Interphone implementation system based on automatic voice recognition

Abstract

The invention discloses an interphone implementation system based on voice automatic recognition, and relates to the technical field of voice recognition, and the system comprises the steps: collecting a voice signal of a user through an interphone microphone, and converting the voice signal into an analog electric signal; converting the analog electric signal into a digital audio signal by using an AAC audio coding algorithm, and carrying out coding processing; voice data packets lost in the transmission process are repaired in a data redundancy mode by using a forward error correction technology; framing the audio signal to be processed by using an overlapped sliding window method; detecting a voice area endpoint in the voice frame by using a double-threshold judgment method; training a speech recognition hidden Markov model through the speech area data and the corresponding statements in the information base; and obtaining statement information corresponding to the voice signal of the user by using a Viterbi algorithm. By arranging the audio repairing module, the endpoint detection module, the model training module and the voice recognition module, the voice quality is improved, the call distance is increased, and thus the working efficiency of a user is improved.

Description

Technical Field

[0001] The invention relates to the technical field of speech recognition, and in particular to an intercom implementation system based on automatic speech recognition. Background Art

[0002] Wireless walkie-talkies belong to private network mobile communications. They have the advantages of no need for wiring, simple networking, and one-touch communication. At the same time, they do not require the participation of third-party operators, only require a one-time investment and no regular payment, saving a lot of call costs. After World War II, wireless walkie-talkies gradually moved from the military field to the civilian field, making them useful in transportation, public utilities, industry, construction, manufacturing and other fields. To this day, the application scope and demand of wireless walkie-talkies are still increasing. The research on speech recognition began in the 1950s with the study of Audry, the world's first speech recognition system that can recognize English and numbers, which opened a chapter in the history of speech recognition.

[0003] With the continuous expansion of the application field of walkie-talkies and people's different business needs, walkie-talkie products have begun to show diversified development. At present, in busy business places such as small business offices and bank business halls, office workers have a short call distance, frequent communication and are very sensitive to product prices. Therefore, they need a walkie-talkie that can be operated freely even in busy business, with low price, low power consumption and excellent voice quality to meet their needs. Summary of the invention

[0004] In order to solve the above technical problems, a walkie-talkie implementation system based on automatic speech recognition is provided. This technical solution solves the problems raised in the above background technology.

[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0006] A walkie-talkie implementation system based on automatic speech recognition, comprising:

[0007] A voice acquisition module, which is used to collect the user's voice signal through the intercom microphone and convert it into an analog electrical signal;

[0008] An audio encoding module, wherein the audio encoding module is used to convert analog electrical signals into digital audio signals using an AAC audio encoding algorithm and perform encoding processing;

[0009] An audio repair module, wherein the audio repair module is used to repair voice data packets lost during transmission by using forward error correction technology and data redundancy to obtain an audio signal to be processed;

[0010] An audio framing module, wherein the audio framing module is used to perform framing processing on the audio signal to be processed by using an overlapping sliding window method to obtain at least two speech frames;

[0011] An endpoint detection module, the endpoint detection module is used to detect the endpoint of the speech area in the speech frame by using a double threshold decision method, and separate the speech area data from the noise area data;

[0012] A model training module, wherein the model building module is used to train a speech recognition hidden Markov model through speech area data and corresponding sentences in the information library;

[0013] The speech recognition module is used to input the speech area data obtained by processing the user's speech signal into the speech recognition hidden Markov model, and use the Viterbi algorithm to obtain the sentence information corresponding to the user's speech signal.

[0014] Preferably, the voice collection module collects the user's voice signal through the intercom microphone and converts it into an analog electrical signal, specifically including:

[0015] The user's voice signal is transmitted to the intercom microphone, compressing the air and causing the diaphragm inside the intercom microphone to vibrate;

[0016] The vibration of the diaphragm changes the distance between the diaphragm and the electrode, thereby changing the capacitance of the capacitor;

[0017] The change in the capacitance of the capacitor inside the intercom microphone generates a weak electrical signal;

[0018] Weak electric signal input electric signal amplifier;

[0019] At least one semiconductor transistor inside the electrical signal amplifier regulates the flow of current in proportion to changes in the input signal, amplifies the weak electrical signal, and obtains an analog electrical signal.

