CN114822574A - Human voice recognition and enhancement method, device and storage medium for speech equipment - Google Patents

Abstract

Embodiments of the present invention disclose a method, device and storage medium for human voice recognition and enhancement for speech equipment, relate to the field of communication technologies, and can be applied to the emergency rescue where the voice of a walkie-talkie needs to be amplified and enhanced in a mask scene. The present invention includes: performing echo cancellation on the collected sound signal through the NLMS algorithm, wherein the collected sound signal includes environmental noise and voice signal; performing spectrum analysis on the sound processed in step 1, and extracting features that conform to the human voice sound signal; enhance the sound signal conforming to human voice characteristics through IIR; send the enhanced sound signal to the intercom module.

Description

Human voice recognition and enhancement method, device and storage medium for speech equipment

technical field

The present invention relates to the field of communication technologies, and in particular, to a method, device and storage medium for human voice recognition and enhancement for speech equipment.

Background technique

At present, walkie-talkies are widely used as a means of voice communication in the on-site coordination of emergency rescue, and shoulder microphone devices are usually used in the execution of tasks. When using the walkie-talkie, an additional shoulder microphone is required to facilitate real-time communication.

Therefore, rescuers need to be equipped with a large number of communication equipment, the use environment is also more complex, and the sound quality and denoising effect of calls are poor. In addition, due to the special application scenarios of walkie-talkies, ultra-long standby and reliability are the most important requirements, so such communication equipment generally adopts a low-power, low-performance but relatively reliable system architecture. The ability is low, so many of the current algorithm functions for sound optimization are not suitable for implementation on this type of architecture. In addition, there is also a lack of effective processing methods for scenarios in which it is necessary to amplify and enhance the sound of the walkie-talkie in the mask in emergency rescue.

SUMMARY OF THE INVENTION

Embodiments of the present invention provide a human voice recognition and enhancement method, device and storage medium for voice equipment, so as to be easily applied to scenarios in which the voice of a walkie-talkie needs to be amplified and enhanced in a mask in emergency rescue.

To achieve the above object, the embodiments of the present invention adopt the following technical solutions:

In a first aspect, a method provided by an embodiment of the present invention includes:

Step 1, performing echo cancellation on the collected sound signal by NLMS algorithm, wherein, the collected sound signal includes environmental noise and voice signal;

Step 2, performing spectrum analysis on the sound processed in step 1, and extracting a sound signal that conforms to the characteristics of the human voice;

Step 3. Perform enhancement processing on the sound signal conforming to human voice characteristics through IIR;

Step 4. Send the enhanced sound signal to the intercom module.

The method further includes: playing and recording the enhanced sound signal; and then using the recorded sound data to perform echo cancellation on the collected sound signal through the NLMS algorithm.

Step 1 includes: inputting the collected sound signal into the codec for encoding; storing the encoded sound signal in the speech signal and noise buffer area, and storing the recorded sound data in the reference buffer area, wherein the processor converts the sound signal from the speech signal and the noise buffer to the reference buffer area. Extract data from noise buffer and echo reference buffer.

m buffer areas are established in the voice signal and noise buffer areas, wherein when the nth buffer area is recording the sound signal, the processor is processing the [(n+m-1)mod m]th buffer area at the same time and play the data of the [(n+m-2)mod m]th buffer area at the same time, wherein, the [(n+m-1)mod m] buffer area represents the previous buffer area of the nth buffer area , [(n+m-2)mod m] buffer area represents the first two buffer areas of the nth buffer area.

The performing echo cancellation on the collected sound signal by using the NLMS algorithm includes: after the collected sound signal is processed by the codec, performing NLMS normalization filtering with the system output signal as a reference, wherein the collected sound signal is processed by the codec. After the codec is processed, the output data type is a 24-bit precision signed integer; then the main signal is subjected to NLMS normalization filtering with the noise signal as a reference to obtain a preliminarily processed denoising signal; The denoised signal is notch filtered, where IIR work is performed at the specified frequency at compile time.

