequency being higher than the second preset frequency.

[0090] The sampling submodule includes a compression unit for compressing the sample voice data using a compression algorithm, including: pre-emphasis processing on the sample voice data; dividing the pre-emphasized sample voice data into multiple frames of fixed length, wherein the number of iterations required for the target neural network model is equal to the number of voice data samples contained in each fixed-length frame; and compressing the multiple fixed-length frames to obtain the compressed voice data.

[0091] The compression module 62 includes: a training submodule, which is used to determine the sample voice data, the channel state when sampling the sample voice data, and the compressed voice data as a training data set, and randomly select part of the data from the training data set to pre-train the initial neural network model to obtain initial model parameters; after setting the initial neural network model according to the initial model parameters, multiple fixed-length frames corresponding to the sample voice data are respectively input into the initial neural network model to perform forward propagation to obtain a model output value; based on the comparison between the model output value and the compressed voice data, the loss function value is determined, and the model parameters of the initial neural network model are adjusted according to the loss function value until the loss function converges to obtain the target neural network model.

[0092] The training submodule includes: an adjustment unit for adjusting the model parameters of the initial neural network model according to the loss function value, including: obtaining the gradient of each model parameter in the initial neural network model with respect to the loss function; and completing the update of each model parameter according to the gradient of each model parameter with respect to the loss function.

[0093] The compression unit includes: a pre-emphasis subunit, which is used to perform pre-emphasis processing on the sample voice data, including: obtaining the voice signal at each time point in the sample voice data; and performing first-order high-pass filtering on the voice signal at each time point to complete the pre-emphasis processing.

[0094] The satellite communication voice compression device also includes: a sending submodule, which is used to deploy an initial neural network model in a communication satellite; sending model parameters corresponding to the target neural network model to the communication satellite to adjust the initial neural network model according to the model parameters to obtain the target neural network model; receiving the compression result of the communication satellite using the target neural network model on the voice data to be compressed, and updating the target neural network model according to the compression result.

[0095] It should be noted that Figure 6 The satellite communication voice compression device shown is used to perform Figure 2 The satellite communication voice compression method shown in the figure, therefore the relevant explanations in the above satellite communication voice compression method are also applicable to the satellite communication voice compression device, and will not be repeated here.

[0096] An embodiment of the present application also provides a computer device, including: 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 above-mentioned satellite communication voice compression method.

[0097] The satellite communication voice compression method executed by the above-mentioned computer device obtains sample voice data by sampling the voice signal according to a preset sampling frequency, wherein the preset sampling frequency is determined according to the channel state of the voice signal; and compresses the voice data to be compressed using a target neural network model to obtain target voice data, wherein the target neural network model is trained in the following manner: taking the sample voice data and the channel state when the sample voice data is sampled as input, and taking the compressed voice data obtained by compressing the sample voice data using a compression algorithm as output, and training the target neural network model, thereby achieving the purpose of compressing the voice data using a neural network model, and then ensuring the voice quality while improving the compression rate, thereby achieving the technical effect of improving the bandwidth utilization rate during voice transmission, and thus solving the technical problem in the related art that the bandwidth resource utilization rate is reduced due to excessive bandwidth resources occupied in order to ensure voice quality.

[0098] An embodiment of the present application also 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 above-mentioned satellite communication voice compression method by running the computer program.

[0099] The satellite communication voice compression method program stored in the above-mentioned non-volatile storage medium obtains sample voice data by sampling the voice signal according to a preset sampling frequency, wherein the preset sampling frequency is determined according to the channel state of the voice signal transmitted; the target neural network model is used to compress the voice data to obtain target voice data, wherein the target neural network model is trained in the following manner: taking the sample voice data and the channel state when the sample voice data is sampled as input, and taking the compressed voice data obtained by compressing the sample voice data using a compression algorithm as output, and training the target neural network model, thereby achieving the purpose of compressing the voice data using a neural network model, and then ensuring the voice quality while improving the compression rate, thereby achieving the technical effect of improving the bandwidth utilization rate during voice transmission, and thus solving the technical problem in the related art that the bandwidth resource utilization rate is reduced due to excessive bandwidth resources occupied in order to ensure voice quality.

[0100] An embodiment of the present application also provides a computer program product, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the satellite communication voice compression method in the present application are implemented.

[0101] An embodiment of the present application further provides a computer program product, comprising computer instructions, which, when executed by a processor, implement the steps of the satellite communication voice compression method in the present application.

[0102] An embodiment of the present application also provides a computer program, which, when executed by a processor, implements the steps of the satellite communication voice compression method in the present application.

