Intelligent telephone customer service interaction method for power enterprise based on artificial intelligence

Abstract

The invention belongs to the technical field of voice signal quantization, and particularly relates to a power enterprise smart phone customer service interaction method based on artificial intelligence, which comprises the following steps: acquiring a voice frame in real time and converting the voice frame into a frequency domain, extracting a main masking tone and a total masking threshold value, and combining with Bark sub-band weighted fusion to obtain a main masking tone and a total masking threshold value; and generating the comprehensive perception masking intensity reflecting the voice masking capability. And dynamically adjusting the sensitivity weight and the quantization relaxation coefficient of each section of the A-law thirteen-line quantizer based on the intensity, adaptively adjusting the range of each section and the quantization step length, and realizing non-uniform quantization of perception driving on the premise of keeping the input range compatible with the G.711 standard. When voice energy is concentrated, quantization is relaxed to improve compression efficiency, and when voice is sensitive, fine quantization is maintained to protect sharpness. According to the invention, on the basis of not changing the decoding compatibility, both the subjective voice quality and the coding efficiency are considered.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of speech signal quantization. More particularly, the present application relates to an intelligent telephone customer service interaction method based on artificial intelligence for power enterprises. BACKGROUND

[0002] In the intelligent telephone customer service system of modern power enterprises, speech interaction is the core channel of user service. A large amount of voice call data needs to be transmitted in real time through the public switched telephone network (PSTN) or IP network, which puts forward higher requirements on the real-time performance, compatibility and bandwidth efficiency of speech coding. In order to realize effective compression and high-quality restoration of speech signals, A-law or mu-law pulse code modulation (PCM) coding standards are widely used, among which A-law thirteen-fold line quantization, as the core technology of ITU-T G.711 protocol, is widely used in traditional telephone systems and embedded speech devices due to its simple algorithm, low delay and strong decoding compatibility.

[0003] A-law thirteen-fold line is a non-uniform quantization method, which divides the amplitude range of the input signal into 13 linear segments (8 segments on the positive and negative half axes, respectively, with a shared zero segment), to realize fine quantization of small signals and rough compression of large signals, thereby expanding the dynamic range under 8-bit PCM code stream and improving the overall auditory quality. This method has a good deployment basis in the existing customer service communication architecture of power enterprises, supports seamless connection with traditional PSTN devices, and is suitable for high-reliability and low-delay speech interaction scenarios.

[0004] However, the traditional A-law quantization adopts a fixed segmentation structure and quantization step, lacks perceptual analysis of speech content, and cannot realize dynamic bit allocation, resulting in high average code rate, which is difficult to meet the bandwidth optimization requirements in high-concurrency customer service scenarios, especially in remote areas or network resource limited environments, affecting system stability and service quality. SUMMARY

[0005] To solve the technical problem that the traditional A-law quantization adopts a fixed segmentation structure and quantization step, which is difficult to meet the bandwidth optimization requirements in high-concurrency customer service scenarios, the present application provides an intelligent telephone customer service interaction method based on artificial intelligence for power enterprises, comprising: The speech frames in the telephone customer service interaction process are collected in real time and converted into the frequency domain; the main masking sound of the speech frame is obtained, and the total masking threshold of the main masking sound on each frequency component of the speech frame is determined. The comprehensive perceptual masking strength of the speech frame is determined according to the total masking threshold of each frequency component of the speech frame in the frequency domain; for each positive segment in the A-law thirteen-fold line, the sensitivity weight of each positive segment is determined according to the comprehensive perceptual masking strength of the speech frame and the sequence number of the positive segment, the quantization relaxation coefficient of each positive segment is determined according to the comprehensive perceptual masking strength of the speech frame and the sensitivity weight of each positive segment, the range size of each positive segment is adjusted according to the quantization relaxation coefficient, and the speech frame is quantized and encoded according to the adjusted range size.

[0006] The present invention uses a psychoacoustic model to dynamically analyze the masking characteristics of speech frames, generates a comprehensive perceptual masking strength that reflects the human ear's ability to mask noise, and adjusts the sensitivity weight and quantization relaxation coefficient of each quantization segment accordingly, so that high-energy, high-masking speech frames are moderately relaxed in quantization in high-frequency and low-amplitude segments to improve compression efficiency, while maintaining fine quantization in sensitive areas such as unvoiced consonants and transition sounds to protect speech clarity. This effectively avoids the waste of bit resources and auditory distortion caused by traditional fixed quantization, takes into account the subjective quality and compression efficiency of speech coding, and at the same time maintains full compatibility of the decoding end with existing PSTN and IP communication systems, thereby improving the real-time and compatibility of voice interaction of smart phone customer service of power companies.