In the process of performing the spectrum analysis, it includes: determining the sound threshold by the maximum power density (which can also be referred to as single tone intensity); and, distinguishing the sound signal belonging to the wind noise and the sound signal belonging to the human voice by the signal-to-noise ratio; and , and distinguish the sound signal belonging to the respiration valve resonance from the sound signal belonging to the spray resonance by the power spectrum concentration.

In the process of IIR enhancement processing, it includes: collecting 1024 points as a cache each time, taking out 4 caches to form a set of 4096 sampling points according to the chronological order of recording the caches; using the sampling point set to activate the voice The degree statistics function is enhanced.

The IIR is a five-segment sound equalizer; through the IIR, according to different sound frequency sections, the sound signal conforming to human voice characteristics is enhanced, wherein the sound frequency section includes: 40-200Hz subwoofer, 200-800Hz mid-bass, 800-1600Hz mid-high, 1600-4000Hz super high and over 4000Hz pan treble.

In a second aspect, the device provided by the embodiment of the present invention includes:

a preprocessing module for performing echo cancellation on the collected sound signal through the NLMS algorithm, wherein the collected sound signal includes environmental noise and voice signals;

The analysis module is used to perform spectrum analysis on the sound processed in step 1, and extract the sound signal conforming to the characteristics of human voice;

The enhancement module is used to enhance the sound signal conforming to the human voice through IIR;

The transmission module is used to send the enhanced sound signal to the intercom module.

In a third aspect, an embodiment of the present invention provides a storage medium storing a computer program or instruction, and when the computer program or instruction is executed, the method in this embodiment is implemented.

In the method, device, and storage medium for human voice recognition and enhancement for speech equipment provided by the embodiments of the present invention, whether or not there is an auxiliary microphone, echo suppression is effective at all times. After the collected signal is converted by A/D, the output data type is a signed integer with 24-bit precision. After the noise and the main signal are normalized and filtered by NLMS with the system output signal as the reference, the noise signal is used as the reference. The main signal is subjected to NLMS to obtain the denoised signal after preliminary processing. The denoised signal will be subjected to IIR notch filtering at the frequency specified at compile time to remove noise at specific frequencies. Therefore, it can be conveniently applied to the scene in which it is necessary to amplify and enhance the sound of the walkie-talkie in the mask in emergency rescue.

Description of drawings

In order to illustrate the technical solutions in the embodiments of the present invention more clearly, the following briefly introduces the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without any creative effort.

1 is a schematic diagram of a system architecture provided by an embodiment of the present invention;

2 and 3 are schematic diagrams of principles provided by an embodiment of the present invention;

4 is a schematic flowchart of human voice recognition and enhancement provided by an embodiment of the present invention;

5 is a schematic flowchart of a spectrum comparison provided by an embodiment of the present invention;

6 is a schematic flowchart of a method provided by an embodiment of the present invention;

FIG. 7 is a schematic flowchart of a possible spectrum analysis provided by an embodiment of the present invention.

Detailed ways

In order to make those skilled in the art better understand the technical solutions of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Hereinafter, embodiments of the present invention will be described in detail, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, but not to be construed as a limitation of the present invention. It will be understood by those skilled in the art that the singular forms "a", "an", "the" and "the" as used herein can include the plural forms as well, unless expressly stated otherwise. It should be further understood that the word "comprising" used in the description of the present invention refers to the presence of stated 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 we refer to an element 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. Furthermore, "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. It will be understood by those skilled in the art 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 general dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and, unless defined as herein, are not to be taken in an idealized or overly formal sense. explain.

An embodiment of the present invention provides a human voice recognition and enhancement method for a voice device, as shown in FIG. 6 , including:

Step 1. Perform echo cancellation on the collected sound signal through the NLMS algorithm.

Among them, the collected sound signal includes ambient noise and voice signal. The NLMS (Normalized Least Mean Square, Normalized Least Mean Square) algorithm is a real-time processing algorithm that filters out the noise of the target signal under the premise of a reference with known noise signal characteristics. It has the advantages of fast convergence and low load.

