An AI voice analysis method and system for IoT card control

Abstract

The present invention relates to the technical field of AI voice analysis controlled by the Internet of Things, and specifically discloses an AI voice analysis method and system for Internet of Things card control, which collects user voice commands through a high-precision audio input device to obtain original audio data, performs time domain and frequency domain analysis on the collected original audio data, and obtains a voice clarity score through comprehensive analysis, calculates the signal-to-noise ratio under different noise intensities, determines whether the noise will affect the voice clarity, and dynamically adjusts noise suppression parameters based on the influence of noise on voice clarity to optimize the voice signal quality, uses a multi-channel audio input device to collect voice signals, calibrates each channel, and performs time domain and frequency domain analysis, calculates the overall voice clarity score based on the analysis results of each channel, and dynamically adjusts the noise suppression parameters of each channel as needed.

Description

Technical Field

[0001] The present invention relates to the field of AI voice technology for Internet of Things control, and in particular to an AI voice analysis method and system for Internet of Things card control. Background Art

[0002] With the continuous development of IoT technology and AI speech analysis technology, AI speech analysis methods for IoT card control will be applied in more fields. For example, the demand for speech recognition technology and speech interaction in application scenarios such as smart home and smart office is increasing. Speech recognition technology is an important part of human-computer interaction. However, existing speech recognition technology faces a common problem: in complex noise environments, the clarity of speech signals is often seriously disturbed, resulting in a decrease in speech recognition accuracy and a poor user experience.

[0003] In real environments, the presence of various noises such as background music, crowd conversations, and machine operation sounds seriously affects the quality of speech signals; these noises not only reduce the recognizability of speech, but also cause the speech recognition system to make misjudgments; in some cases, it is necessary to process the speech signals of multiple speakers at the same time, which puts higher requirements on the signal processing algorithm; traditional single-channel speech processing technology is often unable to effectively cope with speech recognition tasks in multi-source environments; in order to provide a smooth interactive experience, the speech recognition system must be able to complete the acquisition, processing and recognition of speech signals in a short time; this means that efficient signal processing algorithms are needed to ensure the response speed of the system.

[0004] In response to the above-mentioned technical problems, AI speech analysis methods can effectively improve the clarity of speech signals, dynamically adjust noise suppression parameters, and support multi-channel audio signal processing, so as to improve the accuracy and robustness of speech recognition systems, optimize user experience, and promote the widespread application of smart devices in more fields.

[0005] Based on this, the present invention proposes an AI speech analysis method and system for IoT card control, which realizes high-definition speech signal acquisition and processing in complex noise environments through comprehensive time domain and frequency domain analysis, noise suppression technology and dynamic parameter adjustment. Summary of the invention

[0006] The purpose of the present invention is to provide an AI voice analysis method and system for IoT card control, which first includes collecting user voice commands and obtaining original audio data, performing time domain and frequency domain analysis on the original audio data, and comprehensively obtaining a voice clarity score; analyzing the influence of noise on voice clarity, and obtaining a target signal-to-noise ratio interval and a target voice signal clarity score interval; based on the target voice signal clarity score interval, calculating a noise suppression gain adjustment interval; using a multi-channel audio input device to collect voice signals, and analyzing the multi-channel voice signals, and dynamically adjusting the noise suppression parameters of each channel according to the analysis results. Specifically, by calculating the distortion index D and the spectral smoothness S, and combining them with the preset weight coefficient, the speech clarity score C is obtained, and C is compared with the preset clarity score threshold to distinguish between high-definition and low-definition speech signals; by calculating the signal-to-noise ratio SNR, the signal-to-noise ratio ranges of high-definition and low-definition speech signals are integrated to determine the target signal-to-noise ratio range; based on the minimum clarity score Cmin under the maximum signal-to-noise ratio state, the distortion index D and the spectral smoothness S are obtained, and the noise suppression gain G is calculated by the formula, and the noise suppression gain is dynamically adjusted to optimize the speech signal clarity; in a multi-channel environment, by calibrating each channel to ensure data consistency, the time domain and frequency domain features of each channel are extracted, and the signal fusion algorithm is combined to generate a comprehensive speech clarity score, and the noise suppression parameters are dynamically adjusted according to the analysis results of the multi-channel signal, thereby optimizing the overall speech clarity and improving the reliability of voice control and user experience.

[0007] The purpose of the present invention can be achieved through the following technical solutions:

[0008] An AI voice analysis method for IoT card control includes the following steps:

[0009] S1: Collect the user's voice commands and obtain the original audio data;

[0010] S2: Perform time domain analysis and frequency domain analysis on the original audio data, and obtain the speech intelligibility score through comprehensive analysis;

[0011] S3: Analyze the influence of noise on speech clarity, and obtain the target signal-to-noise ratio range and the target speech signal clarity score range;

[0012] S4: Calculate the noise suppression gain adjustment interval based on the target speech signal clarity score interval;

[0013] S5: Use a multi-channel audio input device to collect voice signals, analyze the multi-channel voice signals, and dynamically adjust the noise suppression parameters of each channel according to the analysis results.