[0007] Preferably, the method of determining the comprehensive perceptual masking strength of the speech frame based on the total masking threshold of each frequency component of the speech frame in the frequency domain includes: dividing the frequency components into a number of Bark subbands of equal width according to the values ​​of each frequency component of the speech frame in the frequency domain under the Bark scale; for any subband, obtaining the mean of the total masking thresholds at all frequency components contained in the subband as the total masking threshold of the subband; using a Gaussian function to dynamically generate the perceptual weight of each subband, and performing weighted summation on the total masking threshold of the subband according to the perceptual weight of each subband to obtain the comprehensive perceptual masking strength of the speech frame.

[0008] The present invention obtains the comprehensive perceptual masking strength by weighted summation of sub-band masking thresholds, which not only preserves the integrity of frequency domain perceptual information, but also integrates it into a scalar parameter that can be used for time domain quantization control, realizing an efficient mapping from multidimensional psychoacoustic analysis to one-dimensional coding adjustment, and providing a scientific and stable perceptual basis for the subsequent adaptive adjustment of the A-law quantization strategy.

[0009] Preferably, the perceptual weight of the sub-band satisfies the expression: ;in, Indicates the The perceptual weight of each sub-band; Indicates the The central Bark value of the sub-band, Indicates the preset benchmark Bark value; is the Gaussian attenuation factor.

[0010] Preferably, the sensitivity weight of each positive segment satisfies the expression: ;in, Indicates the The first paragraph Sensitivity weight under speech frames; Indicates the sequence number of the positive segment; Indicates the The normalized comprehensive perceptual masking strength of the speech frames.

[0011] The present invention utilizes the nonlinear characteristics of the power function to maintain high sensitivity of low-amplitude segments to protect voiceless consonants and transition sounds when the masking intensity is low, and significantly reduces the sensitivity of each segment, especially the low-amplitude segment, when the masking intensity is high, so that the quantization step size can be appropriately increased, avoiding bit waste caused by excessively fine quantization in the maskable area. It retains the A-law's protection mechanism for small signals and introduces perception-driven adaptive capabilities, significantly improving the balance between coding efficiency and subjective listening quality.

[0012] Preferably, the quantization relaxation coefficient satisfies the expression: ;in, Indicates the The first paragraph Quantization relaxation coefficient under speech frames; is the preset gain coefficient; Indicates the The normalized comprehensive perceptual masking strength of the speech frames; Indicates the The first paragraph Sensitivity weight under each speech frame.

[0013] The present invention uses the normalized comprehensive perceptual masking strength as the enabling factor of the masking capability, ensuring that the relaxation mechanism is activated only when the speech is maskable, so that the adjustment of the range of each positive segment is consistent with the auditory characteristics of the human ear and has engineering controllability. In high-masking speech frames, the quantization step size of high-amplitude segments and some low-amplitude segments is effectively increased to improve compression efficiency, while in sensitive speech frames, the quantization accuracy of each segment is kept close to the standard A-law to avoid distortion, achieving a dynamic balance of coding granularity and taking into account both speech quality and bit utilization.

[0014] Preferably, the adjusting the range size of each positive segment according to the quantization relaxation coefficient includes: for each positive segment, calculating the product between the initial range size of each positive segment and the quantization relaxation coefficient, and normalizing the product to obtain the adjusted range size of each positive segment.

[0015] The present invention realizes dynamic redistribution of quantization intervals by normalizing the product of the initial range size of each positive segment and the quantization relaxation coefficient. This not only retains the adaptive adjustment capability based on perceptual intensity, but also ensures that the sum of the adjusted ranges of all positive segments remains 1, strictly maintains the input definition domain within the range of [0,1], avoids signal overflow or decoding mismatch caused by parameter adjustment, ensures full compatibility with the standard A-law decoder, improves the robustness and stability of the algorithm, and supports flexible optimization of coding granularity under different speech content, taking into account both compression efficiency and speech fidelity.

[0016] Preferably, the quantization encoding of the speech frame according to the adjusted range size includes: determining the adjusted range and quantization step size of each positive segment according to the adjusted range size of each positive segment; The first speech frame The time domain data are normalized and the results are used express; confirm The symbol is searched based on the adjusted range of all positive segments. The positive segment to which it belongs is quantized according to the quantization step size of the positive segment. Perform uniform quantization, calculate its quantization level, and combine it with the sign bit to generate an 8-bit nonlinear PCM code word to achieve quantization coding.