[0020] Preferably, the audio encoding module converts the analog electrical signal into a digital audio signal using an AAC audio encoding algorithm, and performs encoding processing specifically including:

[0021] A model acquisition unit, wherein the model acquisition unit is used to acquire a psychoacoustic model;

[0022] A threshold calculation unit, the threshold calculation unit is used to calculate a masking threshold of the analog electrical signal based on a psychoacoustic model;

[0023] A signal-to-mask ratio calculation unit, the signal-to-mask ratio calculation unit is used to process the signal strength of the analog electrical signal and the masking threshold value to obtain the signal-to-mask ratio of the analog electrical signal;

[0024] A noise control unit, the noise control unit is used to filter the analog electrical signal based on the signal-to-mask ratio of the analog electrical signal to obtain a noise-reduced electrical signal;

[0025] The noise judgment unit is used to judge whether the signal-to-mass ratio of the noise reduction electrical signal is lower than a set threshold. If so, a framing operation is performed on the noise reduction electrical signal and the operation result is output as a digital audio signal. If not, a secondary filtering process is performed on the noise reduction electrical signal.

[0026] Preferably, the audio repair module uses forward error correction technology to repair voice data packets lost during transmission in a data redundancy manner to obtain the audio signal to be processed, specifically including:

[0027] Performing an XOR operation on each group of voice data packets in the digital audio signal to generate a parity check packet;

[0028] Copying the data of the previous group of voice data packets to the first half of the next group of voice data packets, and outputting them as redundant data packets of the next group of voice data packets;

[0029] When transmitting each group of voice data packets, a parity check packet and a redundant data packet of the group of voice data packets are additionally transmitted;

[0030] The data receiving end detects the parity check packets of each group of voice data packets;

[0031] Determine whether the parity check packets of adjacent groups of voice data packets have different attributes. If so, output the parity check packets with the same attributes corresponding to the voice data packets. If not, do not output;

[0032] The lost voice data packet is repaired by querying the redundant data packets of a group of voice data packets after the lost position;

[0033] The repaired voice data packets are aggregated and output as audio signals to be processed.

[0034] Preferably, the audio framing module performs framing processing on the audio signal to be processed by using an overlapping sliding window method to obtain at least two speech frames, specifically comprising:

[0035] S101: Preset the frame length of the voice frame;

[0036] S102: taking two-thirds of the frame length of the speech frame as the frame shift of the speech frame;

[0037] S103: Selecting a Hamming window function as a sliding window function;

[0038] S104: Aligning the starting position of the window with the starting position of the audio signal to be processed;

[0039] S105: multiplying the window with the data at the corresponding position of the audio signal to be processed to achieve weighted processing and reduce the influence of signal mutation at the window boundary;

[0040] S106: Number the weighted results and record them as speech frames;

[0041] S107: Determine whether the window slides to the end of the audio signal to be processed. If so, output at least two speech frames. If not, slide the window forward along the time axis of the audio signal to be processed by one frame, and repeat step S105.

[0042] The Hamming window function is:

[0043]

[0044] Where ω(n) is the function value of the Hamming window function of the nth signal, and a is the Hamming window parameter, which is N is the frame length of the speech frame and also the width of the Hamming window.