Step 2: Perform spectrum analysis on the sound processed in Step 1, and extract a sound signal conforming to the characteristics of the human voice.

Step 3. Perform enhancement processing on the sound signal conforming to the characteristics of the human voice through the IIR.

Among them, IIR refers to a digital filter, and a digital filter is a linear time-invariant system for filtering digital signals. Digital filtering is essentially an operation process, which realizes the operation and processing of the signal. The input digital signal (digital sequence) is transformed into the output digital sequence through a specific operation. Therefore, a digital filter is essentially a digital calculation process that completes a specific operation, and can also be understood as a computer. The convolution and difference equations that describe the relationship between the output and the input of a discrete system only provide the digital signal filter with an operation rule, so that it can complete the processing of the input data according to this rule.

Step 4. Send the enhanced sound signal to the intercom module.

In this embodiment, the method further includes: playing and recording the enhanced sound signal, recording the enhanced sound signal so as to be used as a reference for echo cancellation in the next cycle, wherein the specific sound data is recorded in the echo reference buffer .

Then use the recorded sound data, and perform echo cancellation on the collected sound signal through the NLMS algorithm. For example: play the enhanced sound signal 1, and store the sound signal 1 in the echo reference buffer area, where the sound signal 1 refers to the enhanced sound after noise cancellation and echo cancellation, which can be played and sent to the intercom module the sound of. If there are main and auxiliary microphones, the sound signal 2 picked up by the main microphone includes the voice signal 2-1, the echo signal 3 (that is, the echo generated after the above-mentioned sound signal 1 is played) and the environmental noise 4. After the A/D conversion of the sound signal 2 Stored in the voice signal and noise buffer area, the voice signal 5 picked up by the secondary microphone includes echo signal 3 and ambient noise 4, and the voice signal 5 is also stored in the voice signal and noise buffer area after A/D conversion.

Among them, the process of performing echo cancellation is specifically adopted in the manner as shown in FIG. 4 , regardless of whether there is an auxiliary microphone or not, the echo suppression is effective for the whole period of time. For example, performing echo cancellation on the collected sound signal through the NLMS algorithm includes: converting the collected signal into a 24-bit precision signed integer after A/D conversion, and taking the system output signal as a reference after the noise and the main signal pass through After the NLMS is normalized and filtered, the noise signal is used as a reference to perform NLMS on the main signal to obtain a preliminarily processed denoised signal. Wherein, the system output signal can be the sound signal 1 played by the system, wherein, if the dual microphone (main and sub-microphone) architecture as shown in FIG. Voice signal, among which the voice signal strength collected by the main microphone is relatively high because the main microphone is close to the user's mouth and nose; the sound signal collected by the secondary microphone also includes echo signals, environmental noise signals and voice signals, but due to the distance between the secondary microphone and the user The mouth and nose are farther away, so the strength of the collected voice signal is not as good as that of the main microphone. As an infinite impulse response filter, the IIR works at the specified frequency at compile time. If the signal needs to filter out the noise of a specified frequency, it will be sent to the parameter specified at the time of programming after it is processed by NLMS. IIR notch. The specific processing principle can be shown in FIG. 2 , wherein: S1 : The signal picked up by the main microphone and the aforementioned sound signal 1 are processed by the NLMS algorithm, and the echo signal 3 can be eliminated. S2: The signal picked up by the secondary microphone and the aforementioned sound signal 1 are processed by the NLMS algorithm, and the echo signal 3 can be eliminated. S3: Perform NLMS algorithm processing on the two signals of S1 and S2 after the echo signal 3 is eliminated to obtain a voice signal 2-1, and then perform IIR notch filtering on the voice signal 2-1 to eliminate noise of a specific frequency (S4) , and then perform FFT transformation and spectrum analysis (S5). The spectrum analysis is mainly to analyze human voice, spray wheat, breathing valve, and white noise. If there is a human voice, enhance it (S6), and then send it to the intercom module and, or local Play (S7).