[0017] Preferably, the step of determining the adjusted range and quantization step of each positive segment according to the adjusted range size of each positive segment includes: positive segments, and the quantization step size is ;when When, The adjusted range of the positive segment is ;when When, The adjusted range of the positive segment is ;when When, The adjusted range of the positive segment is ,in Indicates the The first paragraph The adjusted range size under each speech frame is Indicates the The first paragraph The adjusted range size under each speech frame.

[0018] Preferably, obtaining the main masker sound of the speech frame includes: For each speech frame, the amplitudes corresponding to all frequency components of the speech frame in the frequency spectrum are arranged in ascending order of frequency to form an amplitude sequence, and local maximum values in the amplitude sequence are obtained; all local maximum values are sorted in descending order, and the frequency components corresponding to the first K local maximum values are taken as a primary masking sound respectively, wherein K is a preset number of primary masking sounds.

[0019] Preferably, the normalized integrated perceptual masking strength satisfies the expression: ; wherein, denotes the normalized integrated perceptual masking strength of the i-th speech frame; denotes the normalized integrated perceptual masking strength of the i-th speech frame; denotes the integrated perceptual masking strength of the i-th speech frame before normalization; denotes the integrated perceptual masking strength of the i-th speech frame before normalization; denotes a reference masking level; denotes a preset maximum masking level; denotes a minimum value function, denotes a maximum value function.

[0020] The present application has the advantages that: the present application can moderately relax quantization when the speech energy is concentrated to improve compression efficiency, and maintain fine quantization in sensitive areas such as clear consonants to protect the intelligibility of speech, effectively avoiding the bit waste and hearing distortion caused by traditional fixed quantization, balancing subjective speech quality and coding efficiency, and the encoding end is intelligent and the decoding end is transparent, improving the real-time performance and compatibility of the power enterprise intelligent telephone customer service voice interaction. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is a flowchart schematically showing a power enterprise intelligent telephone customer service interaction method based on artificial intelligence in the present application; Figure 2 is a flowchart schematically showing step 2 of a power enterprise intelligent telephone customer service interaction method based on artificial intelligence in the present application. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0023] The specific embodiments of the present application will be described in detail below with reference to the drawings.

[0024] The embodiments of the present application disclose a power enterprise intelligent telephone customer service interaction method based on artificial intelligence, which is described with reference toFigure 1 , including steps S1 to S3: S1. Real-time acquisition of voice frames during telephone customer service interactions and conversion of the voice frames into the frequency domain.

[0025] Specifically, during the power company's customer service interaction with a smart phone, user voice signals are collected in real time through the phone interface. These voice signals are converted from acoustic vibrations into continuously varying analog electrical signals via a microphone. These signals are then sampled by an analog-to-digital converter (ADC) to produce discrete-time digital voice signals. The sampling rate is set by the implementer based on actual implementation circumstances, for example, 8000Hz, 16000Hz, or 48000Hz.

[0026] It should be noted that speech signals are time-varying and non-stationary. Directly performing a Fourier transform on the entire signal will result in spectral leakage. To accurately analyze local characteristics, the present invention frames the speech signal and uses a window function to smooth the frame boundaries, reducing spectral leakage.

[0027] Specifically, the input speech signal is divided into a series of overlapping speech frames. In this embodiment, the duration of each frame (i.e., the frame length) is set to 20ms, and the inter-frame shift (i.e., the frame shift) is set to 10ms. In other embodiments, implementers can set the frame length and frame shift based on actual implementation.

[0028] Furthermore, each speech frame is windowed, and the window function is set by the implementer according to the actual implementation, such as Hamming window, Hanning window, etc. It should be noted that framing and windowing are well-known technologies and will not be described in detail here.

[0029] Furthermore, a short-time Fourier transform (STFT) is performed on each speech frame after the windowing process to convert the speech frame from the time domain to the frequency domain to obtain several frequency components.

[0030] S2. Obtain a main masker of the speech frame, determine a total masking threshold of the main masker for each frequency component of the speech frame, and determine a comprehensive perceptual masking strength of the speech frame based on the total masking threshold at each frequency component of the speech frame in the frequency domain.

[0031] See the flowchart of step S2 Figure 2 , including steps S201 to S203, specifically: S201: Obtain a main masker of a speech frame and a power level corresponding to the main masker.