[0014] As a further solution of the present invention: the process of obtaining the speech intelligibility score C is:

[0015] Process the distortion index D and spectrum smoothness S;

[0016] The weight of the distortion index D is assigned to w1, and the weight of the spectrum smoothness S is assigned to w2, and both w1 and w2 are greater than 0;

[0017] By formula The speech intelligibility score C is calculated.

[0018] As a further solution of the present invention: the process of obtaining the distortion index D is:

[0019] By formula The distortion index D is calculated;

[0020] Where N is the number of sample points, is the i-th sample point of the original signal, and yi is the i-th sample point of the processed signal.

[0021] As a further solution of the present invention: the process of obtaining the spectrum smoothness S is:

[0022] By formula The spectrum smoothness S is calculated;

[0023] Where M is the number of components in the spectrum, and are the jth and j+1th components of the spectrum respectively.

[0024] As a further solution of the present invention: the target signal-to-noise ratio interval acquisition process is: through the formula The signal-to-noise ratio SNR is calculated;

[0025] in, is the i-th sample point of the original speech signal, is the i-th sample point of the noise signal, and N is the number of sample points.

[0026] The clarity score of each speech signal Compare with clarity score threshold;

[0027] A plurality of high-definition speech signals and a plurality of low-definition speech signals are obtained respectively;

[0028] For all high-definition speech signals, the corresponding signal-to-noise ratio is calculated, and the obtained signal-to-noise ratios are integrated in order of magnitude to obtain the signal-to-noise ratio interval corresponding to the high-definition speech signal, which is recorded as the high-definition low signal-to-noise ratio interval;

[0029] For all low-definition speech signals, the corresponding signal-to-noise ratio is calculated, and the obtained signal-to-noise ratios are integrated in order of magnitude to obtain the signal-to-noise ratio interval corresponding to the low-definition speech signal, which is recorded as the low-definition high signal-to-noise ratio interval;

[0030] If there is an intersection between the high-definition low signal-to-noise ratio interval and the low-definition high signal-to-noise ratio interval, the minimum signal-to-noise ratio in the high-definition low-noise ratio interval is obtained and recorded as , get the minimum signal-to-noise ratio in the low-resolution and high-signal-to-noise ratio range and record it as ,Will Recorded as the first target signal-to-noise ratio interval;

[0031] If the low-resolution high-SNR interval completely contains the high-resolution low-SNR interval, then the minimum SNR in the high-resolution low-SNR interval is obtained and recorded as , the maximum signal-to-noise ratio in the high-definition low signal-to-noise ratio range is obtained and recorded as ,but will be recorded as the second target signal-to-noise ratio interval;

[0032] If the low-definition high signal-to-noise ratio interval does not intersect with the high-definition low signal-to-noise ratio interval, the minimum signal-to-noise ratio in the high-definition low signal-to-noise ratio interval is obtained and recorded as , the maximum signal-to-noise ratio in the high-definition signal-to-noise ratio range is obtained and recorded as but It will be recorded as the second target signal-to-noise ratio interval.

[0033] As a further solution of the present invention: the noise suppression gain adjustment interval acquisition process is:

[0034] The first target signal-to-noise ratio interval As an example, the first target signal-to-noise ratio interval is divided into several acquisition points at the same time, and the signal-to-noise ratio corresponding to each acquisition point is obtained respectively. , Distortion Index and spectrum smoothness ; Where a represents the number of acquisition points, and a is a positive integer greater than 0;

[0035] Based on signal-to-noise ratio , get the clarity score , and then according to the clarity score Get the distortion index and spectrum smoothness , the noise suppression gain is obtained by comprehensive calculation ;

[0036] The target speech signal clarity score interval , in order from small to large, get the corresponding signal-to-noise ratio With clarity score ; where N represents the number of speech signals in the target interval; intelligibility score is the dependent variable, signal-to-noise ratio As the independent variable; the maximum signal-to-noise ratio corresponds to the minimum clarity score in order, and all the clarity scores SNR One-to-one correspondence, a rectangular coordinate system is established to observe the distribution of signal-to-noise ratio and clarity score on the coordinate axis;

[0037] By formula Calculate the rate of change ;

[0038] If the rate of change If it is a positive number, it means that all noise suppression gains in the first target interval are calculated and the maximum noise suppression gain is obtained. and minimum noise suppression gain G1; obtain the noise suppression gain adjustment range ;

[0039] If the rate of change If it is a negative number, calculate all noise suppression gains in the first target interval and obtain the maximum noise suppression gain and the minimum noise suppression gain G3; obtain the noise suppression gain adjustment range .

