CN112233686B - Voice data processing method of NVOCPLUS high-speed broadband vocoder - Google Patents
Voice data processing method of NVOCPLUS high-speed broadband vocoder Download PDFInfo
- Publication number
- CN112233686B CN112233686B CN202011047245.1A CN202011047245A CN112233686B CN 112233686 B CN112233686 B CN 112233686B CN 202011047245 A CN202011047245 A CN 202011047245A CN 112233686 B CN112233686 B CN 112233686B
- Authority
- CN
- China
- Prior art keywords
- voice
- signal
- value
- parameter
- voice data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000003672 processing method Methods 0.000 title claims abstract description 16
- 238000013139 quantization Methods 0.000 claims abstract description 33
- 238000001228 spectrum Methods 0.000 claims abstract description 22
- 239000013598 vector Substances 0.000 claims abstract description 22
- 238000012545 processing Methods 0.000 claims abstract description 14
- 230000015572 biosynthetic process Effects 0.000 claims abstract description 10
- 238000003786 synthesis reaction Methods 0.000 claims abstract description 10
- 230000002194 synthesizing effect Effects 0.000 claims abstract description 7
- 238000004458 analytical method Methods 0.000 claims abstract description 6
- 238000011084 recovery Methods 0.000 claims abstract description 4
- 230000001629 suppression Effects 0.000 claims description 26
- 238000000034 method Methods 0.000 claims description 25
- 238000001914 filtration Methods 0.000 claims description 24
- 230000005284 excitation Effects 0.000 claims description 16
- 238000001514 detection method Methods 0.000 claims description 12
- 238000005311 autocorrelation function Methods 0.000 claims description 11
- 238000005516 engineering process Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 230000021615 conjugation Effects 0.000 claims description 3
- 238000005314 correlation function Methods 0.000 claims description 3
- 230000003111 delayed effect Effects 0.000 claims description 3
- 230000001419 dependent effect Effects 0.000 claims description 3
- 210000005069 ears Anatomy 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 230000003595 spectral effect Effects 0.000 claims description 3
- 230000001502 supplementing effect Effects 0.000 claims description 3
- 238000012546 transfer Methods 0.000 claims description 3
- 108090000623 proteins and genes Proteins 0.000 description 13
- 238000005070 sampling Methods 0.000 description 6
- 230000001755 vocal effect Effects 0.000 description 6
- 238000004891 communication Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 230000006835 compression Effects 0.000 description 3
- 238000007906 compression Methods 0.000 description 3
- 210000000056 organ Anatomy 0.000 description 3
- 230000000737 periodic effect Effects 0.000 description 3
- 230000007613 environmental effect Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 210000004704 glottis Anatomy 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000002238 attenuated effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000001208 nuclear magnetic resonance pulse sequence Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000005236 sound signal Effects 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
- G10L19/04—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using predictive techniques
- G10L19/16—Vocoder architecture
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
- G10L19/02—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using spectral analysis, e.g. transform vocoders or subband vocoders
- G10L19/0212—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using spectral analysis, e.g. transform vocoders or subband vocoders using orthogonal transformation
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
- G10L19/04—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using predictive techniques
- G10L19/26—Pre-filtering or post-filtering
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L21/0216—Noise filtering characterised by the method used for estimating noise
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L21/0216—Noise filtering characterised by the method used for estimating noise
- G10L21/0232—Processing in the frequency domain
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/78—Detection of presence or absence of voice signals
- G10L25/84—Detection of presence or absence of voice signals for discriminating voice from noise
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/93—Discriminating between voiced and unvoiced parts of speech signals
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Quality & Reliability (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
Abstract
The invention relates to a voice data processing method of an NVOCPLUS high-speed broadband vocoder, which comprises the following steps: step 1, an encoding end carries out initialization configuration and analysis processing on an original voice digital signal, judges whether the current voice signal is voice, and calculates a fundamental tone period and unvoiced and voiced numerical parameters of each sub-band after extracting fundamental tones in the voice if the current voice signal is voice; step 2, extracting and quantizing parameters of the line spectrum pair, the pitch value, the gain parameter, the residual compensation gain and the codebook vector to obtain a voice quantization parameter; and 3, after extracting the voice quantization parameters in the step 2, synthesizing the voice quantization parameters into voice, increasing the voice quality through noise pressing, and performing voice reconstruction after the parameter recovery fails or the voice synthesis fails. The invention can provide good voice quality and strong adaptability to speech in low speed and in the application of losing voice frequency below 300 Hz.
Description
Technical Field
The invention belongs to the technical field of vocoder digital voice compression, and particularly relates to a voice data processing method of a high-speed wideband vocoder of a NVOCPLUS.
Background
With the rapid development of communication technology, frequency and resources are precious, and compared with an analog voice communication system, a digital voice communication system has the characteristics of strong anti-interference performance, good confidentiality, easy integration and the like, and a low-speed vocoder plays an important role in the digital voice communication system.