[0032] It should be noted that the human auditory system is more sensitive to energy-concentrated areas in the spectrum. These areas often correspond to resonance peaks or strong consonant components (such as plosives and fricatives) in speech and have strong masking capabilities. For any frequency component in the spectrum of a speech frame, the amplitude is the complex modulus value of that frequency component, reflecting the vibration intensity of that frequency component in the speech frame. The larger the amplitude, the stronger the energy of that frequency component. In speech, this usually corresponds to energy-concentrated areas such as the resonance peaks (Formants) of vowels, fundamental harmonics, or strong consonants (such as plosives and fricatives). Therefore, the present invention locates the sound components that play a dominant role in perception based on the amplitude of each frequency component in the spectrum of a speech frame.

[0033] Specifically, for each speech frame, the amplitudes corresponding to all frequency components in the speech frame's spectrum are sorted from smallest to largest frequency to form an amplitude sequence, and the local maxima in the amplitude sequence are obtained. All local maxima are sorted from largest to smallest, and the frequency components corresponding to the first K local maxima are each used as a primary masker. K is the preset number of primary maskers. In this embodiment, K = 6. In other embodiments, implementers can flexibly set the value of K based on the actual application scenario, for example, K = 5-10, to accommodate different speech content (e.g., segments with dense voiceless consonants require more primary maskers).

[0034] It should be noted that the human ear is particularly sensitive to areas of concentrated energy in the spectrum. These areas are not only loud themselves but can also have a strong masking effect on adjacent frequencies. In other words, a strong sound can render nearby weaker sounds inaudible. Therefore, the present invention locates the primary masking sound by using local maxima in the amplitude sequence.

[0035] It's important to note that the human ear's perception of sound intensity is nonlinear, approximately following a logarithmic law (the Weber-Fechner law). Therefore, directly using linear amplitude cannot accurately reflect subjective loudness. Power level, on the other hand, converts linear amplitude to a logarithmic scale, which more closely reflects the true perceptual characteristics of the human ear. Furthermore, since the masking effect is essentially an energy-level phenomenon—the energy of a loud sound suppresses the energy of a soft sound—the present invention converts each primary masking tone into a power level.

[0036] Specifically, obtain the power level corresponding to each main masking sound: .

[0037] Where, Indicates the The first speech frame The power level corresponding to the main masking sound, in decibels (dB); Indicates that The first speech frame The amplitude corresponding to the main masker sound; is a smoothing constant used to prevent Appears , which leads to formula errors. In this embodiment, In other embodiments, the implementer can set ,For example ~ It should be noted that the calculation of the power level is a well-known technology, and its specific principle will not be described in detail here.

[0038] S202: Determine a total masking threshold of the main masker sound for each frequency component of the speech frame.

[0039] Specifically, the physical frequency of each main masker is converted from the Hz scale to the Bark perceptual scale.

[0040] It should be noted that the human ear's perception of frequency is nonlinear, with high resolution for low-frequency components and low resolution for high-frequency components. The frequency unit Hz is a linear unit and cannot reflect the true perception of the human ear. Bark is a psychoacoustic frequency scale. A "critical bandwidth" is the minimum frequency bandwidth that the human ear can distinguish. Therefore, the present invention converts the physical frequency of each frequency component of the speech frame from the Hz scale to the Bark perceptual scale. The specific conversion process is well known in the art and will not be detailed here.

[0041] Furthermore, for any frequency component of a speech frame , according to the asymmetric masking model, the main masking sound is determined in the frequency component The resulting masking threshold at: .

[0042] Where, Indicates the The first speech frame The main masking sound has a frequency component The masking threshold generated at Indicates the The first speech frame The power level corresponding to the main masking sound; Indicates the The first speech frame Main masking tones and frequency components The distance on the Bark scale, , Represents frequency components The value on the Bark scale, Indicates the The first speech frame The value of the main masker sound on the Bark scale; the coefficient 27 is an empirical constant that reflects the masking attenuation rate.

[0043] Furthermore, all the main masking sounds of the speech frame are The masking threshold generated at the superposition is used to obtain the frequency component The synthetic masking threshold at: .

[0044] in, Indicates the The frequency components of a speech frame The synthetic masking threshold at ; Indicates the The first speech frame The main masking sound has a frequency component The masking threshold generated at Indicates the The number of frequency components corresponding to each speech frame in the frequency domain. The main masking tones of each speech frame are in the frequency components The masking thresholds generated at are converted to linear energy units, summed, and then converted back to logarithmic scale to obtain the synthetic masking threshold .