[0040] As a further solution of the present invention: time domain and frequency domain analysis are performed on multi-channel audio signals to extract the time domain waveform and spectrum characteristics of each channel respectively; the analysis results of all channels are summarized to calculate the overall distortion index D and spectrum smoothness S; the audio data of each channel are combined to calculate the comprehensive speech clarity score C; if the speech clarity score C is lower than a preset threshold, the system will further analyze the signal-to-noise ratio SNR of each channel, identify the main noise source and perform targeted processing; according to the analysis results of the multi-channel signal, the noise suppression parameters of each channel are dynamically adjusted.

[0041] An AI speech analysis system for IoT card control, comprising a speech signal acquisition module, a speech signal analysis module, a speech clarity score calculation module, a noise impact analysis module, a noise adjustment and optimization module, and a multi-channel audio signal processing module;

[0042] A voice signal acquisition module, wherein the voice signal acquisition module acquires original voice signals through an audio input device;

[0043] A speech signal analysis module, which performs time domain analysis and frequency domain analysis on the original audio data;

[0044] A speech clarity score calculation module, wherein the speech clarity score calculation module calculates the speech clarity score by comprehensively analyzing the time domain and frequency domain features of the speech signal;

[0045] A noise impact analysis module, wherein the noise impact analysis module obtains a target signal-to-noise ratio interval and a target speech signal clarity score interval by analyzing the influence of noise on speech clarity;

[0046] A noise adjustment and optimization module, wherein the noise adjustment and optimization module calculates a noise suppression gain adjustment interval based on a target speech signal clarity score interval;

[0047] A multi-channel audio signal processing module collects voice signals by using a multi-channel audio input device, analyzes the multi-channel voice signals, and dynamically adjusts the noise suppression parameters of each channel according to the analysis results.

[0048] Beneficial effects of the present invention:

[0049] (1) The present invention distinguishes high-definition and low-definition speech signals by collecting multiple speech signals and calculating their respective clarity scores; the target signal-to-noise ratio interval is determined by calculating the signal-to-noise ratio and integrating the signal-to-noise ratio intervals of the high-definition and low-definition speech signals. Based on the minimum clarity score under the maximum signal-to-noise ratio state, the distortion index and spectrum smoothness are obtained, and the noise suppression gain is calculated by the formula; this method can dynamically adjust the noise suppression gain and significantly improve the clarity of the speech signal; by observing the changes in the signal-to-noise ratio and clarity score in a rectangular coordinate system, the noise suppression parameters can be adjusted more accurately, thereby maintaining the high quality of the speech signal in different noise environments;

[0050] (2) The present invention analyzes the speech signal within the target speech signal clarity score interval and adjusts the noise suppression gain according to the change of the signal-to-noise ratio; specifically, by calculating the change rate, the relationship between the signal-to-noise ratio and the clarity score can be determined; when the change rate is a positive number, it indicates that as the signal-to-noise ratio increases, the clarity score also increases, and a noise suppression gain adjustment interval can be formed at this time; when the change rate is a negative number, it indicates that as the signal-to-noise ratio increases, the clarity score decreases, and a corresponding adjustment interval can also be formed. This dynamic adjustment mechanism enables the system to better adapt to different noise environments, improves the effect of noise suppression, and thus improves the accuracy of speech recognition.

[0051] (3) The present invention supports multi-channel audio signal processing, uses a multi-channel audio input device to collect voice signals, and ensures that the audio signals of all channels have the same sampling rate and bit depth; performs time domain and frequency domain analysis on each channel, extracts the time domain waveform and spectrum characteristics, summarizes the analysis results, and calculates the overall distortion index and spectrum smoothness; combines the audio data of each channel to generate a comprehensive speech clarity score; this method not only improves the overall clarity of the speech signal, but also enhances the sound in the target speech direction by analyzing the directionality of the multi-channel signal, while suppressing noise in other directions; by dynamically adjusting the noise suppression parameters of each channel, the overall speech clarity is optimized, and the robustness and reliability of the system are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The present invention will be further described below in conjunction with the accompanying drawings.

[0053] Figure 1 This is a flowchart of the specific steps of an AI voice analysis method for IoT card control in the present invention;

[0054] Figure 2 This is a flow chart of a standard for determining language clarity in an AI speech analysis method for IoT card control according to the present invention;

[0055] Figure 3 This is a flowchart of an AI voice analysis system for IoT card control in the present invention. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] See also Figure 1 As shown, the present invention is an AI voice analysis method for IoT card control, comprising the following steps:

[0058] S1: Collect the user's voice commands and obtain the original audio data;

[0059] S2: Perform time domain analysis and frequency domain analysis on the original audio data, and obtain the speech intelligibility score through comprehensive analysis;

[0060] S3: Analyze the influence of noise on speech clarity, and obtain the target signal-to-noise ratio range and the target speech signal clarity score range;

[0061] S4: Calculate the noise suppression gain adjustment interval based on the target speech signal clarity score interval;

[0062] S5: Use a multi-channel audio input device to collect voice signals, analyze the multi-channel voice signals, and dynamically adjust the noise suppression parameters of each channel according to the analysis results.