At present, most speech coding algorithms are established on the basis of acoustic models of human vocal organs. The vocal organs of a person consist of the glottis, the vocal tract and other auxiliary organs. The actual speech generation process is that the vibration generated by glottis is modulated by sound channel filter and radiated via mouth and nose, etc., and can be expressed as the following formula
s(n)＝h(n)*e(n)
Wherein s (n) represents a voice signal, h (n) is a unit impulse response of a vocal tract filter, and e (n) is a glottal vibration signal.
In order to clearly represent speech signals, the glottal and the vocal tract can be respectively described from the spectral characteristics, and how to efficiently quantize the characteristic parameters of the glottal and the vocal tract is the target to be achieved by the algorithm of parameter coding.
Vocoders belong to the class of parametric coding, which is a method of compressing the digital representation of a speech signal to recover the most similar speech to the original speech with fewer bits (bits). With the explosive increase in the efficiency of digital signal processing hardware, vocoders have been used in large quantities in addition to the accelerated research of vocoders.
Different from the existing NVOC narrowband vocoder, the existing NVOC narrowband vocoder comprises two code rates of 2.4kbps and 2.2kbps (used for encryption), the channel FEC code rate is 1.2kbps, and both the voice codec and the FEC perform encoding and decoding on a frame with 20 milliseconds of 8K sampling. The NVOC wideband vocoder realizes high speed of 12.2kbps, more narrow band (40 + bit) after compression (200 + bit), and the encoded data carries more information which is helpful for restoring sound.
In the field of the existing wideband vocoder, because the compression ratio of the voice coding is not high, the following problems still exist on the premise of obtaining better tone quality and accuracy: (1) The gene parameters are extracted by utilizing the time domain correlation, so that the calculation is easy to be wrong; (2) Because the sound is not subjected to noise reduction, the extracted sound parameters are inaccurate when noise exists; and (4) neglecting the compatibility with low-speed narrow-band vocoder.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a voice data processing method of an NVOCPLUS high-speed broadband vocoder, which has reasonable design, high voice quality and strong adaptability to the dialect.
The invention solves the practical problem by adopting the following technical scheme:
a voice data processing method of NVOCPLUS high-speed wideband vocoder comprises the following steps:
step 1, an encoding end carries out initialization configuration and analysis processing on an original voice digital signal, firstly carries out noise suppression processing on the original voice digital signal, then judges whether the current voice signal is voice, if the current voice signal is voice, calculates a pitch period and unvoiced and voiced numerical parameters of each sub-band after extracting a pitch in the voice;
step 2, extracting and quantizing parameters of a line spectrum pair, a base pitch value, a gain parameter, a residual compensation gain and a codebook vector on the basis of the numerical parameters of the pitch period, the unvoiced sound and the voiced sound calculated in the step 1 to obtain a sound quantization parameter;
and 3, after the voice quantization parameters in the step 2 are extracted, synthesizing the voice quantization parameters into voice, increasing the voice quality through noise pressing, and performing voice reconstruction after the parameter recovery fails or the voice synthesis fails.
The step 1 specifically includes:
(1) Noise suppression processing is carried out on the original voice digital signal S (n) to obtain voice data S after noise suppression 1 (n) and sound spectrum characteristics of 0 to 4000Hz of the original data S (n);
(2) Judging whether the current voice signal after noise suppression processing is voice by adopting VAD activation detection technology to obtain voice data S 2 (n)；
(3) Extracting voice data S 2 A fundamental tone of (n);
(4) And calculating the parameters of the pitch period and the unvoiced and voiced values of each sub-band.
Moreover, the specific steps in step (1) of step 1 include:
(1) a high-pass filter is adopted to remove direct current components from voice data, improve high-frequency components and attenuate low frequency;
(2) windowing signals, adopting a Hamming window with the window length N, and obtaining energy distribution on a frequency spectrum through overlapped Fourier transform to obtain voice data S after noise suppression 1 (n), noise suppression result parameters and sound spectrum characteristics of 0-4000 Hz of the original voice digital signal S (n).
Moreover, the specific method in the step 1 and the step (2) is as follows:
according to the auditory characteristics of human ears, the voice data S after noise suppression 1 (n) filtering the sub-band and calculating the level of the sub-band signal, estimating the signal-to-noise ratio according to the formula shown below, comparing the signal-to-noise ratio with a preset threshold value, and judging whether the signal is correctWhether the preceding speech signal is speech:
wherein, a is the signal level value of the current frame, and b is the current signal level value estimated from the previous frames;
moreover, the specific method in the step 1 and the step (3) is as follows:
dialogue data S using a low-pass filter with a cut-off frequency of B Hz 2 (n) low-pass filtering is carried out, after the voice data after low-pass filtering is carried out by adopting a second-order inverse filter, the self-phase function of the output signal of the second-order inverse filtering is calculated according to the following formula, and fundamental tones are extracted:
wherein N is the window length of the mentioned window function in the step (1) in the step 1, S w (i) And (3) outputting a signal for the second-order inverse filtering in the step (1) and the step (3).
