CN105513610A - Voice analysis method and device - Google Patents

Abstract

The embodiment of the present invention discloses a sound analysis method and device, relates to the technical field of sound recognition, and can improve the accuracy of source device recognition of audio files at a relatively low cost. The method of the present invention comprises: with the sound signal of collecting, obtain the audio file corresponding to different compression algorithm respectively with the same sampling rate and bit rate according to the sound signal collected by different compression algorithms; From the audio file corresponding to different compression algorithm extract the silent segment, and obtain the speech feature signal according to the extracted silent segment; use the speech feature signal as training data to train the BP neural network, and analyze the test signal by completing the training of the BP neural network to identify the signal that generates the test signal Recording Devices The present invention is applicable to identifying the source device of an audio file.

Description

A sound analysis method and device

technical field

The invention relates to the technical field of voice recognition, in particular to a voice analysis method and device.

Background technique

With the popularization of various electronic devices, recording equipment has been widely used in the field. Especially in the practice of judiciary and law enforcement, the collection of audio files has become an important means of investigation and evidence collection. However, due to the easy forgery of audio files and the low ability to restore case scenes, audio files can only be used as a reference in many cases.

The equipment used to record the audio file reflects the recording occasion and situation to a certain extent, which is very important for judging whether the audio file can be used as valid evidence. However, at present, the effective identification of recording equipment for audio files is mainly based on the experience of the case investigators, and the accuracy is difficult to guarantee. The cost of professional voiceprint analysis equipment is very high, and the cost of voice identification analysis remains high. . It can be seen that at present, the identification of the source equipment of audio files is difficult and the accuracy rate is low, and the cost of professional voiceprint analysis and identification is very high, and it is difficult to popularize it in grassroots law enforcement and judiciary.

Contents of the invention

Embodiments of the present invention provide a sound analysis method and device, which can improve the accuracy of identifying the source device of an audio file at a relatively low cost.

In order to achieve the above object, embodiments of the present invention adopt the following technical solutions:

In a first aspect, embodiments of the present invention provide a sound analysis method, including:

Using different compression algorithms to obtain audio files corresponding to different compression algorithms according to the collected sound signals at the same sampling rate and bit rate;

Extracting silent segments from audio files corresponding to different compression algorithms, and obtaining speech feature signals according to the extracted silent segments;

Using the speech feature signal as training data to train a BP (BackPropagation, multi-layer feed-forward) neural network, and analyzing the test signal through the trained BP neural network to identify the recording device that generated the test signal.

In a second aspect, an embodiment of the present invention provides a sound analysis device, including: a system main control module, a voice recording and playback module, a TFT touch screen module, a compression algorithm realization module, a storage module and a host computer module connected to each other through a bus;

The voice recording and playback module is used to play sound signals;

The compression algorithm implementation module is used to obtain audio files respectively corresponding to different compression algorithms according to the collected sound signals with the same sampling rate and bit rate through different compression algorithms;

The storage module is used to store the audio files corresponding to different compression algorithms;

The upper computer module is used to extract silent segments from audio files corresponding to different compression algorithms, and obtain speech feature signals according to the extracted silent segments; and use the speech feature signals as training data to train BP neural network, and The test signal is analyzed through the trained BP neural network, and the recording device generating the test signal is identified.

The sound analysis method and device provided by the embodiments of the present invention aim at adopting different compression algorithms with the same sampling rate and bit rate according to the collected sound signal, extracting the MFCC parameters of the silent segment of the recording and improving it respectively, and using different baud Input the high-frequency audio file into Matlab to obtain the corresponding MFCC characteristic parameters, and then use the MFCC characteristic parameters to train the BP neural network, use the trained BP neural network to classify the speech characteristic signals, and identify the recording device according to the classification results. Due to STM32 and Matlab, etc. The device used in the present invention has low cost, so the accuracy of identifying the source device of the audio file can be improved at a relatively low cost.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the accompanying drawings that need to be used in the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For Those of ordinary skill in the art can also obtain other drawings based on these drawings without making creative efforts.

Fig. 1 is the flowchart of the sound analysis method that the embodiment of the present invention provides;

Fig. 2 is a schematic diagram of a specific device for implementing the sound analysis method provided by the embodiment of the present invention;

3 is a schematic flow diagram of a silent segment extraction solution provided by an embodiment of the present invention;

Fig. 4 is the schematic flow chart of the improved MFCC parameter extraction scheme that the embodiment of the present invention provides;

Fig. 5 is the schematic flow chart of the speech characteristic signal classification algorithm based on BP neural network that the embodiment of the present invention provides;

FIG. 6 is a schematic flowchart of a recording device identification solution provided by an embodiment of the present invention;

Fig. 7 is a schematic structural diagram of a sound analysis device provided by an embodiment of the present invention.

detailed description

In order to enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. Hereinafter, embodiments of the present invention will be described in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the figures are exemplary only for explaining the present invention and should not be construed as limiting the present invention.