Another example: if there is no secondary microphone, NLMS is performed on the main signal using the ambient noise signal as a reference. As shown in Figure 3, if there is no secondary microphone, the enhanced sound signal 1 is played, and the sound signal 1 Stored in the echo reference buffer area, the sound signal 2 picked up by the main microphone includes the voice signal 2-1, the echo signal 3 and the ambient noise 4, the sound signal 2 is stored in the voice signal and noise buffer area after A/D conversion, S01: main The signal picked up by the microphone and the aforementioned sound signal 1 are processed by the NLMS algorithm, and the echo signal 3 can be eliminated. S02: Then perform IIR notch filtering on the signal obtained from S01 to eliminate noise at a specific frequency, and then perform FFT transformation and spectrum analysis (S03). In this embodiment, the white noise belongs to the noise without obvious spectral characteristics, which is caused by the error generated during AD conversion). S05).

At present, the reference signal is the output signal after the voice processing is completed, but it needs to be connected to the output signal of the intercom module in the following walkie-talkies. In this embodiment, regardless of whether there is an auxiliary microphone, the echo suppression takes effect at all times. After the collected signal is converted by A/D, the output data type is a signed integer with 24-bit precision. After the noise and main signal are normalized and filtered by NLMS with the system output signal as a reference, the denoising after preliminary processing is obtained. Signal, optionally, if a dual-microphone architecture is adopted, NLMS can also be performed on the main signal using the noise signal as a reference. The denoised signal will be subjected to IIR notch filtering at the specified frequency at compile time to remove noise at a specific frequency (such as single-frequency mechanical sound), and then sent to CFFT to separate frequency features. Practically applied to the walkie-talkie, it can realize the voice-activated PTT function without cooperating with the shoulder microphone, and the call effect is good.

In this embodiment, step 1 includes: inputting the collected sound signal into a codec (for example, WM8978) for encoding, storing the encoded sound signal in the voice signal and noise buffer area, and storing the recorded sound data in the reference buffer Area. Among them, a human voice recognition and enhancement method is realized by analyzing the difference in frequency characteristics between human voice and natural noise. As shown in Figure 1, the encoding of the audio is completed by WM8978, and the data stream is transmitted in full duplex with SAI. The signal is processed in real time with ARM-DSP. Specifically, the processor extracts data from the voice signal and the noise buffer and the echo reference buffer, for example, the encoded and decoded voice signal 2 and voice signal 5 are stored in the voice signal and noise buffer, and the voice signal 1 is stored in the echo Reference buffer.

In this embodiment, m buffer areas are established in the voice signal and noise buffer areas, wherein when the nth buffer area is recording the sound signal, the processor is processing the [(n+m-1)th mod m] data in the buffer area, while playing the data in the [(n+m-2)mod m]th buffer area, where the [(n+m-1)mod m] buffer area represents the nth buffer area The previous buffer area of , [(n+m-2)mod m] buffer area represents the first two buffer areas of the nth buffer area. Among them, "mod" means taking the remainder in the industry, and m takes the value of 4 in the preferred solution. Among them, when a certain buffer area is recording a sound signal, the processor is processing the data of the previous buffer area at the same time, and plays the data in the previous two buffer areas (sound signal 1) at the same time. The way of cache effectively reduces the burden on the processor. In this embodiment, full-duplex is used and the same buffer is used for input and output, so recording and playback must end at the same time, because recording is received by the master device, playback is sent by the slave device, and the playback data stream is synchronized with the recording data stream. , so the cache address of the playback data should be mounted to the DMA before the recording cache is mounted. By using multiple ring buffers, assuming that the sound is currently being recorded in the nth buffer, the data in the [(n+m-1)modm]th buffer must be being processed, and if the sound is being played, there may be The data of the [(n+m-1)modm] or [(n+m-2)modm] (lowest delay mode) block.

Optionally, the data in the [(n+m-2)modm] and [(n+m-3)modm]th buffer areas can also be played at the same time, so as to prevent partial voice loss and obtain a better playback effect.