[0103] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0104] In the above embodiments of the present application, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0105] 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 exemplary. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods, such as 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.

[0106] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0107] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0108] 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 is essentially 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 for enabling a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0109] The above is only a preferred embodiment 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 compressing satellite communication voice, characterized in that: include: Sampling the voice signal at a preset sampling frequency to obtain sample voice data, wherein the preset sampling frequency is determined according to a state of a channel for transmitting the voice signal; Using a target neural network model to compress the speech data to be compressed to obtain target speech data, wherein the target neural network model is trained by taking sample speech data and a channel state when the sample speech data is sampled as input, and taking compressed speech data obtained by compressing the sample speech data using a compression algorithm as output, and training the target neural network model; The target neural network is obtained by the following method, including: Determining the sample speech data, the channel state when the sample speech data is sampled, and the compressed speech data as a training data set, and randomly selecting a portion of data from the training data set to pre-train an initial neural network model to obtain initial model parameters; After setting the initial neural network model according to the initial model parameters, a plurality of fixed-length frames corresponding to the sample speech data are respectively input into the initial neural network model, and forward propagation is performed to obtain a model output value; Based on the comparison between the model output value and the compressed speech data, the loss function value is determined, and the model parameters of the initial neural network model are adjusted according to the loss function value until the loss function converges to obtain the target neural network model.

2. The method according to claim 1, characterized in that Sampling the voice signal at a preset sampling frequency to obtain sample voice data includes: Acquiring a channel state for transmitting the voice signal, the channel state including: a bandwidth and a signal-to-noise ratio of the channel; Determining a channel state for transmitting the voice signal according to the bandwidth of the channel and the signal-to-noise ratio; When the channel state meets the preset conditions, the voice signal is sampled using a first preset frequency; when the channel state does not meet the preset conditions, the voice signal is sampled using a second preset frequency, and the first preset frequency is higher than the second preset frequency.

3. The method according to claim 2, characterized in that The sample voice data is compressed using a compression algorithm, including: Performing pre-emphasis processing on the sample voice data; Dividing the sample speech data after pre-emphasis processing into a plurality of frames of fixed length, wherein the number of iterations required for the target neural network model is equal to the number of speech data samples contained in each frame of fixed length; The multiple fixed-length frames are compressed to obtain the compressed voice data.

4. The method according to claim 1, wherein Adjusting the model parameters of the initial neural network model according to the loss function value includes: Obtaining the gradient of each model parameter in the initial neural network model with respect to the loss function; The updating of each model parameter is completed according to the gradient of each model parameter with respect to the loss function.

5. The method according to claim 3, characterized in that The sample voice data is subjected to pre-emphasis processing, comprising: Acquire a speech signal at each time point in the sample speech data; Performing first-order high-pass filtering on the speech signal at each time point to complete the pre-emphasis processing.

6. The method according to claim 1, characterized in that The method further comprises: Deployment of initial neural network models in communications satellites; Sending model parameters corresponding to the target neural network model to the communication satellite, so as to adjust the initial neural network model according to the model parameters to obtain the target neural network model; Receive the compression result of the communication satellite on the voice data to be compressed using the target neural network model, and update the target neural network model according to the compression result.

7. A satellite communication voice compression device, characterized in that: include: An acquisition module, configured to sample the voice signal at a preset sampling frequency to obtain sample voice data, wherein the preset sampling frequency is determined according to a state of a channel for transmitting the voice signal; a compression module, configured to compress the speech data to be compressed using a target neural network model to obtain target speech data, wherein the target neural network model is trained by taking sample speech data and a channel state when the sample speech data is sampled as input, and taking compressed speech data obtained by compressing the sample speech data using a compression algorithm as output, and training the target neural network model; The target neural network is obtained by the following method, including: Determining the sample speech data, the channel state when the sample speech data is sampled, and the compressed speech data as a training data set, and randomly selecting a portion of data from the training data set to pre-train an initial neural network model to obtain initial model parameters; After setting the initial neural network model according to the initial model parameters, a plurality of fixed-length frames corresponding to the sample speech data are respectively input into the initial neural network model, and forward propagation is performed to obtain a model output value; Based on the comparison between the model output value and the compressed speech data, the loss function value is determined, and the model parameters of the initial neural network model are adjusted according to the loss function value until the loss function converges to obtain the target neural network model.

8. A computer 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 satellite communication voice compression method described in any one of claims 1 to 6.

9. A computer program product comprising computer instructions, characterized in that When the computer instructions are executed by a processor, the satellite communication voice compression method according to any one of claims 1 to 6 is implemented.

Images