[0045] Preferably, the endpoint detection module detects the speech area endpoints in the speech frame using a double threshold decision method, and separates the speech area data from the noise area data, specifically including:

[0046] Calculate the short-time energy of each speech frame;

[0047] Calculate the short-time zero-crossing rate of each speech frame;

[0048] The first threshold is set to the short-time average energy of the speech frame;

[0049] The second threshold is set to the short-time average energy of the noise;

[0050] The third threshold is set to the short-time average zero-crossing rate of the noise;

[0051] Obtain time points A and B with the same short-time average energy as the first threshold, and preliminarily determine that the start and end points of the speech segment are outside the AB segment;

[0052] Search point A to the left and point B to the right to obtain point C where the left side of point A intersects with the second threshold for the first time and point D where the right side of point B intersects with the second threshold for the first time;

[0053] Search leftward from point C and rightward from point D to obtain point E where the left side of point C intersects with the third threshold for the first time and point F where the right side of point D intersects with the third threshold for the first time;

[0054] The EF segment of the output speech frame is the speech area data, and the remaining data is the noise area data;

[0055] The short-time average energy is:

[0056]

[0057] In the formula, E n is the short-time average energy of the nth signal, m is the time sequence number of each speech frame, and x(m is the signal value at the mth moment;

[0058] The short-time average zero-crossing rate is:

[0059]

[0060] In the formula, Z n is the short-time average zero-crossing rate of the nth signal, sgn(x) is the mathematical symbol function;

[0061] The mathematical symbol function is:

[0062]

[0063] Where sgn(x) is a mathematical symbol function.

[0064] Preferably, the model building module trains the speech recognition hidden Markov model through the speech area data and corresponding sentences in the information library, specifically including:

[0065] Determining the number of states of speech zone data in the information base, wherein the states include phonemes and fine-grained speech units;

[0066] Define the observation set as all statement output values ​​that the model may observe;

[0067] The state transition probability matrix of each data in the calculation model from the current state to other states;

[0068] Calculate the observation probability matrix of each data in the model to observe a specific observation value in the current state;

[0069] Initialize the initial probability distribution of each data in the model in various states at the initial moment;

[0070] The number of states, the set of observation values, the state transition probability matrix, the observation probability matrix and the initial probability distribution are used as five-tuple parameters to construct a hidden Markov model.

[0071] Preferably, the speech recognition module inputs the speech area data obtained by processing the user's speech signal into the speech recognition hidden Markov model, and uses the Viterbi algorithm to obtain the sentence information corresponding to the user's speech signal, which specifically includes:

[0072] S201: Acquire the data at each moment and each state of the voice zone data obtained by processing the user's voice signal;

[0073] S202: Initialize the current time;

[0074] S203: Calculate the probability of passing through each state up to the current moment, and output the state with the highest probability as the passing state;

[0075] S204: Determine whether the current moment is the last moment of the voice zone data. If so, output all the passed state sequences as corresponding sentence information. If not, move the current moment back and return to step S203.

[0076] Compared with the prior art, the present invention has the following beneficial effects:

[0077] By setting up the voice acquisition module, audio encoding module, audio repair module, audio framing module, endpoint detection module, model training module and voice recognition module, users can perform common operations on the digital wireless walkie-talkie through voice commands, thereby improving user work efficiency. Through the study of voice characteristics and voice packet loss processing technology, it is proposed to use forward error correction technology at the data receiving end to reduce the voice packet loss rate and the loss caused by packet loss, thereby improving voice quality and increasing call distance. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 This is a schematic diagram of the structure of the intercom system based on automatic speech recognition of the present invention;

[0079] Figure 2 A flow chart of a method for collecting a user's voice signal according to the present invention;

[0080] Figure 3 AAC audio coding algorithm method flow chart of the present invention;

[0081] Figure 4 A flow chart of a method for repairing voice data packets lost during transmission by using forward error correction technology and adopting data redundancy according to the present invention;

[0082] Figure 5 It is a flow chart of a method for performing frame processing on an audio signal to be processed by using an overlapping sliding window method according to the present invention;

[0083] Figure 6 A flow chart of a method for detecting speech area endpoints in a speech frame using a dual threshold decision method according to the present invention;

[0084] Figure 7 A flow chart of the method for training a hidden Markov model for speech recognition of the present invention;

[0085] Figure 8 The present invention is a flow chart of a method for obtaining sentence information corresponding to a user's voice signal by using the Viterbi algorithm. DETAILED DESCRIPTION

[0086] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art may think of other obvious variations.