[0063] Example 2

[0064] On the basis of Example 1, the original voice signal is collected through an audio input device, wherein before the audio input device collects the signal, the audio input device is calibrated to ensure the accuracy of the collected data;

[0065] The audio input device supports high sampling rate (such as 48kHz) and high bit depth (such as 24 bits) to ensure that the details of the voice signal are captured;

[0066] The sampling rate indicates the number of samples collected per second when collecting audio signals, usually expressed in Hz, and the bit depth indicates the accuracy of each sample point, that is, the number of bits occupied by each sample point.

[0067] Perform time domain analysis on the collected raw audio data to extract the time domain waveform characteristics of the sound wave, including time domain amplitude and time domain energy;

[0068] Perform frequency domain analysis on the collected raw audio data to extract spectrum features, including frequency components and spectrum smoothness;

[0069] By formula The distortion index D is calculated;

[0070] Where N is the number of sample points, is the original signal of the i-th sample point, yi is the processed signal of the i-th sample point, i=1, 2, ..., N;

[0071] It should be noted that the original signal represents the time domain waveform characteristics of the collected original speech, including time domain amplitude and time domain energy.

[0072] By formula The spectrum smoothness S is calculated;

[0073] Where M is the number of components in the spectrum, and are the jth and j+1th components of the spectrum, j=1, 2, ..., M;

[0074] It should be noted that: the component represents the spectrum characteristics when the collected original audio data is analyzed in the frequency domain, including frequency components and spectrum smoothness;

[0075] Process the distortion index D and spectrum smoothness S;

[0076] The weight of the distortion index D is assigned to w1, and the weight of the spectrum smoothness S is assigned to w2, and both w1 and w2 are greater than 0;

[0077] By formula The speech intelligibility score C is calculated;

[0078] See also Figure 2 , comparing the speech intelligibility score C with a preset speech intelligibility score threshold;

[0079] If the speech clarity score C is greater than or equal to the preset speech clarity score threshold, it means that the collected original speech has high clarity and a high-definition speech signal is generated;

[0080] If the speech clarity score C is less than the preset speech clarity score threshold, it means that the clarity of the collected original speech is low, and a low-definition speech signal is generated;

[0081] It should be noted that: by performing time domain analysis and frequency domain analysis on the speech signal, we can obtain the distortion index D and spectral smoothness S of the collected speech signal. By processing the distortion index D and spectral smoothness S, we can obtain the speech clarity score C. The larger the speech clarity score C is, the higher the clarity of the speech signal is and the better the speech quality is.

[0082] Example 3

[0083] On the basis of Example 2, since external noise interference will affect the voice quality and cause low-definition voice signals, the degree of influence of noise interference on voice clarity is analyzed and determined. If there is a noise influence relationship, it is convenient to reduce noise interference and improve voice quality during subsequent voice collection.

[0084] Specific:

[0085] By formula The signal-to-noise ratio SNR is calculated;

[0086] in, is the original speech signal of the ith sample point, is the noise signal of the i-th sample point, N is the number of sample points;

[0087] It should be noted that the signal-to-noise ratio (SNR), as a quantitative indicator of noise intensity, reflects the strength of the noise. The larger the signal-to-noise ratio, the higher the noise interference intensity.

[0088] Specific:

[0089] Collect several different speech signals and calculate the clarity score of each speech signal ; Wherein, m represents the number of collected speech signals;

[0090] The clarity score of each speech signal Compare with clarity score threshold;

[0091] A plurality of high-definition speech signals and a plurality of low-definition speech signals are obtained respectively;

[0092] For all high-definition speech signals, the corresponding signal-to-noise ratio is calculated, and the obtained signal-to-noise ratios are integrated in order of magnitude to obtain the signal-to-noise ratio interval corresponding to the high-definition speech signal, which is recorded as the high-definition low signal-to-noise ratio interval;

[0093] For all low-definition speech signals, the corresponding signal-to-noise ratio is calculated, and the obtained signal-to-noise ratios are integrated in order of magnitude to obtain the signal-to-noise ratio interval corresponding to the low-definition speech signal, which is recorded as the low-definition high signal-to-noise ratio interval;

[0094] If there is an intersection between the high-definition low signal-to-noise ratio interval and the low-definition high signal-to-noise ratio interval, the minimum signal-to-noise ratio in the high-definition low-noise ratio interval is obtained and recorded as , get the minimum signal-to-noise ratio in the low-resolution and high-signal-to-noise ratio range and record it as ,Will Recorded as the first target signal-to-noise ratio interval;

[0095] If the low-resolution high-SNR interval completely contains the high-resolution low-SNR interval, then the minimum SNR in the high-resolution low-SNR interval is obtained and recorded as , the maximum signal-to-noise ratio in the high-definition low signal-to-noise ratio range is obtained and recorded as ,but will be recorded as the second target signal-to-noise ratio interval;