Specifically, the input and output use the same buffer, which means that m buffer areas are divided into the voice signal and noise buffer areas. When a certain buffer area is recording a sound signal, the processor is processing the previous buffer area at the same time. The data in the buffer area, play the data in the previous two buffer areas (sound signal 1) at the same time, and a buffer area mentioned above is recording the sound signal, which is the input, and play the data in the previous two buffer areas at the same time. data (sound signal 1), and the data in the buffer area is constantly updated, so the input and output use the same buffer.

In this embodiment, the manner of performing spectrum analysis, as shown in FIG. 5 , includes: performing CFFT processing on the sound signal processed in step 1, and converting the sound signal from a time-domain signal to a frequency-domain signal. In the vocal region of the sound signal, the maximum energy concentration frequency of the human voice region is determined, and the energy intensity at the maximum energy concentration frequency of the human voice region is recorded. The amplitude-frequency characteristic information of the vocal area is acquired, where the amplitude-frequency characteristic information includes: the average value and the standard deviation of the amplitude-frequency of the vocal area. In the noise region of the sound signal, the maximum energy concentration frequency of the noise region is determined, and the energy intensity at the maximum energy concentration frequency of the noise region is recorded. Using the effective sound intensity, dispersion, speech signal-to-noise ratio and total signal-to-noise ratio to compare with the reference data, it is preliminarily judged whether a buffer contains human voice. Among them, the effective sound intensity = the maximum single tone energy intensity in the vocal area. Dispersion = standard deviation of the amplitude and frequency of the vocal area / average value of the amplitude and frequency of the vocal area. Speech signal-to-noise ratio=maximum single tone energy intensity in vocal area/average amplitude frequency in vocal area. The total signal-to-noise ratio = the maximum single-tone energy intensity in the vocal area/the maximum single-tone energy intensity in the noise area. If it is preliminarily determined that the human voice is contained, the TTL stabilizer is triggered once, wherein the number of triggering times of the TTL stabilizer is positively correlated with the confidence. In Fig. 5, "&" represents an AND gate, and "&." represents a NAND gate.

In the specific implementation, the process shown in Figure 7 can be used, in which: record buffering and CFFT (Complex Fast Fourier Transform, complex fast Fourier Transform) processing, convert the signal from the time domain signal to the frequency domain signal; find out the vocal area The maximum energy concentration frequency and record the energy intensity at this frequency; calculate the amplitude-frequency characteristics of the human voice area, and obtain its corresponding average value and standard deviation; find the maximum energy concentration frequency in the noise area and record it at this frequency energy intensity; compare the dimensionless parameters such as effective sound intensity, dispersion, speech signal-to-noise ratio, and total signal-to-noise ratio with relevant data (experimental values), and preliminarily determine whether a buffer contains human voice; For human voice, trigger the TTL stabilizer once. Since the voice itself has continuity, the TTL stabilizer will automatically and stably identify the output result according to the scene and the number of triggers to ensure the confidence in it (that is, call the voice activity statistics function to enhance the signal and output).

In this embodiment, the process of performing spectrum analysis on the processed sound includes:

The sound threshold is determined by the maximum power density (also called tone intensity). And, the sound signal belonging to the wind noise and the sound signal belonging to the human voice are distinguished by the signal-to-noise ratio. And, the sound signal belonging to the respiration valve resonance is distinguished from the sound signal belonging to the spray resonance by the power spectrum concentration. The sound signals other than the sound signals belonging to the human voice are essentially noise.

The sound threshold can be understood as a volume threshold, and the volume threshold is determined by the maximum single tone intensity corresponding to the collected sound signal. A single tone is specifically a single frequency. In practical applications, there are three kinds of noises that need to be distinguished for the situation of sound amplification in the mask: 1. White noise caused by the error generated during AD conversion; 2. Air flow whistling caused by the breathing valve, generally concentrated in the Several fixed frequency bands; 3. Wind noise caused by airflow hitting the microphone when breathing.