[0087] Reference Figure 1 As shown, a walkie-talkie implementation system based on automatic speech recognition includes:

[0088] A voice acquisition module, which is used to collect the user's voice signal through the intercom microphone and convert it into an analog electrical signal;

[0089] An audio encoding module, wherein the audio encoding module is used to convert analog electrical signals into digital audio signals using an AAC audio encoding algorithm and perform encoding processing;

[0090] An audio repair module, wherein the audio repair module is used to repair voice data packets lost during transmission by using forward error correction technology and data redundancy to obtain an audio signal to be processed;

[0091] An audio framing module, wherein the audio framing module is used to perform framing processing on the audio signal to be processed by using an overlapping sliding window method to obtain at least two speech frames;

[0092] An endpoint detection module, the endpoint detection module is used to detect the endpoint of the speech area in the speech frame by using a double threshold decision method, and separate the speech area data from the noise area data;

[0093] A model training module, wherein the model building module is used to train a speech recognition hidden Markov model through speech area data and corresponding sentences in the information library;

[0094] The speech recognition module is used to input the speech area data obtained by processing the user's speech signal into the speech recognition hidden Markov model, and use the Viterbi algorithm to obtain the sentence information corresponding to the user's speech signal.

[0095] Reference Figure 2 As shown, the voice acquisition module collects the user's voice signal through the intercom microphone and converts it into an analog electrical signal, specifically including:

[0096] The user's voice signal is transmitted to the intercom microphone, compressing the air and causing the diaphragm inside the intercom microphone to vibrate;

[0097] The vibration of the diaphragm changes the distance between the diaphragm and the electrode, thereby changing the capacitance of the capacitor;

[0098] The change in the capacitance of the capacitor inside the intercom microphone generates a weak electrical signal;

[0099] Weak electric signal input electric signal amplifier;

[0100] At least one semiconductor transistor inside the electrical signal amplifier regulates the flow of current in proportion to changes in the input signal, amplifies the weak electrical signal, and obtains an analog electrical signal.

[0101] The amplification of electrical signals is achieved through amplifiers, which use the regulation of semiconductor transistors on electrical signals to increase the amplitude of input signals. In wireless walkie-talkies, this process is crucial to ensuring signal transmission quality and communication distance.

[0102] Reference Figure 3 As shown, the audio encoding module uses the AAC audio encoding algorithm to convert the analog electrical signal into a digital audio signal, and performs encoding processing specifically including:

[0103] A model acquisition unit, wherein the model acquisition unit is used to acquire a psychoacoustic model;

[0104] A threshold calculation unit, the threshold calculation unit is used to calculate a masking threshold of the analog electrical signal based on a psychoacoustic model;

[0105] A signal-to-mask ratio calculation unit, the signal-to-mask ratio calculation unit is used to process the signal strength of the analog electrical signal and the masking threshold value to obtain the signal-to-mask ratio of the analog electrical signal;

[0106] A noise control unit, the noise control unit is used to filter the analog electrical signal based on the signal-to-mask ratio of the analog electrical signal to obtain a noise-reduced electrical signal;

[0107] The noise judgment unit is used to judge whether the signal-to-mass ratio of the noise reduction electrical signal is lower than a set threshold. If so, a framing operation is performed on the noise reduction electrical signal and the operation result is output as a digital audio signal. If not, a secondary filtering process is performed on the noise reduction electrical signal.

[0108] The AAC audio coding algorithm can effectively convert analog electrical signals into digital audio signals and perform efficient encoding processing on them, thereby significantly reducing the amount of data while maintaining good sound quality, making it easier to store and transmit. This coding method has been widely used in digital audio broadcasting, streaming media services, mobile devices and other fields.