Those skilled in the art will understand that unless otherwise stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the word "comprising" used in the description of the present invention refers to the presence of said features, integers, steps, operations, elements and/or components, but does not exclude the presence or addition of one or more other features, Integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Additionally, "connected" or "coupled" as used herein may include wirelessly connected or coupled. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.

Those skilled in the art can understand that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It should also be understood that terms such as those defined in commonly used dictionaries should be understood to have a meaning consistent with the meaning in the context of the prior art, and unless defined as herein, are not to be interpreted in an idealized or overly formal sense Explanation.

The embodiment of the present invention provides a sound analysis method, as shown in Figure 1, including:

101. Use different compression algorithms to obtain audio files respectively corresponding to different compression algorithms according to the collected sound signals at the same sampling rate and bit rate.

In this embodiment, the specific execution flow of the sound analysis method can be based on the device of the architecture shown in Figure 2, and the STM32 enhanced series F103VET6 is specifically selected as the system master solution; the storage module includes a CH376 U disk storage circuit and an SD card storage module ; Compression algorithm implementation modules include MP3, AMR, AAC, WMA four audio compression algorithm modules. Coordinate voice recording and playback module, storage module, TFT (ThinFilmTransistor, thin film transistor) touch screen module, compression algorithm implementation module, serial port and other interfaces. Voice recording and playback module includes ISD4004 module, LM386 power amplifier circuit, filter bias module.

When the device is powered on, it can record a piece of voice, press the stop button to end the recording, and go through four different compression algorithms, and then save the recorded four pieces of voice with the same sampling rate and bit rate to the U disk or SD card. Among them, the microSD card used by the SD card is connected to the STM32 main control module by means of SDIO (SecureDigitalInputandOutputCard, a secure digital input and output card), and supports a maximum of 8GSD cards; the U disk storage module uses CH376T as the core, and uses a USBA interface to connect to the U disk , supports up to 8GU disk. The power supply is specifically a 5V power adapter, and the 3.3V voltage is provided by the AMS1117 chip.

102. Extract silent segments from audio files corresponding to different compression algorithms, and obtain speech feature signals according to the extracted silent segments.

Specifically, it can be implemented on the host computer. First, the silent segment is extracted. The process of extracting the silent segment is shown in FIG. 3 .

In this embodiment, the compression algorithm includes 4 different compression algorithms, including MP3, AMR, WMA and AAC. The described according to the extracted silent segment obtains the speech feature signal, specifically comprises: by seeking to improve the flow process of MFCC (MelFrequencyCepstrumCoefficient, Mel frequency cepstral coefficient) parameter as shown in Figure 4, and for each segment of silent segment, adopt cepstrum 500 groups of 24-dimensional speech feature signals were extracted by mathematical method.

103. Use the speech feature signal as training data to train a BP neural network, and analyze a test signal through the trained BP neural network, and identify a recording device that generates the test signal.

In this embodiment, the structure of the BP neural network includes: 24 nodes are set in the input layer, 25 nodes are set in the hidden layer, and 4 nodes are set in the output layer.

For example: as shown in Figure 5. The BP neural network is constructed by the development tool MATLAB2014a, and the silent segment is extracted through programming, and the characteristic parameters are extracted in the speech silent segment, which avoids the interference of the voice signal. Finally, the recognition model BP neural network of the recording device recognition system is determined.

Specifically, the construction of the BP neural network determines the structure of the BP neural network according to the characteristics of the input and output data of the system. Since the speech feature input signal has 24 dimensions, there are four types of speech signals to be classified, so the structure of the BP neural network is 24-25-4 That is, the input layer has 24 nodes, the hidden layer has 25 nodes, and the output layer has 4 nodes.

In the training phase, BP neural network training uses training data to train the BP neural network. For example, there are 2000 sets of speech feature signals, 1500 sets of data are randomly selected as training data to train the network, and 500 sets of data are used as test data to test the classification ability of the network.

In the test phase after training, the BP neural network classification uses the trained neural network to classify the speech category to which the test data belongs. In this way, the overall process shown in Figure 6 is realized, that is, for the collected sound signal, four segments of speech in different audio formats are obtained, and then after the processing of the host computer is completed, the audio format can be recognized by inputting a segment of speech to determine which audio format is used. recorded by recording equipment.

The sound analysis method that the embodiment of the present invention provides, aiming at adopting different compression algorithms with the same sampling rate and bit rate according to the collected sound signal, extracting the MFCC parameters of the silent segment of the recording and improving it respectively, the MFCC parameters of different baud rates Audio file input in Matlab obtains corresponding MFCC characteristic parameter, utilizes MFCC characteristic parameter to train BP neural network again, classifies voice characteristic signal with trained BP neural network, recognizes recording equipment according to classification result, due to the present invention such as STM32 and Matlab The cost of the equipment used is low, so the accuracy of identifying the source equipment of the audio file can be improved at a relatively low cost.