[0096] If the low-definition high signal-to-noise ratio interval does not intersect with the high-definition low signal-to-noise ratio interval, the minimum signal-to-noise ratio in the high-definition low signal-to-noise ratio interval is obtained and recorded as , the maximum signal-to-noise ratio in the high-definition signal-to-noise ratio range is obtained and recorded as but will be recorded as the second target signal-to-noise ratio interval;

[0097] Calculate the clarity score Cn of each speech signal within each target SNR interval, and sort the clarity scores Cn of the speech signals by size to obtain the target speech signal clarity score interval ; Where n represents the number of speech signals within the target signal-to-noise ratio interval.

[0098] It should be noted that for each target interval of the signal-to-noise ratio, the clarity score of each language signal is calculated and sorted by size, and then the target speech signal clarity score interval can be obtained.

[0099] The technical solution of this embodiment is: by collecting multiple speech signals and calculating their respective clarity scores, the clarity score of each signal is compared with a preset clarity score threshold, so as to distinguish high-definition speech signals from low-definition speech signals. Then, the signal-to-noise ratios of the two types of signals are calculated respectively, and integrated into corresponding signal-to-noise ratio intervals in order of size. If there is an overlapping part between the high-definition low-signal-to-noise ratio interval and the low-definition high-signal-to-noise ratio interval, and the former has a larger range, the maximum signal-to-noise ratio of the overlapping part is selected as the first target signal-to-noise ratio interval; if the low-definition high-signal-to-noise ratio interval is completely located in the high-definition low-signal-to-noise ratio interval, the minimum signal-to-noise ratio of the two intervals is selected as the second target signal-to-noise ratio interval; if the two have no intersection, the low-definition high-signal-to-noise ratio interval is directly used as the third target signal-to-noise ratio interval. Finally, the clarity scores of the speech signals in each target signal-to-noise ratio interval are calculated and sorted by size to form a target speech signal clarity score interval, thereby providing a basis for the optimization processing of the speech signal.

[0100] Example 4

[0101] Based on Example 3, the first target signal-to-noise ratio interval As an example, the first target signal-to-noise ratio interval is divided into several acquisition points at the same time, and the signal-to-noise ratio corresponding to each acquisition point is obtained respectively. , Distortion Index and spectrum smoothness ; Where a represents the number of acquisition points, and a is a positive integer greater than 0;

[0102] Based on signal-to-noise ratio , get the clarity score , and then according to the clarity score Get the distortion index and spectrum smoothness ; Substitute into the formula:

[0103] ;

[0104] in, is the ath noise suppression gain, is the preset scale factor, is the preset scale factor, is the preset scale factor.

[0105] The target speech signal clarity score interval , in order from small to large, get the corresponding signal-to-noise ratio With clarity score ; where N represents the number of speech signals in the target interval; intelligibility score is the dependent variable, signal-to-noise ratio As the independent variable; the maximum signal-to-noise ratio corresponds to the minimum clarity score in order, and all the clarity scores SNR One-to-one correspondence, a rectangular coordinate system is established to observe the distribution of signal-to-noise ratio and clarity score on the coordinate axis;

[0106] By formula Calculate the rate of change ;

[0107] It should be noted that the rate of change Indicated on the coordinate axis, the clarity score With signal-to-noise ratio an increase or decrease in the change;

[0108] If the rate of change If it is a positive number, it means that the lower the clarity score, the larger the corresponding signal-to-noise ratio, and the larger the noise gain. Calculate all noise suppression gains in the first target interval and obtain the maximum noise suppression gain and minimum noise suppression gain G1; obtain the noise suppression gain adjustment range ;

[0109] If the rate of change If it is a negative number, it means that the higher the clarity score, the lower the corresponding signal-to-noise ratio, and the lower the noise gain. Calculate all noise suppression gains in the first target interval and obtain the maximum noise suppression gain. and the minimum noise suppression gain G3; obtain the noise suppression gain adjustment range .

[0110] The technical solution of the present invention is: for speech signals within the target speech signal clarity score interval, the corresponding signal-to-noise ratio and clarity score are obtained in order from small to large signal-to-noise ratio; wherein NN represents the number of speech signals within the target interval; with the clarity score as the dependent variable and the signal-to-noise ratio as the independent variable, the maximum signal-to-noise ratio corresponds to the minimum clarity score in order, all clarity scores are matched to the signal-to-noise ratio one by one, a rectangular coordinate system is established, and the distribution of the signal-to-noise ratio and the clarity score on the coordinate axis is observed; the change rate is calculated by a formula, and the change rate represents the change of the clarity score with the signal-to-noise ratio on the coordinate axis; if the change rate is a positive number, it means that as the signal-to-noise ratio increases, the clarity score also increases, forming a noise suppression gain adjustment interval; if the change rate is a negative number, it means that as the signal-to-noise ratio increases, the clarity score decreases, forming a noise suppression gain adjustment interval; through this method, the noise suppression gain can be dynamically adjusted to optimize the clarity of the speech signal.