[0109] Reference Figure 4 As shown, the audio repair module uses forward error correction technology and data redundancy to repair the voice data packets lost during the transmission process, and obtains the audio signal to be processed, which specifically includes:

[0110] Performing an XOR operation on each group of voice data packets in the digital audio signal to generate a parity check packet;

[0111] Copying the data of the previous group of voice data packets to the first half of the next group of voice data packets, and outputting them as redundant data packets of the next group of voice data packets;

[0112] When transmitting each group of voice data packets, a parity check packet and a redundant data packet of the group of voice data packets are additionally transmitted;

[0113] The data receiving end detects the parity check packets of each group of voice data packets;

[0114] Determine whether the parity check packets of adjacent groups of voice data packets have different attributes. If so, output the parity check packets with the same attributes corresponding to the voice data packets. If not, do not output;

[0115] The lost voice data packet is repaired by querying the redundant data packets of a group of voice data packets after the lost position;

[0116] The repaired voice data packets are aggregated and output as audio signals to be processed.

[0117] Forward error correction technology is a technology used to control the bit error rate (packet loss, garbled code) of received data packets when data is transmitted in a channel with low transmission channel reliability and strong noise interference.

[0118] Reference Figure 5 As shown, the audio framing module uses an overlapping sliding window method to perform framing processing on the audio signal to be processed, and obtains at least two speech frames, specifically including:

[0119] S101: Preset the frame length of the voice frame;

[0120] S102: taking two-thirds of the frame length of the speech frame as the frame shift of the speech frame;

[0121] S103: Selecting a Hamming window function as a sliding window function;

[0122] S104: Aligning the starting position of the window with the starting position of the audio signal to be processed;

[0123] S105: multiplying the window with the data at the corresponding position of the audio signal to be processed to achieve weighted processing and reduce the influence of signal mutation at the window boundary;

[0124] S106: Number the weighted results and record them as speech frames;

[0125] S107: Determine whether the window slides to the end of the audio signal to be processed. If so, output at least two speech frames. If not, slide the window forward along the time axis of the audio signal to be processed by one frame, and repeat step S105.

[0126] The Hamming window function is:

[0127]

[0128] Where ω(n) is the function value of the Hamming window function of the nth signal, and a is the Hamming window parameter, which is N is the frame length of the speech frame and also the width of the Hamming window.

[0129] When processing speech signals in frames, the overlapping sliding window method is a commonly used method. This method slides a weighted window of finite length on the speech signal. Each time the window slides, it partially overlaps with the previous frame to maintain the continuity and smoothness of the speech signal.

[0130] Reference Figure 6 As shown, the endpoint detection module uses a double threshold decision method to detect the endpoints of the speech area in the speech frame, and separates the speech area data from the noise area data, specifically including:

[0131] Calculate the short-time energy of each speech frame;

[0132] Calculate the short-time zero-crossing rate of each speech frame;

[0133] The first threshold is set to the short-time average energy of the speech frame;

[0134] The second threshold is set to the short-time average energy of the noise;

[0135] The third threshold is set to the short-time average zero-crossing rate of the noise;

[0136] Obtain time points A and B with the same short-time average energy as the first threshold, and preliminarily determine that the start and end points of the speech segment are outside the AB segment;

[0137] Search point A to the left and point B to the right to obtain point C where the left side of point A intersects with the second threshold for the first time and point D where the right side of point B intersects with the second threshold for the first time;

[0138] Search leftward from point C and rightward from point D to obtain point E where the left side of point C intersects with the third threshold for the first time and point F where the right side of point D intersects with the third threshold for the first time;

[0139] The EF segment of the output speech frame is the speech area data, and the remaining data is the noise area data;

[0140] The short-time average energy is:

[0141]

[0142] In the formula, E n is the short-time average energy of the nth signal, m is the time sequence number of each speech frame, and x(m is the signal value at the mth moment;

[0143] The short-time average zero-crossing rate is:

[0144]

[0145] In the formula, Z n is the short-time average zero-crossing rate of the nth signal, sgn(x) is the mathematical symbol function;

[0146] The mathematical symbol function is:

[0147]

[0148] Where sgn(x) is a mathematical symbol function.