The speech recognition in the industry is mostly semantic recognition and voice recognition in a quiet environment. Since the amplification in the mask has high requirements on the accuracy of noise filtering, traditional time-frequency analysis methods such as Morlet wavelet transform real-time processing. The computing power requirements are too high, so it is not suitable to work on the ARM-CortexM4F processor with a main frequency of 60-120Mhz. For fast spectrum analysis techniques, the first step is to convert the signal from the time domain to the frequency domain. Under the sampling rate of 48Khz, the kernel time occupied by the fast Fourier transform is generally about 33%, and the efficiency of the processor for the full time domain fundamental wave sinθ transform is much higher than that of the morlet wavelet. The energy of human voice is mainly concentrated in 160-2000Hz, so we can focus on this frequency analysis. Now use mathematical features to analyze it. First of all, white noise is noise without obvious spectral characteristics, its energy distribution is uniform, and there is no concentrated segment. Its spectral dispersion is very high, which is reflected in the normal distribution diagram as a very steep t distribution curve, in which the energy is all concentrated near the expected value (mean), so it can be calculated according to the standard deviation of the spectrum σ=sqrt(((x1-x) ^2+(x2-x)^2+...(xn-x)^2)/n) to identify white noise. This function can be achieved by setting a reasonable threshold. For airflow impact howling, the concentrated energy of this frequency band can be weighted down. Due to the concentration of energy in the airflow impact whistle, the impact noise can be distinguished by limiting the sudden change of the amplitude-frequency curve and observing the characteristics of μ and σ in the t distribution. When the airflow impacts the microphone, the spectrum of wind noise is similar to white noise after analysis, so it will not be described in detail.

The human voice is a special sound source. Due to the structural characteristics of the vocal cords, we cannot emit a single tone (ie, a single frequency sound). The so-called tone change is the effect of multiple fundamental frequency synchronous frequency shifts. From the distribution point of view, its spectrum has large energy and more overtones, which makes the amplitude-frequency distribution more scattered. If the normal distribution X～N(μ,σ ² ) is taken, it can be seen that the curve is relatively steep, but there is no sudden change . Therefore, after the change, the appropriate standard deviation can be selected to distinguish the vocal frequency band. Specifically, the vocal timbre is divided into the following sections: 40-200Hz super bass, 200-800Hz mid-bass, 800-1600Hz mid-high, 1600-4000Hz super high, 4000Hz+pan treble. Generally speaking, the sound can be processed with a five-band IIR lattice filter: 1) Subwoofer - attenuated to avoid dullness. 2) Mid-bass - Amplify, make the sound clear. 3) Middle and high-pitched-amplification to make the sound transparent. 4) Super high-pitched - attenuation, no obvious effect but can filter out mechanical noise. 5) Pan treble - notch, no obvious effect, isolation of high frequency noise. Among them, the statistical function of voice activity is used to compare the current spectral features with a given threshold to obtain an activity weight, and then make a comprehensive judgment based on the recent activity weights. For example: the spectrum comparison process shown in Figure 5: the main energy of human voice is concentrated in the fundamental wave of 160-2000Hz, with high signal-to-noise ratio (ratio of peak to average) and low energy concentration (generally falls in the standard normal state) distribution (μ,σ ² ) in the range of 1σ to 2σ), and the energy of wind noise is extremely dispersed, and the signal-to-noise ratio is extremely low, which can be filtered out by the signal-to-noise ratio. Although the overall energy level of the respiration valve and the spray-mike resonance also obeys the t distribution and has a relatively high signal-to-noise ratio, most of the energy peaks are relatively weak and the energy level is close to the center point μ (less than 1σ), and the overall standard deviation is small. After the above analysis, the following conclusions are drawn: a. The sound threshold can be judged by the maximum power density (also called single tone intensity). b. Wind noise and human voice can be judged by the signal-to-noise ratio. c. The breathing valve and the microphone sound can be judged by the power spectrum concentration σ/μ.