[0111] It should be noted that the strength of noise suppression can be increased by adjusting the noise suppression parameter to increase the suppression gain Gm, thereby increasing the noise suppression effect, thereby improving the speech clarity score and making the speech clarity higher.

[0112] Example 5

[0113] On the basis of Example 4, the system is expanded to support multi-channel audio signal processing, and a multi-channel audio input device is used for voice signal acquisition, and each channel provides an independent audio signal; each channel is calibrated to ensure that the audio signals of all channels have the same sampling rate and bit depth; the multi-channel audio signal is analyzed in the time domain and frequency domain, and the time domain waveform and spectrum characteristics of each channel are extracted respectively; the analysis results of all channels are summarized, and the overall distortion index D and spectrum smoothness S are calculated; the audio data of each channel are combined, and a comprehensive voice clarity score C is generated through a signal fusion algorithm; if the voice clarity score C is lower than a preset threshold, the system will further analyze the signal-to-noise ratio SNR of each channel, identify the main noise source and perform targeted processing; according to the analysis results of the multi-channel signal, the noise suppression parameters of each channel are dynamically adjusted to optimize the overall voice clarity;

[0114] It should be noted that when expanding the system to support multi-channel audio signal processing, it is necessary to ensure that the signals from different channels are synchronized in time to avoid interference caused by time deviation. This goal can be achieved by using hardware-triggered synchronization or software synchronization mechanisms, using the time difference and phase difference of multi-channel signals to locate the location of the noise source, and then taking measures to reduce the noise impact of these locations; by analyzing the directionality of multi-channel signals, the sound in the direction of the target voice is enhanced while suppressing noise in other directions; providing a user-friendly graphical interface to display key indicators such as the signal-to-noise ratio and clarity score of each channel, so that users can monitor the operating status of the system.

[0115] Example 6

[0116] See also Figure 3 , an AI speech analysis system for IoT card control, comprising a speech signal acquisition module, a speech signal analysis module, a speech clarity score calculation module, a noise impact analysis module, a noise adjustment and optimization module, and a multi-channel audio signal processing module;

[0117] A voice signal acquisition module, wherein the voice signal acquisition module acquires original voice signals through an audio input device;

[0118] A speech signal analysis module, which performs time domain analysis and frequency domain analysis on the original audio data;

[0119] A speech clarity score calculation module, wherein the speech clarity score calculation module calculates the speech clarity score by comprehensively analyzing the time domain and frequency domain features of the speech signal;

[0120] A noise impact analysis module, wherein the noise impact analysis module obtains a target signal-to-noise ratio interval and a target speech signal clarity score interval by analyzing the influence of noise on speech clarity;

[0121] A noise adjustment and optimization module, wherein the noise adjustment and optimization module calculates a noise suppression gain adjustment interval based on a target speech signal clarity score interval;

[0122] A multi-channel audio signal processing module collects voice signals by using a multi-channel audio input device, analyzes the multi-channel voice signals, and dynamically adjusts the noise suppression parameters of each channel according to the analysis results.

[0123] The working principle of the present invention includes: collecting the user's voice command and obtaining original audio data, performing time domain and frequency domain analysis on the original audio data, and comprehensively obtaining a voice clarity score; analyzing the influence of noise on voice clarity, and obtaining a target signal-to-noise ratio interval and a target voice signal clarity score interval; based on the target voice signal clarity score interval, calculating a noise suppression gain adjustment interval; using a multi-channel audio input device to collect voice signals, and analyzing the multi-channel voice signals, and dynamically adjusting the noise suppression parameters of each channel according to the analysis results. Specifically, by calculating the distortion index D and the spectral smoothness S, and combining them with the preset weight coefficient, the speech clarity score C is obtained, and C is compared with the preset clarity score threshold to distinguish between high-definition and low-definition speech signals; by calculating the signal-to-noise ratio SNR, the signal-to-noise ratio ranges of high-definition and low-definition speech signals are integrated to determine the target signal-to-noise ratio range; based on the minimum clarity score Cmin under the maximum signal-to-noise ratio state, the distortion index D and the spectral smoothness S are obtained, and the noise suppression gain G is calculated by the formula, and the noise suppression gain is dynamically adjusted to optimize the speech signal clarity; in a multi-channel environment, by calibrating each channel to ensure data consistency, the time domain and frequency domain features of each channel are extracted, and the signal fusion algorithm is combined to generate a comprehensive speech clarity score, and the noise suppression parameters are dynamically adjusted according to the analysis results of the multi-channel signal, thereby optimizing the overall speech clarity and improving the reliability of voice control and user experience.