Further, an embodiment of the present invention provides a sound analysis device, as shown in FIG. 7 , including: a system main control module, a voice recording and playback module, a TFT touch screen module, a compression algorithm realization module, a storage module and PC module.

The voice recording and playback module is used for playing sound signals.

The compression algorithm implementation module is used to obtain audio files respectively corresponding to different compression algorithms according to the collected sound signals through different compression algorithms at the same sampling rate and bit rate.

The storage module is used for storing the audio files corresponding to different compression algorithms.

The host computer module is used to extract silent segments from audio files corresponding to different compression algorithms, and obtain speech feature signals according to the extracted silent segments. And use the speech feature signal as training data to train the BP neural network, and analyze the test signal through the trained BP neural network to identify the recording device that generates the test signal.

Specifically, the main control module of the system is an STM32 enhanced F103VET6 chip, which is a 32-bit enhanced MCU, adopts the cortex-M3 core of ARM Company, has 512KFlash, 64KRAM, 3 SPI ports, an SDIO port, 5 USARTs, up to 72M main frequency. The voice recording and playback module is ISD4004, which performs audio amplification through the LM386 integrated audio power amplifier circuit, and the recording time is set to 8-16 minutes. Since the voice signal needs to be collected and synthesized, multiple acquisitions and quantization will cause certain quantization errors. ISD4004 is used For recording, through the multi-level direct analog storage technology, each sampling value is directly stored in the on-chip flash memory, so the voice can be reproduced very realistically and naturally. Among them, LM386 integrated audio power amplifier circuit is selected for audio amplification, and AMS1117-3.3 is selected for voltage stabilization. Described host computer module specifically extracts silent segment from the audio file of corresponding different compression algorithms by MATLAB2014a, and obtains speech characteristic signal according to the silent segment extracted, and utilizes described speech characteristic signal as training data training BP neural network, and The test signal is analyzed through the trained BP neural network, and the recording device generating the test signal is identified.

The sound analysis device that the embodiment of the present invention provides, aiming at adopting different compression algorithms with the same sampling rate and bit rate according to the collected sound signals, extracting the MFCC parameters of the silent segment of the recording and improving it respectively, the MFCC parameters of different baud rates Audio file input in Matlab obtains corresponding MFCC characteristic parameter, utilizes MFCC characteristic parameter to train BP neural network again, classifies voice characteristic signal with trained BP neural network, recognizes recording equipment according to classification result, due to the present invention such as STM32 and Matlab The cost of the equipment used is low, so the accuracy of identifying the source equipment of the audio file can be improved at a relatively low cost.

Each embodiment in this specification is described in a progressive manner, the same and similar parts of each embodiment can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and for relevant parts, please refer to part of the description of the method embodiment.

The above is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Anyone skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present invention. All should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the protection scope of the claims.

Claims (5)

1. A sound analysis method, characterized in that, comprising: Using different compression algorithms to obtain audio files corresponding to different compression algorithms according to the collected sound signals at the same sampling rate and bit rate; Extracting silent segments from audio files corresponding to different compression algorithms, and obtaining speech feature signals according to the extracted silent segments; Using the speech feature signal as training data to train the BP neural network, and analyzing the test signal through the trained BP neural network, and identifying the recording device that generates the test signal. 2. The method according to claim 1, wherein the compression algorithm comprises 4 different compression algorithms, including MP3, AMR, WMA and AAC; The obtaining the speech feature signal according to the extracted silent segment includes: for each silent segment, extracting 500 groups of 24-dimensional speech feature signals by using the cepstral coefficient method. 3. The method according to claim 2, wherein the structure of the BP neural network comprises: the input layer is provided with 24 nodes, the hidden layer is provided with 25 nodes, and the output layer is provided with 4 nodes. 4. A sound analysis device, characterized in that it comprises: a system main control module, a voice recording and playback module, a TFT touch screen module, a compression algorithm realization module, a storage module and a host computer module connected to each other by a bus; The voice recording and playback module is used to play sound signals; The compression algorithm implementation module is used to obtain audio files respectively corresponding to different compression algorithms according to the collected sound signals with the same sampling rate and bit rate through different compression algorithms; The storage module is used to store the audio files corresponding to different compression algorithms; The upper computer module is used to extract silent segments from audio files corresponding to different compression algorithms, and obtain speech feature signals according to the extracted silent segments; and use the speech feature signals as training data to train BP neural network, and The test signal is analyzed through the trained BP neural network, and the recording device generating the test signal is identified. 5. The method according to claim 4, wherein the system main control module is an STM32 enhanced F103VET6 chip; Described voice recording and playback module is ISD4004, carries out audio amplification by LM386 integrated audio power amplifier circuit; Described host computer module specifically extracts silent segment from the audio file of corresponding different compression algorithms by MATLAB2014a, and obtains speech characteristic signal according to the silent segment extracted, and utilizes described speech characteristic signal as training data training BP neural network, and The test signal is analyzed through the trained BP neural network, and the recording device generating the test signal is identified.