[0149] In practical applications, it is basically impossible to ensure that the speech signal is in an environment with a high signal-to-noise ratio. In an environment with a low signal-to-noise ratio, only using the short-time average energy method to determine the starting and ending points of the speech segment can easily cause a large offset in the starting and ending points of the speech segment. Therefore, the short-time average zero-crossing rate method is also required for the next step of judgment. This endpoint detection method that combines the short-time average energy method and the short-time average zero-crossing rate method is called a double threshold judgment method.

[0150] Reference Figure 7 As shown, the model building module trains the speech recognition hidden Markov model through the speech area data and corresponding sentences in the information library, specifically including:

[0151] Determining the number of states of speech zone data in the information base, wherein the states include phonemes and fine-grained speech units;

[0152] Define the observation set as all statement output values ​​that the model may observe;

[0153] The state transition probability matrix of each data in the calculation model from the current state to other states;

[0154] Calculate the observation probability matrix of each data in the model to observe a specific observation value in the current state;

[0155] Initialize the initial probability distribution of each data in the model in various states at the initial moment;

[0156] The number of states, the set of observation values, the state transition probability matrix, the observation probability matrix and the initial probability distribution are used as five-tuple parameters to construct a hidden Markov model.

[0157] The Markov chain used by the hidden Markov model can handle the time-varying characteristics that often appear in speech signals. It is an ideal speech model. Therefore, it has been an important model in the field of speech recognition since it was introduced into the field of speech recognition in the 1970s.

[0158] Reference Figure 8 As shown, the speech recognition module inputs the speech area data obtained by processing the user's speech signal into the speech recognition hidden Markov model, and uses the Viterbi algorithm to obtain the sentence information corresponding to the user's speech signal, which specifically includes:

[0159] S201: Acquire the data at each moment and each state of the voice zone data obtained by processing the user's voice signal;

[0160] S202: Initialize the current time;

[0161] S203: Calculate the probability of passing through each state up to the current moment, and output the state with the highest probability as the passing state;

[0162] S204: Determine whether the current moment is the last moment of the voice zone data. If so, output all the passed state sequences as corresponding sentence information. If not, move the current moment back and return to step S203.

[0163] The Viterbi algorithm is a dynamic programming algorithm used to find the Viterbi path, i.e., the sequence of hidden states, that is most likely to produce a sequence of observed events, particularly in the context of Markov information sources and hidden Markov models.

[0164] Furthermore, the present solution also proposes a computer-readable storage medium on which a computer-readable program is stored. When the computer-readable program is called, the above-mentioned intercom implementation system based on automatic speech recognition is executed.

[0165] It is understandable that the storage medium may be a magnetic medium, such as a floppy disk, a hard disk, or a magnetic tape; an optical medium, such as a DVD; or a semiconductor medium, such as a solid state drive (SSD).

[0166] In summary, the advantages of the present invention are: by setting a voice acquisition module, an audio encoding module, an audio repair module, an audio framing module, an endpoint detection module, a model training module and a voice recognition module, the user can realize common operations on the digital wireless walkie-talkie through voice commands, thereby improving the user's work efficiency. Through the study of voice characteristics and voice packet loss processing technology, it is proposed to use forward error correction technology at the data receiving end to reduce the voice packet loss rate and the loss caused by packet loss, thereby improving the voice quality and increasing the call distance.