Moreover, the specific steps of the step 1, the step (4) comprise:
(1) dividing the frequency domain into 5 frequency bands with equal intervals of 0-4000, wherein the frequency bands are respectively [0-500] Hz,
[500-1000] Hz, [1000-2000] Hz, [2000-3000] Hz, [3000-4000] Hz, and the autocorrelation function of the bandpass signal in each interval is calculated using the following formula:
where "t" is a continuous time argument, "τ" is an input signal delay ". Cndot. * f * () To take conjugation;
(2) taking the average value of the product of two values of the same time function at the moment t and t + a as the function of time t, wherein the function is the measurement of the similarity between the signal and the delayed signal, when the delay time is zero, the average value of the signal is the mean square value, the value of the mean square value is the maximum value at the moment, and the maximum value of the function is used as the intensity of voiced sound to calculate the value of the unvoiced sound of each sub-band;
further, the specific steps of step 2 include:
(1) Filtering the voice data subjected to noise suppression by adopting a high-pass filter with cut-off frequency of A Hz to obtain S 3 (n), windowing, calculating autocorrelation coefficients, solving line spectrum pair parameters by using a Levinson-Durbin recursive algorithm, and performing parameter quantization on the obtained line spectrum pair parameters by using a three-level vector quantization scheme;
(2) Quantizing the pitch value calculated in the step (3) in the step 1: linearly mapping integer intervals containing pitch values to [ 0-z ]]In the above, the number of z is m 1 Bit representation;
(3) Voice data S detected by voice in step 1 (2) 2 (n) obtaining a prediction error signal r (n) without the influence of formants through a second order inverse filter, wherein the coefficient of the second order filter is a 1 、a 2 1, the gain parameter is expressed by RMS of r (n), and the quantization is completed in a logarithmic domain;
(4) Quantizing the maximum value obtained by the correlation function of the band-pass signal value after the frequency domain segmentation of the step 1 and the step 4 into m 2 A bit;
(5) Calculating residual compensation gain, calculating linear prediction coefficient by using quantized LSF parameter to form prediction error filter for input speech S 2 (n) filtering to obtain a residual signal, wherein the length of the residual signal is 160 points;
(6) Using a Hamming window with the window length of 160 points to window the prediction residual error, supplementing 0 to 512 points to a windowed signal, performing 512-point complex FFT on the windowed signal, and finding out a Fourier transform value corresponding to the first x-order harmonic by using a spectrum peak point detection algorithm;
(7) Let P be the quantized fundamental, given that the initial position of the ith harmonic is 512i/P, peak point detection finds the maximum peak value within 512/P frequency samples centered around the initial position of each subharmonic, the width being truncated to an integer; the harmonic times of searching are limited to be the smaller of x and P/4; the harmonic corresponding coefficients are thenNormalization, using an m for the x-dimensional vector 3 ∈[0,48]Quantizing the vector codebook of bits to obtain m 3 ∈[0,48]A bit.
Moreover, the specific method for synthesizing the voice quantization parameter into the voice in the step 3 is as follows:
by dividing into several frequency bands to form excitations, adding them and passing them through a synthesis filter to obtain synthesized speech, and then post-filtering the synthesized speech to obtain decoded synthesized speech data, where the synthesis filter H (z) and the post-filter H pf The z-transform transfer function of (z) is as follows:
H(z)＝1/A(z)
wherein A (z) is 1-az -1 A is the filter coefficient, z in all the above equations is a complex variable having real and imaginary parts, let z = e jw γ =0.56, β =0.75, μ being determined by the reflection coefficient, the value of μ being dependent on
Furthermore, the method further comprises the following steps before the step 3:
and initializing and configuring a decoding end, wherein the initialization and configuration comprises rate selection, parameters of an algorithm of the decoding end and filter coefficients.
Before the step of initializing and configuring the decoding end, the method also comprises the following steps:
expanding the linear prediction coefficient, the excitation gain parameter and the gene period parameter in the step 3 to respectively obtain expanded parameters;
the method comprises the following specific steps:
(1) Enlarging the gain value quantization interval obtained in the step 3, and calculating the molecular frame to obtain an expansion excitation gain parameter;
(2) Respectively subtracting the LSP parameter mean value from the current frame LSP parameter and the quantized previous frame LSP parameter in the step 3 to obtain the mean value-removed vectors, and respectively recording the mean value-removed vectors
And
as the input of the hierarchical vector quantization, the quantization is carried out, namely the expanded LSP linear prediction parameters are obtained;
(3) Enlarging the gene value quantization bit obtained in the step 3, calculating a molecular frame once every two subframes, namely dividing the interval set in the step 3 (2) into two parts of the corresponding subframes, respectively calculating the maximum value and the index i according to the autocorrelation function in the step 2 (3), and respectively using the maximum value and the index i
And (5) carrying out normalization to obtain the period parameter of the expanded gene.
The invention has the advantages and beneficial effects that:
1. the invention provides excellent voice quality under low speed condition, provides good voice quality in application losing voice frequency under 300Hz and has strong adaptability to speech by analyzing the continuity of voice in time domain and the correlation of voice in frequency domain.