[0045] It should be noted that even in a quiet environment with no other sounds present, the human ear cannot perceive sounds below a certain minimum intensity. This minimum audible sound intensity varies with frequency and is called the Absolute Threshold of Hearing (ATH). For example, the human ear is most sensitive in the 2–5 kHz region, while higher sound pressure is required for detection in the low-frequency (<200 Hz) and high-frequency (>15 kHz) regions. Therefore, the present invention determines the total masking threshold for each frequency component based on the absolute hearing threshold of the speech frame at each frequency component and the composite masking threshold.

[0046] Specifically, for any frequency component of a speech frame , obtain the frequency components of the speech frame through the ATH model It should be noted that the ATH model is a well-known technology and will not be described in detail here.

[0047] Frequency components of speech frames The synthetic masking threshold at is compared with the absolute hearing threshold, and the maximum value is taken as the frequency component The total masking threshold at .

[0048] It should be noted that the total masking threshold represents the maximum noise intensity that cannot be perceived by the human ear in the current speech background.

[0049] S203 : Determine the comprehensive perceptual masking strength of the speech frame according to the total masking threshold of each frequency component of the speech frame in the frequency domain.

[0050] It should be noted that to achieve perceptual modeling consistent with the human ear's auditory characteristics, the present invention divides the spectrum of the speech frame into several Bark subbands. The frequency range of each subband is divided according to the critical bandwidth characteristics of the human ear, so that the subbands have equal widths on the Bark scale, thereby more realistically simulating the filtering response mechanism of the human ear's basilar membrane.

[0051] Specifically, the frequency components of the speech frame in the frequency domain are divided into several Bark subbands of equal width based on the Bark scale value of each frequency component. In this embodiment, the total number of subbands is set to 16 to balance perceptual modeling accuracy and computational complexity. In other embodiments, the total number of subbands can be adjusted based on actual implementation.

[0052] Furthermore, for any sub-band, an average of the total masking thresholds at all frequency components included in the sub-band is obtained as the total masking threshold of the sub-band.

[0053] It should be noted that for different sub-bands, different frequency bands contribute differently to speech clarity and cannot be simply weighted averaged. The human ear has significant frequency selectivity in its perception of sound loudness: it is most sensitive in the 1000-1500Hz frequency band, which is the key frequency band for the transition between vowel resonance peaks and voiced sounds; and towards low and high frequencies, the perceptual sensitivity gradually decreases, forming a "bell-shaped" response curve. If the same weight is given to all sub-bands, the coding strategy will deviate from the real perception law of the human ear. Therefore, the present invention uses a Gaussian function to dynamically generate the perceptual weights of each sub-band.

[0054] Specifically, the perceptual weight of each subband satisfies the expression: .

[0055] in, Indicates the The perceptual weight of each sub-band; Indicates the The central Bark value of the sub-band, Indicates the preset benchmark Bark value; is a Gaussian attenuation factor used to control the deviation of the perceptual weight from the reference Bark value as the frequency changes The decay rate.

[0056] For the benchmark Bark value , when the frequency components 1000–1500 Hz belong to the same sub-band, the reference Bark value Set to the Bark value of the sub-band center where the frequency component 1000–1500Hz is located. When the frequency component 1000–1500Hz belongs to different sub-bands, the base Bark value Set to the mean of the Bark values ​​at the center of the sub-band where the frequency component 1000–1500 Hz is located. , the experience value is 3.0, which means that when the Bark distance deviates When the value exceeds 3 units, the perception weight decays to below about 60%, which is consistent with the rapid decline of high-frequency / low-frequency sensitivity of the human ear. In other embodiments, the implementer can adjust the value according to the actual application scenario. and To adapt to different auditory needs, but it is necessary to ensure Cannot be 0.

[0057] Furthermore, the comprehensive perceptual masking strength of the speech frame is obtained according to the total masking threshold and perceptual weight of each subband: .

[0058] in, Indicates the The comprehensive perceptual masking strength of the speech frame is used to reflect the The ability of a speech frame to mask noise; Indicates the The first speech frame The total masking threshold for subbands; Indicates the number of subbands; Indicates the The perceptual weight of each subband.

[0059] S3. For each positive segment in the A-law thirteen-fold line, determine the sensitivity weight of each positive segment according to the comprehensive perceptual masking strength of the speech frame and the sequence number of the positive segment, determine the quantization relaxation coefficient of each positive segment according to the comprehensive perceptual masking strength of the speech frame and the sensitivity weight of each positive segment, adjust the range size of each positive segment according to the quantization relaxation coefficient, and quantize and encode the speech frame according to the adjusted range size.