[0124] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. An AI voice analysis method for IoT card control, characterized in that: The following steps are involved: S1: Collect the user's voice commands and obtain the original audio data; S2: Perform time domain analysis and frequency domain analysis on the original audio data, and obtain the speech intelligibility score through comprehensive analysis; S3: Analyze the influence of noise on speech clarity, and obtain the target signal-to-noise ratio range and the target speech signal clarity score range; S4: Calculate the noise suppression gain adjustment interval based on the target speech signal clarity score interval; S5: Use a multi-channel audio input device to collect voice signals, analyze the multi-channel voice signals, and dynamically adjust the noise suppression parameters of each channel according to the analysis results; The target signal-to-noise ratio interval acquisition process is: The clarity score of each speech signal Compare with clarity score threshold; A plurality of high-definition speech signals and a plurality of low-definition speech signals are obtained respectively; For all high-definition speech signals, the corresponding signal-to-noise ratio is calculated, and the obtained signal-to-noise ratios are integrated in order of magnitude to obtain the signal-to-noise ratio interval corresponding to the high-definition speech signal, which is recorded as the high-definition low signal-to-noise ratio interval; For all low-definition speech signals, the corresponding signal-to-noise ratio is calculated, and the obtained signal-to-noise ratios are integrated in order of magnitude to obtain the signal-to-noise ratio interval corresponding to the low-definition speech signal, which is recorded as the low-definition high signal-to-noise ratio interval; If there is an intersection between the high-definition low signal-to-noise ratio interval and the low-definition high signal-to-noise ratio interval, the minimum signal-to-noise ratio in the high-definition low-noise ratio interval is obtained and recorded as , get the minimum signal-to-noise ratio in the low-resolution and high-signal-to-noise ratio range and record it as ,Will Recorded as the first target signal-to-noise ratio interval; If the low-resolution high-SNR interval completely contains the high-resolution low-SNR interval, then the minimum SNR in the high-resolution low-SNR interval is obtained and recorded as , the maximum signal-to-noise ratio in the high-definition low signal-to-noise ratio range is obtained and recorded as ,Will Recorded as the second target signal-to-noise ratio interval; If the low-definition high signal-to-noise ratio interval does not intersect with the high-definition low signal-to-noise ratio interval, the minimum signal-to-noise ratio in the high-definition low signal-to-noise ratio interval is obtained and recorded as , the maximum signal-to-noise ratio in the high-definition signal-to-noise ratio range is obtained and recorded as ,but will be recorded as the second target signal-to-noise ratio interval; The process of obtaining the target speech signal clarity score interval is as follows: Calculate the clarity score Cn of each speech signal within the target signal-to-noise ratio range, and sort the clarity scores Cn of the speech signals by size to obtain the target speech signal clarity score range ; Where n represents the number of speech signals within the target signal-to-noise ratio interval; Based on the first target signal-to-noise ratio interval , divide the first target signal-to-noise ratio interval into several acquisition points at the same time, and obtain the signal-to-noise ratio corresponding to each acquisition point , Distortion Index and spectral smoothness ; Where a represents the number of acquisition points, and a is a positive integer greater than 0; Based on signal-to-noise ratio , get the clarity score , and then according to the clarity score Get the distortion index and spectral smoothness , the noise suppression gain is obtained by comprehensive calculation ; The target speech signal clarity score interval , in order from small to large, get the corresponding signal-to-noise ratio With clarity score ; By clarity score is the dependent variable, signal-to-noise ratio As the independent variable; the maximum signal-to-noise ratio corresponds to the minimum clarity score in order, and all the clarity scores SNR One-to-one correspondence, a rectangular coordinate system is established to observe the distribution of signal-to-noise ratio and clarity score on the coordinate axis; By formula Calculate the rate of change .

2. The AI ​​voice analysis method for IoT card control according to claim 1, characterized in that: The process of obtaining the speech intelligibility score C is as follows: Processing the distortion index D and spectrum smoothness S of the original audio data; The weight of the distortion index D is assigned to w1, and the weight of the spectrum smoothness S is assigned to w2, and both w1 and w2 are greater than 0; By formula The speech intelligibility score C is calculated.

3. The AI ​​voice analysis method for IoT card control according to claim 2 is characterized in that: The process of obtaining the distortion index D is: By formula The distortion index D is calculated; Where N is the number of sample points, is the i-th sample point of the original signal, is the i-th sample point of the processed signal.

4. The AI ​​voice analysis method for IoT card control according to claim 2 is characterized in that: The process of obtaining the spectrum smoothness S is: By formula The spectrum smoothness S is calculated; Where M is the number of components in the spectrum, and are the jth and j+1th components of the spectrum respectively.

5. The AI ​​voice analysis method for IoT card control according to claim 1 is characterized in that: Signal-to-noise ratio (SNR) acquisition process: By formula The signal-to-noise ratio SNR is calculated; in, is the i-th sample point of the original speech signal, is the i-th sample point of the noise signal, and N is the number of sample points.