2. The invention extracts the actual parameters in two stages, more accurately extracts the parameters, improves the sound quality and saves the calculation resources for users.
3. The invention is different from the low-speed vocoder, expands the linear prediction coefficient, the excitation gain parameter and the gene period parameter, and ensures that the sound reconstruction degree is much higher than the narrow band because the coding result carries more information even under the condition of the error code existing in poor channel quality.
4. The invention inhibits the noise through the noise inhibiting function, improves the accuracy of the extracted sound parameters when the noise exists, and ensures the sound quality.
5. The invention adopts the codebook based on various local conversation training, and has strong adaptability to dialects.
6. The invention is developed based on standard codes, is standard and sustainable, and is easy to be transplanted to various hardware platforms.
Drawings
Fig. 1 is a working principle diagram of the present invention.
Detailed Description
The embodiments of the invention will be described in further detail below with reference to the accompanying drawings:
the input parameter of the voice data processing method of the NVOCPLUS high-speed broadband vocoder is a linear PCM voice digital signal with the sampling rate of 8000Hz (the number of voice signal samples collected per second) and the resolution of 16 bits; in the time domain, every 20 milliseconds of analysis, and in the frequency domain, a plurality of frequency bands are divided into 0-4000 for analysis.
A voice data processing method of NVOCPLUS high-speed wideband vocoder, as shown in fig. 1, includes the following steps:
step 1, initializing and configuring a coding end, wherein the initialization and configuration comprises rate selection, parameters and coefficients used by the coding end and a filter coding end algorithm;
step 2, the coding end initializes configuration and analysis processing of the original voice digital signal, firstly, noise suppression processing is carried out on the original voice digital signal, then whether the current voice signal is voice or not is judged, if the current voice signal is voice, fundamental tone in the voice is extracted, and then fundamental tone period and unvoiced and voiced numerical parameters of each sub-band are calculated;
the step 2 specifically comprises the following steps:
(1) Noise suppression: carrying out noise suppression processing on the original voice digital signal S (n) to obtain voice data S after noise suppression 1 (n) and sound spectrum characteristics of 0 to 4000Hz of the original data S (n);
the step 2, the step (1), comprises the following specific steps:
(1) a high-pass filter is adopted to remove direct-current components from voice data, improve high-frequency components and attenuate low frequency;
(2) windowing signal, using window length ofN, by overlapping Fourier transform to obtain energy distribution on frequency spectrum, obtaining voice data S after noise suppression 1 (n), noise suppression result parameters and sound spectrum characteristics of 0-4000 Hz of the original voice digital signal S (n).
(2) Voice detection: judging whether the current voice signal after noise suppression processing is voice by adopting VAD activation detection technology to obtain voice data S 2 (n)；
The specific method of the step 2 and the step (2) is as follows:
according to the auditory characteristics of human ears, the voice data S after noise suppression 1 (n) filtering the sub-band and calculating the level of the sub-band signal, estimating the signal-to-noise ratio according to the following formula, and comparing the signal-to-noise ratio with a preset threshold value to further judge whether the current speech signal is voice:
wherein, a is the signal level value of the current frame, and b is the current signal level value estimated from the previous frames;
(3) Gene estimation first stage: extracting voice data S 2 A pitch of (n);
the specific method of the step (2) and the step (3) comprises the following steps:
dialogue data S using low-pass filter with cut-off frequency of B Hz 2 (n) low-pass filtering, and after the low-pass filtered voice data is inversely filtered by adopting a second-order inverse filter, calculating a self-phase function of an output signal of the second-order inverse filter according to the following formula, and extracting fundamental tone:
wherein N is the window length of the mentioned window function in the step (1) in the step 1, S w (i) And (3) outputting a signal for the second-order inverse filtering in the step (2) and the step (3).
In the present embodiment, in the frequency domain, the speech signal has a peak valueThe frequency of the sum peak is the multiple relation of the fundamental tones, and possible fundamental tone values or fundamental tone range values are preliminarily calculated; in the time domain, speech has short-term autocorrelation, and if the original signal has periodicity, its autocorrelation function also has periodicity, and the periodicity is the same as that of the original signal. And peaks occur at integer multiples of the period. The unvoiced sound signal is non-periodic, its autocorrelation function is attenuated with the increase of frame length, the voiced sound is periodic, its autocorrelation function has peak value on the integral multiple of gene period, and the low-pass filter whose cut-off frequency is B Hz is used to make voice data S 2 (n) low-pass filtering is carried out to remove the influence of the high-frequency signal on the fundamental tone extraction, then a second-order inverse filter is adopted to carry out inverse filtering on the voice data after the low-pass filtering to remove the influence of formants, an autocorrelation function of an output signal of the second-order inverse filtering is calculated, and the fundamental tone is extracted:
in the autocorrelation function of the frame, the pitch value of the frame is the sampling rate/frame length at which the maximum value appears, after the first maximum value is removed.