[0167] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. A walkie-talkie implementation system based on automatic speech recognition, characterized in that: include: A voice acquisition module, which is used to collect the user's voice signal through the intercom microphone and convert it into an analog electrical signal; An audio encoding module, wherein the audio encoding module is used to convert analog electrical signals into digital audio signals using an AAC audio encoding algorithm and perform encoding processing; An audio repair module, wherein the audio repair module is used to repair voice data packets lost during transmission by using forward error correction technology and data redundancy to obtain an audio signal to be processed; An audio framing module, wherein the audio framing module is used to perform framing processing on the audio signal to be processed by using an overlapping sliding window method to obtain at least two speech frames; An endpoint detection module, the endpoint detection module is used to detect the endpoint of the speech area in the speech frame by using a double threshold decision method, and separate the speech area data from the noise area data; A model training module, wherein the model building module is used to train a speech recognition hidden Markov model through speech area data and corresponding sentences in the information library; The speech recognition module is used to input the speech area data obtained by processing the user's speech signal into the speech recognition hidden Markov model, and use the Viterbi algorithm to obtain the sentence information corresponding to the user's speech signal.

2. According to claim 1, a walkie-talkie implementation system based on automatic speech recognition is characterized in that: The voice acquisition module collects the user's voice signal through the intercom microphone and converts it into an analog electrical signal, which specifically includes: The user's voice signal is transmitted to the intercom microphone, compressing the air and causing the diaphragm inside the intercom microphone to vibrate; The vibration of the diaphragm changes the distance between the diaphragm and the electrode, thereby changing the capacitance of the capacitor; The change in the capacitance of the capacitor inside the intercom microphone generates a weak electrical signal; Weak electric signal input electric signal amplifier; At least one semiconductor transistor inside the electrical signal amplifier regulates the flow of current in proportion to changes in the input signal, amplifies the weak electrical signal, and obtains an analog electrical signal.

3. The intercom implementation system based on automatic speech recognition according to claim 2 is characterized in that: The audio encoding module converts the analog electrical signal into a digital audio signal using the AAC audio encoding algorithm, and performs encoding processing specifically including: A model acquisition unit, wherein the model acquisition unit is used to acquire a psychoacoustic model; A threshold calculation unit, the threshold calculation unit is used to calculate a masking threshold of the analog electrical signal based on a psychoacoustic model; A signal-to-mask ratio calculation unit, the signal-to-mask ratio calculation unit is used to process the signal strength of the analog electrical signal and the masking threshold value to obtain the signal-to-mask ratio of the analog electrical signal; A noise control unit, the noise control unit is used to filter the analog electrical signal based on the signal-to-mask ratio of the analog electrical signal to obtain a noise-reduced electrical signal; The noise judgment unit is used to judge whether the signal-to-mass ratio of the noise reduction electrical signal is lower than a set threshold. If so, a framing operation is performed on the noise reduction electrical signal and the operation result is output as a digital audio signal. If not, a secondary filtering process is performed on the noise reduction electrical signal.

4. The intercom implementation system based on automatic speech recognition according to claim 3 is characterized in that: The audio repair module uses forward error correction technology to repair voice data packets lost during transmission in a data redundancy manner to obtain an audio signal to be processed, specifically including: Performing an XOR operation on each group of voice data packets in the digital audio signal to generate a parity check packet; Copying the data of the previous group of voice data packets to the first half of the next group of voice data packets, and outputting them as redundant data packets of the next group of voice data packets; When transmitting each group of voice data packets, a parity check packet and a redundant data packet of the group of voice data packets are additionally transmitted; The data receiving end detects the parity check packets of each group of voice data packets; Determine whether the parity check packets of adjacent groups of voice data packets have different attributes. If so, output the parity check packets with the same attributes corresponding to the voice data packets. If not, do not output; The lost voice data packet is repaired by querying the redundant data packets of a group of voice data packets after the lost position; The repaired voice data packets are aggregated and output as audio signals to be processed.