[0060] It should be noted that the quantization step size structure of the A-law thirteen-fold line quantizer is a fixed lookup table mode and is not adjusted as the speech content changes. In frames where speech energy is concentrated and the spectrum is complex (such as vowels and voiced sounds), the same quantization accuracy as that of silent or unvoiced consonant segments is still used, resulting in over-fine quantization in some areas that are imperceptible to the human ear, wasting bit resources. The comprehensive perceptual masking strength of the present invention can reflect the ability of the speech frame as a whole to mask noise. Therefore, the present invention dynamically adjusts the range size of each segment in the A-law thirteen-fold line quantizer according to the comprehensive perceptual masking strength, thereby realizing the adjustment of the quantization step size, so as to achieve the technical effect of improving the compression efficiency and subjective listening quality of speech coding while being compatible with the G.711 standard.

[0061] It should be further explained that when the comprehensive perceptual masking intensity is greater, it means that the speech energy in the speech frame is more concentrated and the spectrum is more complex. For example, it is in a vowel segment, a voiced segment or a strong resonance area. At this time, the main sound component can effectively mask the quantization noise near it, allowing a larger quantization error without causing distortion perceptible to the human ear. On the contrary, when the comprehensive perceptual masking intensity is smaller, it indicates that the speech is in a sensitive area of ​​clear consonants, transition sounds or near silence. At this time, the masking effect is weak, and fine quantization must be maintained to protect high-frequency details and speech clarity to avoid "harsh sounds" or recognition errors.

[0062] Specifically, the comprehensive perceptual masking strength is normalized to eliminate its dimension effect. The normalized expression is: .

[0063] in, Indicates the The normalized comprehensive perceptual masking strength of the speech frames; Represents the first The comprehensive perceptual masking strength of each speech frame; represents the reference masking level. In the present invention, dB, this is because when dB usually corresponds to voiceless consonants, transition sounds or speech segments close to silence, and fine quantization needs to be maintained to protect speech clarity and intelligibility; Indicates the preset maximum masking level. In the present invention, dB, because this value corresponds to the typical upper limit of high-energy speech (such as vowels, plosives, and strong resonance) in telephone communication systems. Speech frames exceeding this value have extremely strong masking capabilities, allowing maximum quantization compression. Implementers can also set it according to actual implementation conditions. as well as , but must meet ; represents the minimum function, represents the maximum function, 、 Used to limit The range is between [0,1].

[0064] Furthermore, since the positive and negative semi-axes of the A-law are symmetrical about the origin, the present invention takes the 8 positive segments of the positive semi-axes of the A-law as an example for explanation. For the 8 positive segments, the initial range size of each segment is recorded as 、 、…、 , whose value is determined by the standard A-law function.

[0065] According to the normalized comprehensive perceptual masking strength of the speech frame and the sequence number of the positive segment, the sensitivity weight of each positive segment is determined: .

[0066] in, Indicates the The first paragraph Sensitivity weight under speech frames; Indicates the sequence number of the positive segment; Indicates the The normalized comprehensive perceptual masking strength of the speech frames.

[0067] It should be noted that, in the static case without considering the masking effect, the low amplitude segment ( Small) corresponds to the small signal area, and the human ear is extremely sensitive to its changes (such as voiceless consonants and transition sounds), so it has a high degree of perception sensitivity; while the high amplitude segment ( Large) corresponds to the large signal area, which has strong energy and relatively low human ear resolution, so it has low sensitivity. However, in actual speech, the masking effect will dynamically change the perception sensitivity. When the speech energy is concentrated and the spectrum is complex, that is, When the amplitude is large, the low-amplitude component may be effectively masked by the main sound, and its sensitivity should be appropriately reduced; when the speech is sparse or in a silent transition zone, that is, When the amplitude is small, the low amplitude segment must maintain high sensitivity to prevent the quantization noise exposure. Therefore, the present invention uses the normalized result of the sequence number of the positive segment As the base, As an index, the dynamic adjustment of sensitivity weight is realized by using the form of gamma transformation. The smaller it is, the weaker the masking ability is. The closer it is to 1, the The closer the value is to , so the sensitivity weight near , the sensitivity weight of the low amplitude segment is large, maintaining high sensitivity, requiring fine quantization to protect speech details, and the sensitivity weight of the high amplitude segment is small, allowing coarse quantization. When the normalized comprehensive perceptual masking strength The larger the value, the stronger the masking ability. The smaller, the The greater the increase in The value of The larger the basis, the greater the sensitivity weight At the same time, due to the nonlinear characteristics of gamma transform, at larger In the case of The smaller the time, the The greater the increase in the The bigger the The smaller the increase is, the lower the amplitude segment ( Smaller) sensitivity weight The decrease is more significant, which increases its range and thus increases its quantization step appropriately, avoiding the bit waste caused by "over-protection" in a strong masking environment. Larger) sensitivity weight The decline is relatively gentle, and still maintains a certain coarse quantization characteristic, which is in line with the physiological law that the human ear is not sensitive to high amplitude segments. When it is close to 1, it indicates that the speech frame energy is extremely strong and the spectrum is complex. The main sound component has a very strong masking ability. At this time, even the quantization noise of the low-amplitude segment is almost completely masked. Therefore, the overall quantization can be relaxed to maximize the compression efficiency.