In this embodiment, the process of performing the enhancement processing during the IIR enhancement processing includes: collecting 1024 points as a cache each time, and taking out 4 caches to form a set of 4096 sampling points according to the chronological order of recording the caches . Among them, for direct amplification (that is, local playback is required), if the sampling time is too long, the delay will be too large. When the output delay (playback delay) exceeds 100ms, it will seriously interfere with the user's voice auditory feedback, thereby affecting the user's voice. However, if the sampling points are too few, the accuracy of the system judgment will be affected. Therefore, the scheme of sampling circular storage + chain loading is used, 1024 points are collected each time as a cache, and then 4 caches are taken out according to first-in-first-out to form 4096 recent sampling points, each cache corresponds to a period, according to multiple sampling points The signal characteristics of the period, analyze the possibility of human voice in the continuous period.

Using the set of sampling points, enhancement is performed through a statistical function of speech activity. Among them, for direct sound reinforcement, if the sampling time is too long, the delay will be too large, and when the output delay is greater than 100ms, it will seriously interfere with the user's language auditory feedback, thereby affecting the user's language ability. However, if the sampling points are too few, the accuracy of the system judgment will be affected. Therefore, the scheme of sampling cycle storage + chain loading is used, 1024 points are collected each time as a cache, and then 4 caches are taken out according to first-in-first-out and combined into 4096 recent sampling points, combined with the voice activity statistical function, for speech recognition with play.

The voice activity statistical function can be modeled by those skilled in the art. The functions of the voice activity statistical function include: that is, the human voice is more smooth, continuous and natural, because if there is no such function, after processing the voice, it will eliminate noise, The enhanced sound signal 1 after the echo cancellation will appear intermittent and incoherent, and the incoherent phenomenon is caused by spectrum analysis. In this embodiment, the "voice activity" is a professional term in the industry.

In this embodiment, the IIR is a five-segment sound equalizer. Specifically, the IIR can be used to enhance the sound signal conforming to the characteristics of the human voice according to different sound frequency sections, wherein the sound frequency section includes: 40-200Hz super bass, 200-800Hz mid-bass, 800- 1600Hz mid treble, 1600-4000Hz super treble and 4000Hz and above pan treble.

This embodiment also provides a human voice recognition and enhancement device for speech equipment, including:

The preprocessing module is configured to perform echo cancellation on the collected sound signal through the NLMS algorithm, wherein the collected sound signal includes environmental noise and voice signal.

The analysis module is used for performing spectrum analysis on the sound processed in step 1, and extracting the sound signal conforming to the characteristics of the human voice.

The enhancement module is used to enhance the sound signal conforming to human voice characteristics through IIR.

The transmission module is used to send the enhanced sound signal to the intercom module.

An embodiment of the present invention further provides a storage medium storing a computer program or instruction, and when the computer program or instruction is executed, the method in this embodiment is implemented.

Each embodiment in this specification is described in a progressive manner, and the same and similar parts between the various embodiments may be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, as for the device embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and for related parts, please refer to the partial descriptions of the method embodiments. The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical scope disclosed by the present invention can easily think of changes or substitutions. All should be included within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims (11)

1. A human voice recognition and enhancement method for voice equipment is characterized by comprising the following steps in the processing process of each period:

step 1, carrying out echo cancellation on the collected sound signals through an NLMS algorithm;

step 2, carrying out spectrum analysis on the sound processed in the step 1, and extracting a sound signal according with the human voice characteristic;

step 3, enhancing the sound signals according with the human voice characteristics through the IIR;

and 4, sending the enhanced sound signal to the talkback module.

2. The method of claim 1, further comprising:

playing and recording the sound signal after enhancement processing;

and then, recorded sound data is utilized to perform echo cancellation on the collected sound signals through an NLMS algorithm.

3. The method of claim 2, wherein step 1 comprises:

inputting the collected sound signal into a coder for coding;

the encoded sound signal is stored in a speech signal and noise buffer, and the recorded sound data is stored in a reference buffer, wherein the processor extracts the data from the speech signal and noise buffer and the echo reference buffer.

4. The method of claim 3, wherein m buffers are created in the speech signal and noise buffer, wherein when the nth buffer is recording the audio signal, the processor is simultaneously processing data of the [ (n + m-1) mod m ] buffer and simultaneously playing data of the [ (n + m-2) mod m ] buffer, wherein the [ (n + m-1) mod m ] buffer represents a previous buffer of the nth buffer, and the [ (n + m-2) mod m ] buffer represents a previous buffer of the nth buffer.