(4) A first stage of multi-subband voiced and unvoiced decision: calculating the value of unvoiced and voiced sounds of each sub-band
The step 2, the step (4) comprises the following specific steps:
(1) dividing the frequency domain into 5 frequency bands at equal intervals of 0-4000, wherein the frequency bands are respectively [0-500] Hz, [500-1000] Hz, [1000-2000] Hz, [2000-3000] Hz, [3000-4000] Hz, and calculating the autocorrelation function of the bandpass signals in each interval by using the following formula:
wherein ". Dot" is a convolution operator, (.) * f * () To obtain conjugation;
(2) taking the average value of the product of two values of the same time function at the moment t and t + a as the function of delay time t, wherein the average value is the measure of the similarity between the signal and the delayed signal, when the delay time is zero, the average value becomes the mean square value of the signal, the value of the mean square value is the maximum at the moment, and the maximum value of the function is taken as the voiced intensity to calculate the unvoiced and voiced numerical value of each sub-band;
step 3, extracting and quantizing parameters of a line spectrum pair, a base pitch value, a gain parameter, a residual compensation gain and a codebook vector on the basis of the numerical parameters of the pitch period, the unvoiced sound and the voiced sound calculated in the step 2 to obtain a sound quantization parameter;
the specific steps of the step 3 comprise:
(1) Filtering the voice data subjected to noise suppression by adopting a high-pass filter with cut-off frequency of A Hz to obtain S 3 (N), adding a Hamming window with the window length of N2, calculating an autocorrelation coefficient, solving a line spectrum pair parameter (namely a prediction parameter (namely an LSF parameter)) by using a Levinson-Durbin recursive algorithm, and performing parameter quantization on the obtained line spectrum pair parameter by using a three-level vector quantization scheme to obtain m 1 A bit;
(2) Quantizing the pitch value calculated in the step (3) in the step 2: linearly mapping integer intervals containing pitch values to [ 0-z ]]In the above, the number of z is m 2 Bit representation;
(3) Voice data S detected by voice in step 2 (2) 2 (n) obtaining a prediction error signal r (n) without the influence of formants through a second order inverse filter, wherein the coefficient of the second order filter is a 1 、a 2 1, the excitation gain parameter is expressed by RMS (mean value Ping Fangen of the square) of r (n), and the quantization is completed in a logarithmic domain;
(4) Quantizing the maximum value (namely the unvoiced and voiced state value) obtained by the correlation function of the band-pass signal value after the frequency domain segmentation of the step (4) in the step 2 into m 3 A bit;
(5) Calculating spectral compensation gain, calculating linear prediction coefficient by using quantized LSF parameter to form prediction error filter for input speech S 2 (n) filtering to obtain a residual signal, wherein the length of the residual signal is 160 points;
(6) Windowing the prediction residual error by using a Hamming window with the window length of 160 points, supplementing 0 to 512 points to a windowed signal, performing 512-point complex FFT on the windowed signal, and finding a Fourier transform value corresponding to the first x-order harmonic by using a spectrum peak point detection algorithm;
(7) Let P be the quantized pitch, given an initial position of the ith harmonic of 512i/P, peak detection looks for the largest peak within 512/P frequency samples centered around the initial position of each subharmonic, this width being truncated to an integer. The number of harmonics to be searched is limited to the smaller of x and P/4. The coefficients corresponding to these harmonics are then normalized, using an m for this x-dimensional vector 4 ∈[0,48]Quantizing the vector codebook of bits to obtain m 4 ∈[0,48]A bit.
because the high-speed vocoder has larger bandwidth, more bits can be carried, and in order to improve the accuracy of gene detection, the resolution reliability of pitch detection is improved and the calculation is carried out by using a molecular frame.
In the present embodiment, 12.2kbps is taken as an example, and the meaning of the sub-frame here is expressed as every 40 sampling points (5 ms data).
The specific steps of the step 4 comprise:
(1) Enlarging the gain value quantization interval obtained in the step 3, and calculating the molecular frame to obtain an expansion excitation gain parameter;
(2) Respectively recording the vectors obtained by subtracting the LSP parameter mean value from the current frame LSP parameter and the quantized previous frame LSP parameter in the step 3 and removing the mean value as
And
as the input of the hierarchical vector quantization, the quantization is carried out, namely the expanded LSP linear prediction parameters are obtained;
(3) In the enlargement step 3Quantizing bit of the obtained gene value, calculating a molecular frame once every two subframes, namely dividing the interval set in the step (2) in the step (3) into two parts of corresponding subframes, respectively solving a maximum value and an index i according to the autocorrelation function in the step (3) in the step (2), and respectively using the maximum value and the index i
And (5) carrying out normalization to obtain the period parameter of the expanded gene.
In this embodiment, the extended parameters are the result of adding the extended information bits on the premise of the original parameters. Not solely as a new parameter derived from the encoding.