[0068] Furthermore, the quantization relaxation coefficient of each positive segment is determined according to the normalized comprehensive perceptual masking strength of the speech frame and the sensitivity weight of each positive segment: .

[0069] in, Indicates the The first paragraph The quantization relaxation coefficient under each speech frame is used to control the quantization granularity of each segment of the speech frame. When the quantization granularity is greater than 1, it means that the positive segment can be moderately coarsely quantized. When the quantization granularity is equal to 1, it means that the original accuracy is maintained. is a preset gain coefficient used to limit the step size change to prevent audible distortion caused by excessive adjustment. The value range of is [0.01, 0.05] to ensure smooth adjustment of the quantization step size and avoid audible distortion caused by excessive adjustment. In other embodiments, the implementer can set the value of the gain coefficient within the range of the value of the gain coefficient according to actual implementation; represents the normalized overall perceptual masking strength of the th speech frame; represents the sensitivity weight of the th positive segment under the th speech frame.

[0070] In the formula, the sensitivity weight reflects the compressibility of the th positive segment, and the lower the sensitivity weight , the more “compressible” the th positive segment is, and at this time is larger, and the range of the positive segment should be increased so that the quantization step size of the positive segment is also increased. At the same time, the adjustment only takes effect when the masking ability is strong, so the normalized overall perceptual masking strength is introduced as an enabling factor, and the weaker the adjustment ability is when is smaller, and the stronger the adjustment ability is when is larger.

[0071] Further, the adjusted range size of each positive segment is determined according to the quantization relaxation coefficient: .

[0072] wherein, represents the adjusted range size of the th positive segment under the th speech frame; represents the quantization relaxation coefficient of the th positive segment under the th speech frame; represents the initial range size of the th positive segment. is used to normalize , so as to ensure that the sum of the adjusted range sizes of all positive segments is 1, that is, the input domain remains [0, 1], which is compatible with the standard A-law decoder, avoids sample overflow.

[0073] Further, the adjusted range of each positive segment and the quantization step size are determined according to the adjusted range size of each positive segment, including: For the th positive segment, when , the adjusted range of the th positive segment is , and the quantization step size is ; when , the adjusted range of the th positive segment is , the quantization step size is ;when When, The adjusted range of the positive segment is , the quantization step size is .

[0074] Furthermore, the speech frames are quantized and encoded according to the adjusted ranges of the positive segments, including: For the first The first speech frame The time domain data are normalized and the results are used Indicates. OK The symbol is searched based on the adjusted range of all positive segments. The positive segment to which it belongs is quantized according to the quantization step size of the positive segment. Perform uniform quantization, calculate its quantization level, and generate 8-bit nonlinear PCM codewords by combining the sign bit. Each time domain data in a speech frame is quantized and encoded to achieve speech frame encoding.

[0075] It should be noted that this invention achieves the technical paradigm of "intelligent adaptation on the encoding side and full compatibility on the decoding side" by dynamically adjusting the segmented parameters of the A-law thirteen-fold quantizer. The encoder adaptively optimizes the quantization granularity based on the speech content, while the decoder still uses the standard G.711 inverse quantization table without any modification, allowing direct access to PSTN or IP networks. This significantly improves compression efficiency without sacrificing subjective speech quality.

Claims

1. An artificial intelligence-based smart phone customer service interaction method for electric power enterprises, characterized in that: include: Real-time acquisition of voice frames during telephone customer service interactions and conversion of voice frames into the frequency domain; Obtaining a main masker of the speech frame, determining a total masking threshold of the main masker for each frequency component of the speech frame, and determining a comprehensive perceptual masking strength of the speech frame based on the total masking threshold at each frequency component of the speech frame in the frequency domain; For each positive segment in the A-law thirteen-fold curve, the sensitivity weight of each positive segment is determined according to the comprehensive perceptual masking strength of the speech frame and the sequence number of the positive segment, the quantization relaxation coefficient of each positive segment is determined according to the comprehensive perceptual masking strength of the speech frame and the sensitivity weight of each positive segment, the range size of each positive segment is adjusted according to the quantization relaxation coefficient, and the speech frame is quantized and encoded according to the adjusted range size.