5. The method of claim 1, wherein the performing echo cancellation on the collected sound signal by the NLMS algorithm comprises:

after the collected sound signals are processed by the coder and decoder, NLMS normalization filtering is carried out by taking system output signals as reference, wherein the type of output data is 24-bit precision signed integer after the collected sound signals are processed by the coder and decoder, and the system output signals are sound signals subjected to enhancement processing in the previous period;

performing NLMS normalization filtering on the main signal by taking the noise signal as a reference to obtain a noise elimination signal after primary processing;

notch filtering the noise-canceled signal through an IIR, wherein the IIR works when being executed to specify a frequency in coding.

6. The method of claim 1, wherein the spectral analysis performed comprises:

performing CFFT processing on the sound signal processed in the step 1, and converting the sound signal from a time domain signal into a frequency domain signal;

determining the maximum energy concentration frequency of a human vocal area in the human vocal area of the voice signal, and recording the energy intensity under the maximum energy concentration frequency of the human vocal area;

acquiring amplitude-frequency characteristic information of a human vocal region, wherein the amplitude-frequency characteristic information comprises: average and standard deviation of amplitude and frequency of human vocal tract;

in a noise area of a sound signal, determining the maximum energy concentration frequency of the noise area, and recording the energy intensity of the maximum energy concentration frequency of the noise area;

comparing the effective sound intensity, the dispersion, the voice signal-to-noise ratio and the total signal-to-noise ratio with reference data to preliminarily judge whether a cache contains human voice;

and if the voice is preliminarily judged to be contained, triggering the TTL stabilizer once, wherein the triggering times of the TTL stabilizer are positively correlated with the confidence coefficient.

7. The method of claim 6, wherein the effective sound intensity is the maximum tone energy intensity of the human vocal tract;

the dispersion is the standard deviation of the amplitude frequency of the vocal region/the average value of the amplitude frequency of the vocal region;

the voice signal-to-noise ratio is the average value of the maximum single tone energy intensity in the human voice area/the amplitude frequency in the human voice area;

the total signal-to-noise ratio is the maximum tone energy intensity in the human voice region/the maximum tone energy intensity in the noise region.

8. The method of claim 1 or 7, wherein in performing the spectral analysis, the method comprises:

determining a sound threshold by a maximum power density (single tone strength);

and, distinguishing sound signals belonging to wind noise from sound signals belonging to human voice by signal-to-noise ratio;

and, distinguishing between the acoustic signal belonging to the resonance of the respiratory valve and the acoustic signal belonging to the resonance of the jet microphone by the concentration of the power spectrum.

9. The method of claim 1, wherein during the IIR enhancement processing, the method comprises:

collecting 1024 points as a cache every time, and taking out 4 caches to form 4096 sampling point sets according to the time sequence of the caches;

enhancing through a voice activity statistical function by utilizing the sampling point set;

the IIR is a five-segment type sound equalizer;

carrying out enhancement processing on a sound signal conforming to human sound characteristics according to different sound frequency sections through an IIR, wherein the sound frequency sections comprise: ultralow pitch at 40-200Hz, middle and low pitch at 200-800Hz, middle and high pitch at 800-1600Hz, ultrahigh pitch at 1600-4000Hz, and alto at 4000Hz or above.

10. A human voice recognition and enhancement apparatus for a speech device, comprising:

the preprocessing module is used for carrying out echo cancellation on the collected sound signals through an NLMS algorithm;

the analysis module is used for carrying out spectrum analysis on the sound processed in the step 1 and extracting a sound signal which accords with the characteristics of human voice;

the enhancement module is used for enhancing the sound signals conforming to the human voice characteristics through the IIR;

and the transmission module is used for transmitting the enhanced sound signal to the talkback module.

11. A storage medium, storing a computer program or instructions which, when executed, implement the method of any one of claims 1 to 8.

Images