Step 5, initializing and configuring a decoding end, wherein the initialization and configuration comprises the speed selection (2.2 kbps or 2.4 kbps), and parameters of an algorithm of the decoding end, filter coefficients and the like;
and 6, after the voice quantization parameters in the steps 3 and 4 are extracted, synthesizing the voice quantization parameters into voice, enhancing the voice quality through noise pressing, and performing voice reconstruction after the parameter recovery fails or the voice synthesis fails.
The specific method of the step 6 comprises the following steps:
the result after each frame signal coding is a numerical value formed by equivalently converting a line spectrum pair, gain, a gene period, unvoiced voiced sound and a vector codebook into bits. Among these parameters, the noise suppression result parameter determines whether the audio data segment with excessive environmental noise is replaced by mute or comfortable environmental sound, the pitch period and the unvoiced/voiced value determine the excitation source for synthesizing the speech signal at the decoding end, and according to the step 1 (4) at the encoding end, the unvoiced/voiced signal covers 5 frequency bands, so that the excitation is formed by dividing into several frequency bands, and then the excitation is added and passed through the synthesis filter and post-filtering, so as to obtain the decoded synthesized speech data. If the frame is an unvoiced frame, namely the unvoiced and voiced values bit are all 0, a random number is used as an excitation source, if the frame is a voiced frame, a periodic pulse sequence is selected to generate the excitation source through an all-pass filter, the amplitude of the excitation source is weighted by a gain parameter, and the length of a sampling point depends on the size of a gene period. All-pass filter H 1 (z) synthesis filter H 2 (z) andpost filter H pf The z-transform transfer function of (z) is as follows:
wherein A (z) is 1-az -1 A is filter coefficient, obtained by P transformation of linear prediction parameters in step 4, P is high-level mathematical transformation, z in all the formulas is complex variable and has real part and imaginary part, and z = e can be ensured jw γ =0.56, β =0.75, μ being determined by the reflection coefficient, the value of μ being dependent on
It can be understood that the algorithm of the encoding and decoding is corresponding, the input parameter format of the decoding end and the output parameter format of the encoding end are also corresponding, the decoder decodes a frame and outputs 160 sampling values, and the calling needs to be unified with the encoder speed.
It should be emphasized that the embodiments described herein are illustrative and not restrictive, and thus the present invention includes, but is not limited to, the embodiments described in this detailed description, as well as other embodiments that can be derived by one skilled in the art from the teachings herein, and are within the scope of the present invention.
Claims (8)
1. A voice data processing method of NVOCPLUS high-speed broadband vocoder is characterized in that: the method comprises the following steps:
step 1, an encoding end carries out initialization configuration and analysis processing on an original voice digital signal, firstly carries out noise suppression processing on the original voice digital signal, then judges whether the current voice signal is voice, if the current voice signal is voice, calculates a fundamental tone period and unvoiced and voiced numerical parameters of each sub-band after extracting fundamental tones in the voice;
step 2, extracting and quantizing parameters of a line spectrum pair, a base pitch value, a gain parameter, a residual compensation gain and a codebook vector on the basis of the numerical parameters of the pitch period, the unvoiced sound and the voiced sound calculated in the step 1 to obtain a sound quantization parameter;
step 3, after extracting the voice quantization parameters in the step 2, synthesizing the voice quantization parameters into voice, increasing the voice quality through noise pressing, and performing voice reconstruction after the parameter recovery fails or the voice synthesis fails;
the method also comprises the following steps before the step 3:
initializing and configuring a decoding end, wherein the initialization and configuration comprises rate selection, and the initialization and configuration of parameters and filter coefficients of an algorithm of the decoding end;
before the step of initializing and configuring the decoding end, the method also comprises the following steps:
expanding the linear prediction coefficient, the excitation gain parameter and the pitch period parameter in the step 3 to respectively obtain an expanded parameter;
the method comprises the following specific steps:
(1) Expanding the gain value quantization interval obtained in the step 3, and calculating the molecular frame to obtain an expanded excitation gain parameter;
(2) Respectively subtracting the LSP parameter mean value from the current frame LSP parameter and the quantized previous frame LSP parameter in the step 3 to obtain the mean value-removed vectors, and respectively recording the mean value-removed vectors
And
as the input of the hierarchical vector quantization, the quantization is carried out, namely the expanded LSP linear prediction parameters are obtained;
(3) Enlarging the quantized bit of the pitch value obtained in the step 3, calculating a molecular frame once every two subframes, namely dividing the interval set in the step 3 (2) into two parts of the corresponding subframes, respectively solving the maximum value and the index i according to the autocorrelation function in the step 2 (3), and respectively using the maximum value and the index i
And carrying out normalization to obtain the extended pitch period parameters.
2. The method as claimed in claim 1, wherein the voice data processing method of the NVOCPLUS high-speed wideband vocoder comprises: the step 1 specifically comprises the following steps:
(1) Carrying out noise suppression processing on the original voice digital signal S (n) to obtain voice data S after noise suppression 1 (n) sound spectral characteristics of 0 to 4000Hz of the original data S (n);
(2) Judging whether the current voice signal after noise suppression processing is voice by adopting VAD activation detection technology to obtain voice data S 2 (n)；
(3) Extracting voice data S 2 A pitch of (n);
(4) And calculating the parameters of the pitch period and the unvoiced and voiced values of each sub-band.