6. The AI ​​voice analysis method for IoT card control according to claim 1, characterized in that: The process of obtaining the noise suppression gain adjustment interval is as follows: If the rate of change is a positive number, calculate all noise suppression gains in the first target interval, and obtain the maximum noise suppression gain and minimum noise suppression gain G1; Get the noise suppression gain adjustment range ; If the rate of change If it is a negative number, calculate all noise suppression gains in the first target interval and obtain the maximum noise suppression gain and minimum noise suppression gain G3; obtain the noise suppression gain adjustment range .

7. An AI voice analysis system for IoT card control, characterized in that: It includes a speech signal acquisition module, a speech signal analysis module, a speech clarity score calculation module, a noise impact analysis module, a noise adjustment and optimization module, and a multi-channel audio signal processing module; A voice signal acquisition module, wherein the voice signal acquisition module acquires original voice signals through an audio input device; A speech signal analysis module, wherein the speech signal analysis module performs time domain analysis and frequency domain analysis on the original audio data; A speech clarity score calculation module, wherein the speech clarity score calculation module calculates the speech clarity score by comprehensively analyzing the time domain and frequency domain features of the speech signal; A noise impact analysis module, wherein the noise impact analysis module obtains a target signal-to-noise ratio interval and a target speech signal clarity score interval by analyzing the influence of noise on speech clarity; A noise adjustment and optimization module, wherein the noise adjustment and optimization module calculates a noise suppression gain adjustment interval based on a target speech signal clarity score interval; A multi-channel audio signal processing module, wherein the multi-channel audio signal processing module collects voice signals by using a multi-channel audio input device, analyzes the multi-channel voice signals, and dynamically adjusts the noise suppression parameters of each channel according to the analysis results; The target signal-to-noise ratio interval acquisition process is: The clarity score of each speech signal Compare with clarity score threshold; A plurality of high-definition speech signals and a plurality of low-definition speech signals are obtained respectively; For all high-definition speech signals, the corresponding signal-to-noise ratio is calculated, and the obtained signal-to-noise ratios are integrated in order of magnitude to obtain the signal-to-noise ratio interval corresponding to the high-definition speech signal, which is recorded as the high-definition low signal-to-noise ratio interval; For all low-definition speech signals, the corresponding signal-to-noise ratio is calculated, and the obtained signal-to-noise ratios are integrated in order of magnitude to obtain the signal-to-noise ratio interval corresponding to the low-definition speech signal, which is recorded as the low-definition high signal-to-noise ratio interval; If there is an intersection between the high-definition low signal-to-noise ratio interval and the low-definition high signal-to-noise ratio interval, the minimum signal-to-noise ratio in the high-definition low-noise ratio interval is obtained and recorded as , get the minimum signal-to-noise ratio in the low-resolution and high-signal-to-noise ratio range and record it as ,Will Recorded as the first target signal-to-noise ratio interval; If the low-resolution high-SNR interval completely contains the high-resolution low-SNR interval, then the minimum SNR in the high-resolution low-SNR interval is obtained and recorded as , the maximum signal-to-noise ratio in the high-definition low signal-to-noise ratio range is obtained and recorded as ,Will Recorded as the second target signal-to-noise ratio interval; If the low-definition high signal-to-noise ratio interval does not intersect with the high-definition low signal-to-noise ratio interval, the minimum signal-to-noise ratio in the high-definition low signal-to-noise ratio interval is obtained and recorded as , the maximum signal-to-noise ratio in the high-definition signal-to-noise ratio range is obtained and recorded as ,but will be recorded as the second target signal-to-noise ratio interval; The process of obtaining the target speech signal clarity score interval is as follows: Calculate the clarity score Cn of each speech signal within the target signal-to-noise ratio range, and sort the clarity scores Cn of the speech signals by size to obtain the target speech signal clarity score range ; Where n represents the number of speech signals within the target signal-to-noise ratio interval; Based on the first target signal-to-noise ratio interval , divide the first target signal-to-noise ratio interval into several acquisition points at the same time, and obtain the signal-to-noise ratio corresponding to each acquisition point , Distortion Index and spectral smoothness ; Where a represents the number of acquisition points, and a is a positive integer greater than 0; Based on signal-to-noise ratio , get the clarity score , and then according to the clarity score Get the distortion index and spectral smoothness , the noise suppression gain is obtained by comprehensive calculation ; The target speech signal clarity score interval , in order from small to large, get the corresponding signal-to-noise ratio With clarity score ; By clarity score is the dependent variable, signal-to-noise ratio As the independent variable; the maximum signal-to-noise ratio corresponds to the minimum clarity score in order, and all the clarity scores SNR One-to-one correspondence, a rectangular coordinate system is established to observe the distribution of signal-to-noise ratio and clarity score on the coordinate axis; By formula Calculate the rate of change .

Images