2. The method for interactive customer service of electric power enterprises based on artificial intelligence according to claim 1, characterized in that: The step of determining the comprehensive perceptual masking strength of the speech frame according to the total masking threshold of each frequency component of the speech frame in the frequency domain comprises: Based on the Bark scale values ​​of each frequency component of the speech frame in the frequency domain, the frequency components are divided into several Bark subbands of equal width. For any subband, the average of the total masking thresholds at all frequency components contained in the subband is obtained as the total masking threshold of the subband. The perceptual weight of each sub-band is dynamically generated using a Gaussian function. The total masking threshold of the sub-band is weighted and summed according to the perceptual weight of each sub-band to obtain the comprehensive perceptual masking strength of the speech frame.

3. The method for interactive customer service of electric power enterprise based on artificial intelligence according to claim 2, characterized in that: The perceptual weight of the sub-band satisfies the expression: ； in, Indicates the The perceptual weight of each sub-band; Indicates the The central Bark value of the sub-band, Indicates the preset benchmark Bark value; is the Gaussian attenuation factor.

4. The method for interactive customer service of electric power enterprises based on artificial intelligence according to claim 1, characterized in that: The sensitivity weights of the positive segments satisfy the expression: ； in, Indicates the The first paragraph Sensitivity weight under speech frames; Indicates the sequence number of the positive segment; Indicates the The normalized comprehensive perceptual masking strength of the speech frames.

5. The method for interactive customer service of electric power enterprise based on artificial intelligence according to claim 1, characterized in that: The quantization relaxation coefficient satisfies the expression: ； in, Indicates the The first paragraph Quantization relaxation coefficient under speech frames; is the preset gain coefficient; Indicates the The normalized comprehensive perceptual masking strength of the speech frames; Indicates the The first paragraph Sensitivity weight under each speech frame.

6. The method for interactive customer service of electric power enterprise smart phone based on artificial intelligence according to claim 1, characterized in that: The adjusting of the range of each positive segment according to the quantization relaxation coefficient includes: For each positive segment, the product of the initial range size of each positive segment and the quantization relaxation coefficient is calculated, and the product is normalized to obtain the adjusted range size of each positive segment.

7. The method for interactive customer service of electric power enterprise via smart phone based on artificial intelligence according to claim 1, characterized in that: The quantizing and encoding the speech frame according to the adjusted range size includes: Determine the adjusted range and quantization step of each positive segment according to the adjusted range size of each positive segment; The first speech frame The time domain data are normalized and the results are used express; confirm The symbol is searched based on the adjusted range of all positive segments. The positive segment to which it belongs is quantized according to the quantization step size of the positive segment. Perform uniform quantization, calculate its quantization level, and combine it with the sign bit to generate an 8-bit nonlinear PCM code word to achieve quantization coding.

8. The method for interactive customer service of electric power enterprise via smart phone based on artificial intelligence according to claim 7, characterized in that: Determining the adjusted range and quantization step size of each positive segment according to the adjusted range size of each positive segment includes: For the positive segments, and the quantization step size is ;when When, The adjusted range of the positive segment is ;when When, The adjusted range of the positive segment is ;when When, The adjusted range of the positive segment is ,in Indicates the The first paragraph The adjusted range size under each speech frame is Indicates the The first paragraph The adjusted range size under each speech frame.

9. The method for interactive customer service of electric power enterprise based on artificial intelligence according to claim 1, characterized in that: The obtaining of the main masker sound of the speech frame includes: For each speech frame, the amplitudes corresponding to all frequency components of the speech frame in the spectrum are arranged in ascending order of frequency to form an amplitude sequence, and the local maxima in the amplitude sequence are obtained. All local maxima are sorted in descending order, and the frequency components corresponding to the first K local maxima are each used as a primary masker, where K is the preset number of primary maskers.

10. The method for interactive customer service via smart phone of a power enterprise based on artificial intelligence according to claim 4 or 5, characterized in that: The normalized comprehensive perceptual masking strength satisfies the expression: ； in, Indicates the The normalized comprehensive perceptual masking strength of the speech frames; Represents the first The comprehensive perceptual masking strength of each speech frame; represents the reference masking level; Indicates the preset maximum masking level; represents the minimum function, Represents the maximum function.

Images