3. The method as claimed in claim 2, wherein the voice data processing method of the NVOCPLUS high-speed wideband vocoder comprises: the step 1, the step (1), comprises the following specific steps:
(1) a high-pass filter is adopted to remove direct-current components from voice data, improve high-frequency components and attenuate low frequency;
(2) windowing signals, adopting a Hamming window with the window length N, and obtaining energy distribution on a frequency spectrum through overlapped Fourier transform to obtain voice data S after noise suppression 1 (n), noise suppression result parameter, and sound spectrum characteristic of 0-4000 Hz of the original voice digital signal S (n).
4. The method as claimed in claim 2, wherein the voice data processing method of the NVOCPLUS high-speed wideband vocoder comprises: the specific method of the step (1) and the step (2) comprises the following steps:
according to the auditory characteristics of human ears, the voice data S after noise suppression 1 (n) filtering the sub-band and calculating the level of the sub-band signal, estimating the signal-to-noise ratio according to the following formula, and comparing the signal-to-noise ratio with a preset threshold value to further judge whether the current speech signal is voice:
where a is the signal level value of the current frame and b is the current signal level value estimated from the previous frames.
5. The method as claimed in claim 3, wherein the voice data processing method of the NVOCPLUS high-speed wideband vocoder comprises: the specific method of the step 1 and the step (3) comprises the following steps:
dialogue data S using a low-pass filter with a cut-off frequency of B Hz 2 (n) low-pass filtering, and after the low-pass filtered voice data is inversely filtered by adopting a second-order inverse filter, calculating a self-phase function of an output signal of the second-order inverse filter according to the following formula, and extracting fundamental tone:
wherein N is the window length of the mentioned window function in the step 1, S w (i) And (3) outputting a signal for the second-order inverse filtering in the step (1) and the step (3).
6. The method as claimed in claim 2, wherein the voice data processing method of the NVOCPLUS high-speed wideband vocoder comprises: the step 1, the step (4) comprises the following specific steps:
(1) dividing the frequency domain into 5 frequency bands at equal intervals of 0-4000, wherein the frequency bands are respectively [0-500] Hz, [500-1000] Hz, [1000-2000] Hz, [2000-3000] Hz, [3000-4000] Hz, and calculating the autocorrelation function of the bandpass signals in each interval by using the following formula:
where "t" is a continuous time argument, "τ" is an input signal delay ". The" is a convolution operator ".) * f * () To take conjugation;
(2) the average value of the product of two values of the same time function at the instant t and t + a is taken as the function of the time t, which is the measure of the similarity between the signal and the delayed signal, when the delay time is zero, the average square value of the signal is obtained, the value of the average square value is the maximum value at the moment, and the maximum value of the function is taken as the voiced sound intensity to calculate the unvoiced and voiced sound value of each sub-band.
7. The method as claimed in claim 1, wherein the voice data processing method of the NVOCPLUS high-speed wideband vocoder comprises: the specific steps of the step 2 comprise:
(1) Filtering the voice data subjected to noise suppression by adopting a high-pass filter with cut-off frequency of A Hz to obtain S 3 (n), windowing, calculating autocorrelation coefficients, solving line spectrum pair parameters by using a Levinson-Durbin recursive algorithm, and performing parameter quantization on the obtained line spectrum pair parameters by using a three-level vector quantization scheme;
(2) Quantizing the pitch value calculated in the step (3) in the step 1: linearly mapping integer intervals containing pitch values to [ 0-z ]]In the above, the number of z is m 1 Bit representation;
(3) Voice data S detected by voice in step 1 (2) 2 (n) obtaining a prediction error signal r (n) without the influence of the formants through a second order inverse filter, wherein the coefficient of the second order inverse filter is a 1 、a 2 1, the gain parameter is expressed by RMS of r (n), and the quantization is completed in a logarithmic domain;
(4) Quantizing the maximum value obtained by the correlation function of the band-pass signal value after the frequency domain segmentation of the step 1 and the step 4 into m 2 A bit;
(5) Computing residual compensation gain, computing linear prediction coefficient by using quantized LSF parameter to form prediction error filter for input speech S 2 (n) filtering to obtain a residual signal, wherein the length of the residual signal is 160 points;
(6) Windowing the prediction residual error by using a Hamming window with the window length of 160 points, supplementing 0 to 512 points to a windowed signal, performing 512-point complex FFT on the windowed signal, and finding a Fourier transform value corresponding to the first x-order harmonic by using a spectrum peak point detection algorithm;
(7) Setting P as quantized fundamental tone, setting the initial position of the ith harmonic as 512i/P, and searching the maximum peak value with the width within 512/P frequency samples and the initial position of each subharmonic as the center by peak point detection, wherein the width is truncated into an integer; the harmonic number of the search is limited to the smaller of x and P/4; the coefficients corresponding to the harmonics are then normalized, using an m for this x-dimensional vector 3 ∈[0,48]Bit vector codebook is quantized to obtain m 3 ∈[0,48]A bit.