5. The intercom implementation system based on automatic speech recognition according to claim 4 is characterized in that: The audio framing module performs framing processing on the processed audio signal using an overlapping sliding window method to obtain at least two speech frames, specifically comprising: S101: Preset the frame length of the voice frame; S102: taking two-thirds of the frame length of the speech frame as the frame shift of the speech frame; S103: Selecting a Hamming window function as a sliding window function; S104: Aligning the starting position of the window with the starting position of the audio signal to be processed; S105: multiplying the window with the data at the corresponding position of the audio signal to be processed to achieve weighted processing and reduce the influence of signal mutation at the window boundary; S106: Number the weighted results and record them as speech frames; S107: Determine whether the window slides to the end of the audio signal to be processed. If so, output at least two speech frames. If not, slide the window forward along the time axis of the audio signal to be processed by one frame, and repeat step S105. The Hamming window function is: Where ω(n) is the function value of the Hamming window function of the nth signal, and a is the Hamming window parameter, which is N is the frame length of the speech frame and also the width of the Hamming window.

6. The intercom implementation system based on automatic speech recognition according to claim 5 is characterized in that: The endpoint detection module uses a double threshold decision method to detect the endpoints of the speech area in the speech frame, and separates the speech area data from the noise area data, specifically including: Calculate the short-time energy of each speech frame; Calculate the short-time zero-crossing rate of each speech frame; The first threshold is set to the short-time average energy of the speech frame; The second threshold is set to the short-time average energy of the noise; The third threshold is set to the short-time average zero-crossing rate of the noise; Obtain time points A and B with the same short-time average energy as the first threshold, and preliminarily determine that the start and end points of the speech segment are outside the AB segment; Search point A to the left and point B to the right to obtain point C where the left side of point A intersects with the second threshold for the first time and point D where the right side of point B intersects with the second threshold for the first time; Search leftward from point C and rightward from point D to obtain point E where the left side of point C intersects with the third threshold for the first time and point F where the right side of point D intersects with the third threshold for the first time; The EF segment of the output speech frame is the speech area data, and the remaining data is the noise area data; The short-time average energy is: In the formula, E n is the short-time average energy of the nth signal, m is the time sequence number of each speech frame, and x(m is the signal value at the mth moment; The short-time average zero-crossing rate is: In the formula, Z n is the short-time average zero-crossing rate of the nth signal, sgn(x) is the mathematical symbol function; The mathematical symbol function is: Where sgn(x) is a mathematical symbol function.

7. The intercom implementation system based on automatic speech recognition according to claim 6 is characterized in that: The model building module trains the speech recognition hidden Markov model through the speech area data and corresponding sentences in the information library, specifically including: Determining the number of states of speech zone data in the information base, wherein the states include phonemes and fine-grained speech units; Define the observation set as all statement output values ​​that the model may observe; The state transition probability matrix of each data in the calculation model from the current state to other states; Calculate the observation probability matrix of each data in the model to observe a specific observation value in the current state; Initialize the initial probability distribution of each data in the model in various states at the initial moment; The number of states, the set of observation values, the state transition probability matrix, the observation probability matrix and the initial probability distribution are used as five-tuple parameters to construct a hidden Markov model.

8. The intercom implementation system based on automatic speech recognition according to claim 7 is characterized in that: The speech recognition module inputs the speech area data obtained by processing the user's speech signal into the speech recognition hidden Markov model, and uses the Viterbi algorithm to obtain the sentence information corresponding to the user's speech signal, which specifically includes: S201: Acquire the data at each moment and each state of the voice zone data obtained by processing the user's voice signal; S202: Initialize the current time; S203: Calculate the probability of passing through each state up to the current moment, and output the state with the highest probability as the passing state; S204: Determine whether the current moment is the last moment of the voice zone data. If so, output all the passed state sequences as corresponding sentence information. If not, move the current moment back and return to step S203.

Images