8. The method as claimed in claim 1, wherein the voice data processing method of the NVOCPLUS high-speed wideband vocoder comprises: the specific method for synthesizing the voice quantization parameter into the voice in the step 3 comprises the following steps:
by dividing into several frequency bands to form excitations, adding them and passing them through a synthesis filter to obtain synthesized speech, and then post-filtering the synthesized speech to obtain decoded synthesized speech data, where the synthesis filter H (z) and the post-filter H pf The z-transform transfer function of (z) is as follows:
H(z)＝1/A(z)
wherein A (z) is 1-az -1 A is the filter coefficient, z in all the above equations is a complex variable having real and imaginary parts, let z = e jw γ =0.56, β =0.75, μ being determined by the reflection coefficient, the value of μ being dependent on
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011047245.1A CN112233686B (en) | 2020-09-29 | 2020-09-29 | Voice data processing method of NVOCPLUS high-speed broadband vocoder |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011047245.1A CN112233686B (en) | 2020-09-29 | 2020-09-29 | Voice data processing method of NVOCPLUS high-speed broadband vocoder |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112233686A CN112233686A (en) | 2021-01-15 |
| CN112233686B true CN112233686B (en) | 2022-10-14 |
Family
ID=74120236
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011047245.1A Active CN112233686B (en) | 2020-09-29 | 2020-09-29 | Voice data processing method of NVOCPLUS high-speed broadband vocoder |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112233686B (en) |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6094629A (en) * | 1998-07-13 | 2000-07-25 | Lockheed Martin Corp. | Speech coding system and method including spectral quantizer |
| CN103050121A (en) * | 2012-12-31 | 2013-04-17 | 北京迅光达通信技术有限公司 | Linear prediction speech coding method and speech synthesis method |
| CN109729553B (en) * | 2017-10-30 | 2021-12-28 | 成都鼎桥通信技术有限公司 | Voice service processing method and device of LTE (Long term evolution) trunking communication system |
| CN108597529A (en) * | 2018-01-22 | 2018-09-28 | 北京交通大学 | A kind of police digital cluster system air interface speech monitoring system and method |
| CN111107501A (en) * | 2018-10-25 | 2020-05-05 | 普天信息技术有限公司 | Group call service processing method and device |
| CN111243610A (en) * | 2020-01-19 | 2020-06-05 | 福建泉盛电子有限公司 | Method for realizing intercommunication of different vocoder and mobile stations in digital intercommunication system |
-
2020
- 2020-09-29 CN CN202011047245.1A patent/CN112233686B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN112233686A (en) | 2021-01-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Shrawankar et al. | Techniques for feature extraction in speech recognition system: A comparative study | |
| US5450522A (en) | Auditory model for parametrization of speech | |
| JP4308345B2 (en) | Multi-mode speech encoding apparatus and decoding apparatus | |
| JP3277398B2 (en) | Voiced sound discrimination method | |
| KR100348899B1 (en) | The Harmonic-Noise Speech Coding Algorhthm Using Cepstrum Analysis Method | |
| KR101378696B1 (en) | Determining an upperband signal from a narrowband signal | |
| JP4624552B2 (en) | Broadband language synthesis from narrowband language signals | |
| JP2002516420A (en) | Voice coder | |
| Milner et al. | Speech reconstruction from mel-frequency cepstral coefficients using a source-filter model | |
| EP1395978A2 (en) | Method and apparatus for speech reconstruction in a distributed speech recognition system | |
| JP2001525079A (en) | Audio coding system and method | |
| JP3687181B2 (en) | Voiced / unvoiced sound determination method and apparatus, and voice encoding method | |
| KR100216018B1 (en) | Method and apparatus for encoding and decoding of background sounds | |
| Bhatt | Simulation and overall comparative evaluation of performance between different techniques for high band feature extraction based on artificial bandwidth extension of speech over proposed global system for mobile full rate narrow band coder | |
| CN112270934B (en) | Voice data processing method of NVOC low-speed narrow-band vocoder | |
| JPH10105195A (en) | Pitch detecting method and method and device for encoding speech signal | |
| Robinson | Speech analysis | |
| CN112233686B (en) | Voice data processing method of NVOCPLUS high-speed broadband vocoder | |
| Shahnaz et al. | Robust pitch estimation at very low SNR exploiting time and frequency domain cues | |
| KR20060067016A (en) | Apparatus and method for voice coding | |
| CN114550741A (en) | Semantic recognition method and system | |
| EP0713208B1 (en) | Pitch lag estimation system | |
| Schlien et al. | Acoustic tube interpolation for spectral envelope estimation in artificial bandwidth extension | |
| Srivastava | Fundamentals of linear prediction | |
| Tan et al. | Speech feature extraction and reconstruction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |