JP2010186190A - Quantized lsp parameter dynamic feature extractor and quantized lsp parameter dynamic feature extracting method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To change the coding mode of a sound source section and efficiently code a speech signal without coding/transmitting mode information. <P>SOLUTION: An adder 606 calculates the difference between a smooth quantized LPS parameter at present processing unit time and another smooth quantized LPS parameter of the preceding processing unit time. A square sum calculating means 607 calculates the square sum of the difference for each order about output of the adder 606. An AR type average calculating means 611 calculates the average LSP parameter in a noise section. An adder 612 calculates the difference of the quantized LSP parameter at the present processing unit time and the average quantized LSP parameter in the noise section for each order. A square sum calculating means 613 calculates the square sum for each order about the adder 612. A speech section detecting means 619 determines whether or not an input signal in the present processing unit time be in the speech section. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

The present invention relates to a low bit rate speech encoding apparatus in a mobile communication system or the like that encodes and transmits a speech signal, and in particular, CELP (Code Excited Linear Prediction) that expresses a speech signal separately into vocal tract information and sound source information. ) Type speech encoding apparatus and the like. In particular, the present invention relates to a quantized LSP parameter dynamic feature extractor and a quantized LSP parameter dynamic feature extraction method for extracting dynamic features of parameters used for detecting a speech section .

  In the fields of digital mobile communication and voice storage, voice coding apparatuses for compressing voice information and coding with high efficiency for effective use of radio waves and storage media are used. Among them, a method based on the CELP (Code Excited Linear Prediction) method has been widely put into practical use at medium and low bit rates. For the CELP technology, see M.R. Schroeder and B.S. Atal: “Code-Excited Linear Prediction (CELP): High-quality Speech at Very Low Bit Rates”, Proc. ICASSP-85, 25.1.1, pp.937-940, 1985 ".

  The CELP speech coding method divides speech into a certain frame length (about 5 ms to 50 ms), performs speech linear prediction for each frame, and knows the prediction residual (excitation signal) by linear prediction for each frame. The encoding is performed using the adaptive code vector and the noise code vector having the waveform. The adaptive code vector is from an adaptive code book that stores drive excitation vectors generated in the past, and the noise code vector is from a noise code book that stores a vector having a predetermined number of predetermined shapes prepared in advance. Selected and used. As the noise code vector stored in the noise code book, a random noise sequence vector, a vector generated by arranging several pulses at different positions, or the like is used.

  FIG. 13 shows a configuration example of a basic block of a conventional CELP encoding apparatus. In this CELP encoding apparatus, LPC analysis / quantization, pitch search, noise codebook search, and gain codebook search are performed using the input digital signal, and the quantized LPC code (L) and pitch period ( P), the noise codebook index (S), and the gain codebook index (G) are transmitted to the decoder.

  However, in the above conventional speech coding apparatus, it is necessary to deal with voiced speech, unvoiced speech, background noise, etc. with one kind of noise codebook, and encodes all these input signals with high quality. It was difficult.

The present invention has been made in view of the above circumstances, and can achieve multi-mode excitation coding without newly transmitting mode information, particularly for determination of voiced / unvoiced intervals. In addition, a multi-mode speech encoding apparatus and speech decoding apparatus that can also determine speech sections / non-speech sections and that can further improve the improvement in encoding / decoding performance by multi-mode conversion. The purpose is to provide. The present invention also provides a quantized LSP parameter dynamic feature extractor and a quantized LSP parameter dynamic feature extraction method for extracting dynamic features of parameters for accurately detecting a speech section of an input signal. Objective.

The present invention performs mode determination using static / dynamic features of quantization parameters representing spectral characteristics, and codes of driving sound sources based on mode determination results indicating speech / non-speech and voiced / unvoiced intervals. The mode of various codebooks used for conversion was changed. In addition, mode information used for encoding is used for decoding, and various codebook modes used for decoding are switched. In the present invention, the distance between the average quantized LSP parameter and the current quantized LSP parameter is calculated.

According to the present invention, the mode is switched between excitation coding and / or post-decoding processing using static and dynamic features in the quantized data of the parameter representing the spectral characteristics, so that mode information is not newly transmitted. In addition, multi-mode excitation coding can be achieved. In particular, since it is possible to determine voice / non-speech intervals in addition to determination of voiced / unvoiced intervals, a speech coding apparatus and speech that can further improve the degree of improvement in coding performance by multi-mode conversion A decoding device can be provided. In addition, according to the present invention, a quantized LSP parameter dynamic feature extractor and a quantized LSP parameter dynamic feature extraction capable of extracting a dynamic feature with a parameter for accurately detecting a speech section of an input signal. Can provide a method .

FIG. 1 is a block diagram showing the configuration of a speech coding apparatus according to Embodiment 1 of the present invention. The block diagram which shows the structure of the audio | voice decoding apparatus in Embodiment 2 of this invention. Flow chart showing the flow of speech coding processing in Embodiment 1 of the present invention The flowchart which shows the flow of the audio | voice decoding process in Embodiment 2 of this invention. The block diagram which shows the structure of the audio | voice signal transmission apparatus and receiver in Embodiment 3 of this invention. The block diagram which shows the structure of the mode selector in Embodiment 4 of this invention. The block diagram which shows the structure of the multi-mode post-processor in Embodiment 5 of this invention. The flowchart which shows the flow of the mode selection process of the front | former stage in Embodiment 4 of this invention Flow chart showing the flow of the subsequent mode selection process in Embodiment 4 of the present invention Flow chart showing the overall flow of mode selection processing in Embodiment 4 of the present invention Flowchart showing the flow of the previous mode selection process in the fifth embodiment of the present invention Flowchart showing the flow of subsequent mode selection processing in Embodiment 5 of the present invention The block diagram which shows the structure of the conventional audio | voice encoding apparatus.

  According to a first aspect of the present invention, there is provided first encoding means for encoding at least one parameter representing vocal tract information included in an audio signal, and at least one type indicating sound source information included in the audio signal. The second encoding means capable of encoding the parameters in several modes and the mode switching of the second encoding means based on the dynamic characteristics of the specific parameter encoded by the first encoding means A mode switching unit and a synthesizing unit that synthesizes an input speech signal with a plurality of types of parameter information encoded by the first and second encoding units are employed.

  According to this configuration, since the encoding result of the first encoding unit is used to determine the encoding mode of the second encoding unit, the second encoding is performed without adding new information for indicating the mode. The means can be made multi-mode and the encoding performance can be improved.

  According to a second aspect of the present invention, in the first aspect, the mode switching unit performs mode switching of the second encoding unit that encodes the driving sound source using a quantization parameter that represents a spectral characteristic of speech. Take.

  According to this configuration, in a speech encoding apparatus that independently encodes a parameter representing spectral characteristics and a parameter representing driving excitation, the encoding of the driving excitation is made multi-mode without increasing new transmission information. Encoding performance can be improved.

  According to a third aspect of the present invention, in the second aspect, the mode switching of the means for encoding the driving sound source using the static and dynamic features of the quantization parameter representing the spectral characteristics of the speech. The structure which performs is taken.

  According to this configuration, since the stationary noise part can be detected by using the dynamic feature, the coding performance for the stationary noise part can be improved by the multi-mode driving excitation coding.

  According to a fourth aspect of the present invention, in the second and third aspects, the mode switching means switches the mode of the means for encoding the driving sound source using the quantized LSP parameter.

  According to this configuration, the present invention can be easily applied to a CELP system that uses LSP parameters as parameters representing spectral characteristics.

  The fifth aspect of the present invention employs a configuration in which, in the fourth aspect, the mode switching means performs mode switching of the means for encoding the driving sound source using the static and dynamic features of the quantized LSP parameter. .

  According to this configuration, the present invention can be easily applied to the CELP system that uses the LSP parameter as a parameter representing the spectrum characteristic, and the LSP parameter that is a frequency domain parameter is used, so that the spectral steadiness can be determined satisfactorily. Encoding performance for stationary noise can be improved.

  According to a sixth aspect of the present invention, in the fourth and fifth aspects, the mode switching means determines the stationarity of the quantized LSP using past and current quantized LSP parameters, and the current quantized LSP. And a means for switching the mode of the means for encoding the driving sound source based on the determination result.

  According to this configuration, since the encoding of the driving sound source can be performed by switching between the stationary noise unit, the unvoiced speech unit, and the voiced speech unit, the encoding is performed by preparing the driving excitation encoding mode corresponding to each unit. Performance can be improved.

  According to a seventh aspect of the present invention, there is provided means for decoding at least one parameter representing vocal tract information included in an audio signal and decoding at least one parameter representing sound source information included in the audio signal. Second decoding means for converting, mode switching means for switching the mode of the second decoding means based on the dynamic characteristics of the specific parameter decoded by the first decoding means, and the first and second And a synthesizing unit that decodes the audio signal based on a plurality of types of parameter information decoded by the decoding unit.

  According to this configuration, the signal encoded by the speech encoding device of the first aspect can be decoded.

  According to an eighth aspect of the present invention, in the seventh aspect, the mode switching unit performs mode switching of the second decoding unit that decodes the driving sound source using a quantization parameter that represents a spectral characteristic of the speech. Take.

  According to this configuration, the signal encoded by the speech encoding apparatus according to the second aspect can be decoded.

  According to a ninth aspect of the present invention, in the seventh aspect, the mode switching of the means for decoding the driving sound source using the static and dynamic features of the quantization parameter representing the spectral characteristics of the speech. The structure which performs is taken.

  According to this configuration, the signal encoded by the speech encoding apparatus of the third aspect can be decoded.

  The tenth aspect of the present invention employs a configuration in which, in the seventh aspect, the mode switching means performs mode switching of the means for decoding the driving sound source using the quantized LSP parameter.

  According to this configuration, it is possible to decode the signal encoded by the speech encoding device according to the fourth aspect.

  The eleventh aspect of the present invention employs a configuration in which, in the seventh aspect, the mode switching means performs mode switching of the means for decoding the driving sound source using the static and dynamic features of the quantized LSP parameter. .

  According to this configuration, the signal encoded by the speech encoding apparatus according to the fifth aspect can be decoded.

  According to a twelfth aspect of the present invention, in the seventh aspect, the mode switching means uses means for determining the continuity of the quantized LSP using past and current quantized LSP parameters, and the current quantized LSP. And means for switching the mode of the means for decoding the driving sound source based on the determination result.

  According to this configuration, the signal encoded by the speech encoding apparatus according to the sixth aspect can be decoded.

  A thirteenth aspect of the present invention employs a configuration in any one of the seventh to twelfth aspects, in which post-processing switching for a decoded signal is performed based on a determination result of a determination unit.

  According to this configuration, it is possible to decode a signal encoded by the multimode audio encoding device according to any one of the first to sixth aspects, and further to encode an audio signal in a steady background noise environment by post-processing. Performance can be improved.

  According to a fourteenth aspect of the present invention, there is provided means for calculating an inter-frame change of a quantized LSP parameter, means for calculating an average quantized LSP parameter in a frame in which the quantized LSP parameter is stationary, and the average quantum And a means for calculating a distance between the quantized LSP parameter and the current quantized LSP parameter.

  According to this configuration, it is possible to extract a dynamic feature for accurately detecting a voice section of an input signal.

  A fifteenth aspect of the present invention employs a configuration including means for calculating linear prediction residual power from quantized LSP parameters and means for calculating intervals between adjacent order quantized LSP parameters.

  According to this configuration, it is possible to extract the features of the peaks and valleys of the spectrum envelope of the input signal, and it is possible to extract static features for detecting a section that is highly likely to be a speech section.

  A sixteenth aspect of the present invention includes the dynamic feature extractor according to the fourteenth aspect and the static feature extractor according to the fifteenth aspect, and the quantized LSP parameter extracted by the dynamic feature extractor. A configuration is adopted in which a speech section is detected using at least one of a dynamic feature and a static feature of a quantized LSP parameter extracted by the static feature extractor.

  According to this configuration, it is possible to accurately separate the speech section and the stationary noise section.

  A seventeenth aspect of the present invention includes the speech section detector of the sixteenth aspect and voiced / unvoiced determination means, and at least one of the detection result of the voice section detector and the determination result of the voiced / unvoiced determination means. A configuration is adopted in which mode determination is performed using this information.

  According to this configuration, it is possible to realize a multi-mode configuration using the information for separating the voice interval / noise interval and the voiced interval / unvoiced interval.

  According to an eighteenth aspect of the present invention, the voiced / unvoiced judging means includes: a means for calculating a reflection coefficient from a quantized LSP parameter; and a means for calculating a linear prediction residual power from the quantized LSP parameter. A configuration using information extracted by a static feature extractor of parameters is adopted.

  According to this configuration, voiced / unvoiced determination can be performed with high accuracy.

  According to a nineteenth aspect of the present invention, in the first aspect, a mode switching means is constituted by the mode selector.

  According to this configuration, excitation encoding can be performed in multimode according to the characteristics of the input speech.

  According to a twentieth aspect of the present invention, in the seventh aspect, a mode switching means is configured by the mode selector.

  According to this configuration, an audio signal encoded using the encoding device according to the nineteenth aspect can be decoded.

  According to a twenty-first aspect of the present invention, a determination unit that determines whether or not a speech section is used by using a decoded LSP parameter, an FFT processing unit that performs FFT processing of a signal, and a phase spectrum obtained by the FFT processing Phase spectrum randomizing means for randomizing according to the determination result of the determination means, amplitude spectrum smoothing means for smoothing the amplitude spectrum obtained by the FFT processing according to the determination result, and the phase spectrum randomizing means And IFFT processing means for performing inverse FFT processing on the phase spectrum randomized by the above and the phase spectrum smoothed by the amplitude spectrum smoothing means.

  According to this configuration, post-processing can be performed in multimode, and in particular, the subjective quality of the stationary noise section can be improved.

  According to a twenty-second aspect of the present invention, in the twenty-first aspect, the frequency of the phase spectrum to be randomized is determined using the average amplitude spectrum in the past non-voice interval in the voice interval, and the auditory weight is determined in the non-voice interval. A configuration is employed in which the phase spectrum to be randomized and the frequency of the amplitude spectrum to be smoothed are determined using the average value of the amplitude spectrum of all frequencies in the subscript region.

  According to this configuration, it is possible to adaptively perform post-processing of the voice section and the noise section.

  A twenty-third aspect of the present invention employs a configuration in which, in the twenty-first aspect, noise generated using an average amplitude spectrum in a past non-voice section is superimposed in the voice section.

  According to this configuration, it is possible to improve the perceptual quality of the decoded speech signal having stationary background noise.

  According to a twenty-fourth aspect of the present invention, in the twenty-first aspect, the speech section detection means in the sixteenth aspect determines whether or not it is the speech section, and the average amplitude spectrum and the current amplitude spectrum in the past non-speech section. And the size of the difference.

  According to this configuration, since it is possible to detect a case where the power of the decoded signal suddenly increases, it is possible to cope with a case where a detection error by the speech section detection means in the sixteenth aspect occurs.

  A twenty-fifth aspect of the present invention employs a configuration in which, in the thirteenth aspect, post-processing is performed using the multimode post-processor in the twenty-first aspect.

  According to this configuration, it is possible to realize a speech decoding apparatus that can improve the subjective quality of the stationary noise section in particular by performing post-processing in multimode.

  A twenty-sixth aspect of the present invention employs a configuration including the speech coding apparatus according to the first aspect and the speech decoding apparatus according to the seventh aspect.

  According to this configuration, it is possible to realize a speech encoding / decoding device including the speech encoding device according to the first aspect and the speech decoding device according to the seventh aspect.

  According to a twenty-seventh aspect of the present invention, an audio input device that converts an audio signal into an electrical signal, an A / D converter that converts a signal output from the audio input device into a digital signal, and the A / D conversion A voice encoding device according to any one of the first to sixth aspects for encoding a digital signal output from the transmitter, and an RF for performing modulation processing or the like on the encoded information output from the speech encoding device A configuration including a modulator and a transmission antenna that converts a signal output from the RF modulator into a radio wave and transmits the signal is employed.

  According to this configuration, it is possible to realize an audio signal transmitting device including the audio encoding device according to any one of the first to sixth aspects, and it is possible to perform high-quality low bit rate audio encoding.

  According to a twenty-eighth aspect of the present invention, there is provided a receiving antenna that receives a received radio wave, an RF demodulator that demodulates a signal received by the receiving antenna, and a decoder that decodes information obtained by the RF demodulator. The speech decoding device according to any of the seventh to thirteenth aspects, a D / A converter for D / A converting a digital speech signal decoded by the speech decoding device, and output by the D / A converter And an audio output device that converts an electrical signal to an audio signal.

  According to this configuration, an audio signal receiving device including the audio decoding device according to any of the seventh to thirteenth aspects can be realized, and a signal transmitted from the audio signal transmitting apparatus according to the twenty-seventh aspect can be received and decoded. Can be

  A twenty-ninth aspect of the present invention employs a configuration including at least one of the sound signal transmitting apparatus according to the twenty-seventh aspect and the sound signal receiving apparatus according to the twenty-eighth aspect.

  According to this configuration, a mobile station apparatus provided with the audio signal transmitting apparatus according to the 27th aspect and / or the audio signal receiving apparatus according to the 28th aspect can be realized, and a mobile station apparatus with high sound quality can be realized.

  The 30th aspect of the present invention employs a configuration comprising at least one of the audio signal transmitting apparatus of the 27th aspect and the audio signal receiving apparatus of the 28th aspect.

  According to this configuration, a base station apparatus including the audio signal transmitting apparatus according to the twenty-seventh aspect and / or the audio signal receiving apparatus according to the twenty-eighth aspect can be realized, and a high-quality base station apparatus can be realized.

  A thirty-first aspect of the present invention provides a computer with a procedure for determining the continuity of a quantized LSP using past and current quantized LSP parameters, and a procedure for determining voicedness using the current quantized LSP. A machine-readable recording medium recording a program for executing a mode switching procedure of a procedure for encoding a driving sound source based on a result determined by the procedure.

  According to this recording medium, a function equivalent to that of the speech coding apparatus according to the sixth aspect can be provided by installing the recorded program in the computer.

  A thirty-second aspect of the present invention provides a computer with a procedure for determining the continuity of a quantized LSP using past and current quantized LSP parameters, and a procedure for determining voicedness using the current quantized LSP. A procedure for switching the mode of the procedure for decoding the driving sound source based on the result determined by the procedure, and a procedure for switching the post-processing procedure for the decoded signal based on the result determined by the procedure. A machine-readable recording medium on which a program to be executed is recorded.

  According to this recording medium, a function equivalent to that of the speech decoding apparatus according to the thirteenth aspect can be provided by installing the recorded program in the computer.

  A thirty-third aspect of the present invention employs a configuration for performing mode switching of a mode for encoding a driving sound source using static and dynamic features of quantization parameters representing the spectral characteristics of speech.

  According to this configuration, since the stationary noise part can be detected by using the dynamic feature, the coding performance for the stationary noise part can be improved by the multi-mode driving excitation coding.

  A thirty-fourth aspect of the present invention employs a configuration for performing mode switching of a mode for decoding a driving sound source using static and dynamic features of a quantization parameter representing a spectral characteristic of speech.

  According to this configuration, it is possible to provide a decoding method capable of decoding a signal encoded by the speech encoding method of the thirty-third aspect.

  A thirty-fifth aspect of the present invention is a speech decoding method according to the thirty-fourth aspect, comprising a step of performing post-processing on a decoded signal and a step of switching the post-processing step based on mode information. Take.

  According to this configuration, it is possible to provide a speech decoding method capable of further improving the stationary noise quality of a signal decoded using the speech decoding method of the 34th aspect.

  A thirty-sixth aspect of the present invention includes a step of calculating a change in the quantized LSP parameter between frames, a step of calculating an average quantized LSP parameter in a frame in which the quantized LSP parameter is stationary, and the average quantum And a step of calculating a distance between the quantized LSP parameter and the current quantized LSP parameter.

  According to this configuration, it is possible to extract a dynamic feature for accurately detecting a voice section of an input signal.

  A thirty-seventh aspect of the present invention employs a configuration including a step of calculating linear prediction residual power from quantized LSP parameters and a step of calculating intervals between adjacent order quantized LSP parameters.

  According to this configuration, it is possible to extract the features of the peaks and valleys of the spectrum envelope of the input signal, and it is possible to extract static features for detecting a section that is highly likely to be a speech section.

  A thirty-eighth aspect of the present invention comprises the dynamic feature extraction step in the thirty-sixth aspect and the static feature extraction step in the thirty-seventh aspect, and the quantized LSP extracted in the dynamic feature extraction step A configuration is adopted in which a speech section is detected using at least one of a dynamic feature of a parameter and a static feature of a quantized LSP parameter extracted in the static feature extraction step.

  According to this configuration, it is possible to accurately separate the speech section and the stationary noise section.

  A thirty-ninth aspect of the present invention employs a configuration in which mode determination is performed using a voice detection result obtained by the voice section detection method according to the thirty-eighth aspect.

  According to this configuration, it is possible to realize a multi-mode configuration using the information for separating the voice interval / noise interval and the voiced interval / unvoiced interval.

  According to a 40th aspect of the present invention, there is provided a determination step of determining whether or not a speech section is detected using a decoded LSP parameter, an FFT processing step of performing FFT processing of a signal, and a phase spectrum obtained by the FFT processing as described above. A phase spectrum randomizing step for randomizing according to a determination result in the determination step, an amplitude spectrum smoothing step for smoothing an amplitude spectrum obtained by the FFT processing according to the determination result, and the phase spectrum randomizing step And an IFFT processing step for performing an inverse FFT process on the phase spectrum randomized in step 1 and the phase spectrum smoothed in the amplitude spectrum smoothing step.

  According to this configuration, post-processing can be performed in multimode, and in particular, the subjective quality of the stationary noise section can be improved.

  Hereinafter, a speech encoding apparatus and the like according to an embodiment of the present invention will be described with reference to FIGS.

(Embodiment 1)
FIG. 1 shows the configuration of a speech encoding apparatus according to Embodiment 1 of the present invention.

  Input data including a digitized audio signal or the like is input to the preprocessor 101. The pre-processor 101 cuts the DC component, limits the bandwidth of the input data, etc. using a high-pass filter, a band-pass filter, etc., and outputs the result to the LPC analyzer 102 and the adder 105. The subsequent encoding process can be performed without performing any process in the pre-processor 101, but the encoding performance is improved by performing the process as described above.

  The LPC analyzer 102 performs linear prediction analysis, calculates a linear prediction coefficient (LPC), and outputs the linear prediction coefficient (LPC) to the LPC quantizer 103.

  The LPC quantizer 103 quantizes the input LPC, and outputs the quantized LPC to the synthesis filter 104 and the mode selector 105, and outputs a code L representing the quantized LPC to the decoder. Note that LPC quantization is generally performed by converting to LSP (Line Spectrum Pair) having good interpolation characteristics.

  The synthesis filter 104 constructs an LPC synthesis filter using the input quantized LPC. The synthesized filter is subjected to filter processing with the driving sound source signal output from the adder 114 as an input, and the synthesized signal is output to the adder 106.

  The mode selector 105 determines the mode of the noise codebook 109 using the quantized LPC input from the LPC quantizer 103.

  Here, the mode selector 105 also stores information on quantized LPC input in the past, and uses both the characteristics of the variation of the quantized LPC between frames and the features of the quantized LPC in the current frame. Make a selection. There are at least two types of modes, for example, a mode corresponding to a voiced voice part, a mode corresponding to an unvoiced voice part, a stationary noise part, and the like. In addition, the information used for mode selection does not need to be the quantized LPC itself, and it is more effective to use information converted into parameters such as quantized LSP, reflection coefficient, and linear prediction residual power.

  The adder 106 calculates an error between the preprocessed input data input from the preprocessor 101 and the synthesized signal and outputs the error to the auditory weighting filter 107.

  The auditory weighting filter 107 performs auditory weighting on the error calculated by the adder 106 and outputs the result to the error minimizer 108.

  The error minimizer 108 adjusts the noise codebook index Si, the adaptive codebook index (pitch period) Pi, and the gain codebook index Gi to the noise codebook 109, the adaptive codebook 110, and the gain codebook 111, respectively. The noise code vector, adaptive code vector, and noise generated by the noise codebook 109, the adaptive codebook 110, and the gain codebook 111 so that the perceptually weighted error input from the perceptual weighting filter 107 is minimized. A codebook gain and an adaptive codebook gain are determined, respectively, and a code S representing a noise code vector, P representing an adaptive code vector, and a code G representing gain information are output to a decoder.

  The random code book 109 stores a predetermined number of different random code vectors, and outputs a random code vector specified by the noise code vector index Si input from the error minimizer 108. The noise codebook 109 has at least two types of modes. For example, in a mode corresponding to a voiced voice part, a more pulsed noise code vector is generated, and it corresponds to an unvoiced voice part or a stationary noise part. In this mode, a more noisy noise code vector is generated. The noise code vector output from the noise codebook 109 is generated from one mode selected by the mode selector 105 out of the two or more modes, and is added after the noise codebook gain Gs is multiplied by the multiplier 112. Is output to the device 114.

  The adaptive codebook 110 performs buffering while sequentially updating drive excitation signals generated in the past, and uses the adaptive codebook index (pitch period (pitch lag)) Pi input from the error minimizer 108 to perform adaptive codebooking. Generate a vector. The adaptive code vector generated by adaptive codebook 110 is multiplied by adaptive codebook gain Ga by multiplier 113 and then output to adder 114.

  The gain codebook 111 stores a predetermined number of sets (gain vectors) of the adaptive codebook gain Ga and the noise codebook gain Gs, and uses the gain codebook index Gi input from the error minimizer 108. The adaptive codebook gain component Ga of the designated gain vector is output to the multiplier 113, and the noise codebook gain component Gs is output to the multiplier 112. If the gain codebook has a multi-stage configuration, it is possible to reduce the amount of memory required for the gain codebook and the amount of calculation required for the gain codebook search. Also, if the number of bits allocated to the gain codebook is sufficient, the adaptive codebook gain and the noise codebook gain can be independently scalar quantized.

  Adder 114 adds the noise code vector and adaptive code vector input from multipliers 112 and 113 to generate a drive excitation signal, and outputs it to synthesis filter 104 and adaptive codebook 110.

  In the present embodiment, only the noise codebook 109 has been converted to the multimode, but the quality can be further improved by converting the adaptive codebook 110 and the gain codebook 111 to the multimode. It is.

  Next, referring to FIG. 3, a processing flow of the speech coding method in the above embodiment will be described. In this description, the speech encoding process is performed for each predetermined time length processing unit (frame: about several tens of milliseconds in time length), and one frame is further divided into a short number of processing units (sub An example in which processing is performed for each frame) will be described.

  In step 301, all the memories such as the contents of the adaptive codebook, the synthesis filter memory, and the input buffer are cleared.

  Next, input data such as a voice signal digitized in step 302 is input for one frame, and an input data offset removal or band limitation is performed by applying a high-pass filter or a band-pass filter. The input data after the preprocessing is buffered in the input buffer and used for the subsequent encoding processing.

  Next, in step 303, LPC analysis (linear prediction analysis) is performed to calculate LPC coefficients (linear prediction coefficients).

  Next, in step 304, the LPC coefficient calculated in step 303 is quantized. Various methods for quantizing LPC coefficients have been proposed. However, when LSP parameters are converted into LSP parameters having good interpolation characteristics and predictive quantization using multi-stage vector quantization or inter-frame correlation is applied, the LPC coefficients can be efficiently quantized. For example, when one frame is divided into two subframes and processed, the LPC coefficient of the second subframe is quantized, and the LPC coefficient of the first subframe is quantized of the second subframe in the immediately preceding frame. This is determined by interpolation using the quantized LPC coefficient and the quantized LPC coefficient of the second subframe in the current frame.

  Next, in step 305, an auditory weighting filter that constructs auditory weighting on the preprocessed input data is constructed.

  Next, in step 306, a perceptual weighting synthesis filter that generates a perceptual weighting region composite signal from the driving sound source signal is constructed. This filter is a filter in which a synthesis filter and an auditory weighting filter are connected in cascade. The synthesis filter is constructed using the quantized LPC coefficients quantized in step 304, and the auditory weighting filter is calculated in step 303. Constructed using LPC coefficients.

  Next, in step 307, a mode is selected. Mode selection is performed using the dynamic and static features of the quantized LPC coefficients quantized in step 304. Specifically, a variation of the quantized LSP, a reflection coefficient calculated from the quantized LPC coefficient, a prediction residual power, and the like are used. A noise codebook search is performed according to the mode selected in this step. There are at least two modes selected in this step. For example, a two-mode configuration of voiced voice mode, unvoiced voice, and stationary noise mode can be considered.

  Next, in step 308, an adaptive codebook search is performed. The search of the adaptive codebook is to search for an adaptive code vector that generates an auditory weighted composite waveform that is closest to the waveform obtained by performing auditory weighting on the preprocessed input data. Minimal error between the signal filtered by the perceptual weighting filter constructed in step 305 and the signal filtered by the perceptual weighting synthesis filter constructed in step 306 using the adaptive code vector cut out from the adaptive codebook as the driving excitation signal The position to cut out the adaptive code vector is determined so that

  Next, in step 309, a noise codebook search is performed. The noise codebook search is performed by selecting a noise code vector that generates a driving sound source signal that generates an auditory weighted composite waveform that is closest to the waveform obtained by performing auditory weighting on input data after preprocessing. There is a search taking into account that the driving excitation signal is generated by adding the adaptive code vector and the noise code vector. Therefore, the drive excitation signal is generated by adding the adaptive code vector already determined in step 308 and the noise code vector stored in the noise codebook, and the generated drive excitation signal is constructed in step 306. The noise code vector is extracted from the noise codebook so that the error between the signal filtered by the auditory weighting synthesis filter and the signal obtained by filtering the preprocessed input data by the auditory weighting filter constructed in step 305 is minimized. Select. When processing such as pitch periodization is performed on the noise code vector, a search that takes that processing into consideration is also performed. In addition, this noise codebook has at least two modes. For example, in a mode corresponding to the voiced voice part, a search is performed using a noise codebook storing a more pulsed noise code vector. In a mode corresponding to an unvoiced voice part, a stationary noise part, etc., a search is performed using a noise codebook storing more noisy noise code vectors. In step 307, the mode of the noise codebook to be used during the search is selected.

  Next, in step 310, a gain codebook search is performed. The search of the gain codebook is performed by obtaining a set of the adaptive codebook gain and the noise codebook gain to be multiplied with respect to each of the adaptive code vector already determined in step 308 and the noise code vector determined in step 309. And the adaptive codebook gain multiplied adaptive code vector and the noise code gain multiplied noise code vector are added to generate a drive excitation signal, and the generated drive excitation signal is sent to step 306. Adaptive codebook gain and noise that minimize the error between the signal filtered by the perceptual weighting synthesis filter constructed in step 305 and the signal filtered by the perceptual weighting filter constructed in step 305 from the preprocessed input data A set of codebook gains is selected from the gain codebook.

  Next, in step 311, a driving sound source signal is generated. The driving excitation signal is selected in step 310 by the vector obtained by multiplying the adaptive code vector selected in step 308 by the adaptive codebook gain selected in step 310 and the noise code vector selected in step 309. It is generated by adding the vector multiplied by the noise codebook gain.

  Next, in step 312, the memory used in the subframe processing loop is updated. Specifically, the adaptive codebook is updated, the state of the auditory weighting filter and the auditory weighting synthesis filter is updated, and the like.

  The above steps 305 to 312 are processing in units of subframes.

  Next, in step 313, the memory used in the frame processing loop is updated. Specifically, the state of the filter used in the preprocessor, the update of the quantized LPC coefficient buffer, the update of the input data buffer, and the like are performed.

  Next, in step 314, encoded data is output. The encoded data is sent to the transmission line after being subjected to bit stream or multiplexing processing according to the transmission form.

  The above steps 302 to 304 and 313 to 314 are processing in units of frames. Further, the processing in units of frames and subframes is repeated until there is no input data.

(Embodiment 2)
FIG. 2 shows the configuration of a speech decoding apparatus according to the second exemplary embodiment of the present invention.

  The code L representing the quantized LPC, the code S representing the noise code vector, the code P representing the adaptive code vector, and the code G representing the gain information transmitted from the encoder are respectively an LPC decoder 201 and The noise codebook 203, the adaptive codebook 204, and the gain codebook 205 are input.

  The LPC decoder 201 decodes the quantized LPC from the code L and outputs the decoded LPC to the mode selector 202 and the synthesis filter 209, respectively.

  The mode selector 202 determines the modes of the noise codebook 203 and the post-processor 211 using the quantized LPC input from the LPC decoder 201, and sends the mode information M to the noise codebook 203 and the post-processor 211, respectively. Output. Note that the mode selector 202 also stores information on quantized LPC input in the past, and selects a mode using both the characteristics of the quantized LPC variation between frames and the quantized LPC characteristics of the current frame. Do. There are at least two types of modes, and for example, a mode corresponding to a voiced voice part, a mode corresponding to an unvoiced voice part, and a mode corresponding to a stationary noise part are included. In addition, the information used for mode selection does not need to be the quantized LPC itself, and it is more effective to use information converted into parameters such as quantized LSP, reflection coefficient, and linear prediction residual power.

  The noise code book 203 stores a predetermined number of different noise code vectors, and outputs a noise code vector designated by a noise code book index obtained by decoding the input code S. The noise codebook 203 has at least two or more modes. For example, in the mode corresponding to the voiced voice part, a more pulsed noise code vector is generated, and the voice code part 203 is compatible with the voiceless voice part and the stationary noise part. In this mode, a more noisy noise code vector is generated. The noise code vector output from the noise codebook 203 is generated from one mode selected by the mode selector 202 from the two or more modes, and is added after the noise codebook gain Gs is multiplied by the multiplier 206. Is output to the device 208.

  The adaptive codebook 204 is buffered while sequentially updating the driving excitation signal generated in the past, and the adaptive codebook 204 uses an adaptive codebook index (pitch period (pitch lag)) obtained by decoding the input code P. Generate a vector. The adaptive code vector generated by adaptive codebook 204 is multiplied by adaptive codebook gain Ga by multiplier 207 and then output to adder 208.

  The gain codebook 205 stores a predetermined number of sets (gain vectors) of the adaptive codebook gain Ga and the noise codebook gain Gs, and uses a gain codebook index obtained by decoding the input code G. The adaptive codebook gain component Ga of the designated gain vector is output to the multiplier 207, and the noise codebook gain component Gs is output to the multiplier 206.

  Adder 208 adds the noise code vector and adaptive code vector input from multipliers 206 and 207 to generate a drive excitation signal, and outputs the generated signal to synthesis filter 209 and adaptive codebook 204.

  The synthesis filter 209 constructs an LPC synthesis filter using the input quantized LPC. The synthesized filter is subjected to filter processing with the driving sound source signal output from the adder 208 as an input, and the synthesized signal is output to the post filter 210.

  The post filter 210 performs processing for improving the subjective quality of the audio signal, such as pitch emphasis, formant emphasis, spectral tilt correction, and gain adjustment, on the synthesized signal input from the synthesis filter 209, and the post-processor 211. Output to.

  The post-processor 211 performs a process for improving the subjective quality of the stationary noise part, such as an inter-frame smoothing process of the amplitude spectrum and a randomization process of the phase spectrum, on the signal input from the post filter 210. This is adaptively performed using mode information M input from 202. For example, the smoothing process and the randomization process are hardly performed in the mode corresponding to the voiced voice part and the unvoiced voice part, and the smoothing process and the randomization process are adaptively performed in the mode corresponding to the stationary noise part. The post-processed signal is output as output data such as a digitized decoded speech signal.

  In the present embodiment, the mode information M output from the mode selector 202 is configured to be used for both mode switching of the noise codebook 203 and mode switching of the post-processor 211. Even if it is used for only mode switching, an effect can be obtained. In this case, only one of them is multimode processing.

  Next, referring to FIG. 4, the flow of processing of the speech decoding method in the above embodiment is shown. In this description, the speech encoding process is performed for each predetermined time length processing unit (frame: about several tens of milliseconds in time length), and one frame is further divided into a short number of processing units (sub An example in which processing is performed for each frame) will be described.

  In step 401, all the memories such as the contents of the adaptive codebook, the synthesis filter memory, and the output buffer are cleared.

  Next, in step 402, the encoded data is decoded. Specifically, the multiplexed received signal is separated or the bit stream is converted into a code that represents a quantized LPC coefficient, an adaptive code vector, a noise code vector, and gain information, respectively. .

  Next, in step 403, the LPC coefficients are decoded. The LPC coefficient is decoded from the code representing the quantized LPC coefficient obtained in step 402 by the reverse procedure of the LPC coefficient quantization method described in the first embodiment.

  Next, in step 404, a synthesis filter is constructed using the LPC coefficients decoded in step 403.

  Next, in step 405, using the static and dynamic features of the LPC coefficients decoded in step 403, a noise codebook and post-processing mode selection is performed. Specifically, a variation of the quantized LSP, a reflection coefficient calculated from the quantized LPC coefficient, a prediction residual power, and the like are used. The noise codebook is decoded and post-processed according to the mode selected in this step. There are at least two types of modes, and for example, a mode corresponding to the voiced voice part, a mode corresponding to the unvoiced voice part, and a mode corresponding to the stationary noise part or the like are included.

  Next, in step 406, the adaptive code vector is decoded. The adaptive code vector is decoded by decoding a position where the adaptive code vector is cut out from the adaptive codebook from a code representing the adaptive code vector, and cutting out the adaptive code vector from the position.

  Next, in step 407, the random code vector is decoded. The noise code vector is decoded by decoding the noise codebook index from the code representing the noise code vector and extracting the noise code vector corresponding to the index from the noise codebook. When applying the pitch periodization of the noise code vector, the decoded noise code vector is obtained after further pitch periodization and the like. This noise codebook has at least two types of modes. For example, in a mode corresponding to a voiced voice part, a more pulsating noise code vector is generated, and it corresponds to an unvoiced voice part or a stationary noise part. In the mode, a more noisy noise code vector is generated.

  Next, in step 408, the adaptive codebook gain and the noise codebook gain are decoded. The gain information is decoded by decoding the gain codebook index from the code representing the gain information and taking out the set of the adaptive codebook gain and the noise codebook gain indicated by this index from the gain codebook.

  Next, in step 409, a driving sound source signal is generated. The driving excitation signal is selected in step 408 by multiplying the adaptive code vector selected in step 406 by the adaptive codebook gain selected in step 408 and the noise code vector selected in step 407. It is generated by adding the vector multiplied by the noise codebook gain.

  Next, in step 410, the decoded signal is synthesized. The decoded signal is synthesized by filtering the driving sound source signal generated in step 409 with the synthesis filter constructed in step 404.

  Next, in step 411, post-filter processing is performed on the decoded signal. The post filter processing includes processing for improving the subjective quality of the decoded signal, particularly the decoded speech signal, such as pitch enhancement processing, formant enhancement processing, spectral tilt correction processing, and gain adjustment processing.

  Next, in step 412, final post-processing is performed on the decoded signal after post-filter processing. This post-processing mainly includes processing for improving the subjective quality of the stationary noise portion in the decoded signal, such as (sub) interframe smoothing processing of the amplitude spectrum and randomization processing of the phase spectrum. The processing corresponding to the mode selected in the above is performed. For example, in the mode corresponding to the voiced voice part and the unvoiced voice part, the smoothing process and the randomizing process are hardly performed, and in the mode corresponding to the stationary noise part and the like, the smoothing process and the randomizing process are adaptively performed. ing. The signal generated in this step is output data.

  Next, in step 413, the memory used in the subframe processing loop is updated. Specifically, the update of the adaptive codebook, the state update of each filter included in the post filter processing, and the like are performed.

  Steps 404 to 413 are processing in units of subframes.

  Next, in step 414, the memory used in the frame processing loop is updated. Specifically, quantization (decoding) LPC coefficient buffer update, output data buffer update, and the like are performed.

  Steps 402 to 403 and 414 are processes in units of frames. Further, the processing for each frame is repeated until there is no encoded data.

(Embodiment 3)
FIG. 5 is a block diagram showing a speech signal transmitter and receiver including the speech coding apparatus according to the first embodiment or the speech decoding apparatus according to the second embodiment. FIG. 5A shows a transmitter, and FIG. 5B shows a receiver.

  In the audio signal transmitter of FIG. 5A, audio is converted into an electrical analog signal by the audio input device 501 and output to the A / D converter 502. The analog speech signal is converted into a digital speech signal by the A / D converter 502 and output to the speech encoder 503. The voice encoder 503 performs a voice encoding process and outputs the encoded information to the RF modulator 504. The RF modulator performs an operation for transmitting the information of the encoded audio signal as a radio wave such as modulation / amplification / code spreading, and outputs it to the transmission antenna 505. Finally, a radio wave (RF signal) 506 is transmitted from the transmission antenna 505.

  On the other hand, in the receiver of FIG. 5B, a radio wave (RF signal) 506 is received by the receiving antenna 507, and the received signal is sent to the RF demodulator 508. The RF demodulator 508 performs processing for converting a radio wave signal such as code despreading / demodulation into encoded information, and outputs the encoded information to the speech decoder 509. The audio decoder 509 performs a decoding process on the encoded information and outputs a digital decoded audio signal to the D / A converter 510. The D / A converter 510 converts the digital decoded speech signal output from the speech decoder 509 into an analog decoded speech signal and outputs it to the speech output device 511. Finally, the audio output device 511 converts the electrical analog decoded audio signal into decoded audio and outputs it.

  The transmission device and the reception device can be used as a mobile device or a base station device of a mobile communication device such as a mobile phone. Note that the medium for transmitting information is not limited to the radio wave as shown in this embodiment mode, and an optical signal or the like can be used, and a wired transmission path can also be used.

  The speech encoding apparatus shown in the first embodiment, the speech decoding apparatus shown in the second embodiment, and the transmission apparatus and transmission / reception apparatus shown in the third embodiment are a magnetic disk, a magneto-optical disk, It can also be realized by recording as software on a recording medium such as a ROM cartridge, and by using the recording medium, a speech encoding apparatus / decoding apparatus and a personal computer using such a recording medium can be realized. A transmitter / receiver can be realized.

(Embodiment 4)
The fourth embodiment is an example showing a configuration example of the mode selectors 105 and 202 in the first and second embodiments described above.

  FIG. 6 shows the configuration of the mode selector according to the fourth embodiment.

  The mode selector according to the present embodiment includes a dynamic feature extraction unit 601 that extracts a dynamic feature of a quantized LSP parameter, and first and second static features that extract a static feature of the quantized LSP parameter. Extractors 602 and 603 are provided.

  The dynamic feature extraction unit 601 performs the smoothing process by inputting the quantized LSP parameter to the AR type smoothing unit 604. The AR-type smoothing unit 604 performs the smoothing process shown in the equation (1) using each next-order quantized LSP parameter input every processing unit time as time-series data.

Ls [i] = (1-α) × Ls [i] + α × L [i], i = 1, 2,..., M, 0 <α <1 (1)
Ls [i]: i-th order smoothed quantized LSP parameter
L [i]: i th order quantized LSP parameter
α: Smoothing coefficient
M: LSP analysis order In the equation (1), the value of α is set to about 0.7 so that the smoothing is not so strong. The smoothed quantized LSP parameter obtained by the above equation (1) is branched into one that is input to the adder 606 via the delay means 605 and one that is directly input to the adder 606.

  The delay means 605 delays the input smoothed quantized LSP parameter by one processing unit time and outputs it to the adder 606.

  The adder 606 receives the smoothed quantized LSP parameter in the current processing unit time and the smoothed quantized LSP parameter in the previous processing unit time. The adder 606 calculates the difference between the smoothed quantized LSP parameter in the current processing unit time and the smoothed quantized LSP parameter in the previous processing unit time. This difference is calculated for each order of the LSP parameter. The calculation result by the adder 606 is output to the square sum calculation means 607.

  The sum-of-squares calculation means 607 calculates the sum of squares of the difference for each order between the smoothed quantized LSP parameter in the current processing unit time and the smoothed quantized LSP parameter in the previous processing unit time. calculate.

  In the dynamic feature extraction unit 601, the quantized LSP parameter is also input to the delay unit 608 in parallel with the AR type smoothing unit 604. The delay unit 608 delays by one processing unit time and outputs the result to the AR type average value calculation unit 611 via the switch 609.

  The switch 609 closes when the mode information output from the delay unit 610 is a noise mode, and operates to input the quantized LSP parameter output from the delay unit 608 to the AR-type average value calculation unit 611. .

  The delay means 610 receives the mode information output from the mode determination means 621, delays it by one processing unit time, and outputs it to the switch 609.

  The AR-type average value calculation unit 611 calculates an average LSP parameter in the noise section based on the equation (1), similarly to the AR-type smoothing unit 604, and outputs the average LSP parameter to the adder 612. However, the value of α in equation (1) is set to about 0.05, and an extremely strong smoothing process is performed to calculate an average LSP parameter.

  The adder 612 calculates, for each order, the difference between the quantized LSP parameter in the current processing unit time and the average quantized LSP parameter in the noise interval calculated by the AR type average value calculating unit 611 for each order. It outputs to the sum calculation means 613.

  The sum of squares calculation means 613 receives the difference information of the quantized LSP parameters output from the adder 612, calculates the square sum of each order, and outputs it to the speech section detection means 619.

  The dynamic feature extraction unit 601 for the quantized LSP parameter is configured by the above elements 604 to 613.

  The first static feature extraction unit 602 calculates linear prediction residual power from the quantized LSP parameter in the linear prediction residual power calculation unit 614. Further, the adjacent LSP interval calculation means 615 calculates an interval for each adjacent order of the quantized LSP parameter as shown in the equation (2).

Ld [i] = L [i + 1] -L [i], i = 1, 2,... M-1 (2)
L [i]: i-th order quantized LSP parameter The calculated value of the adjacent LSP interval calculating means 615 is given to the variance value calculating means 616. The variance value calculation unit 616 calculates the variance value of the quantized LSP parameter interval output from the adjacent LSP interval calculation unit 615. When calculating the variance value, the characteristics of the peaks and valleys of the spectrum existing in the portion other than the lowest range are reflected by excluding the low end (Ld [1]) data without using all the LSP parameter interval data. be able to. When stationary noise with the characteristic that the low range is raised is passed through a high-pass filter, there is always a peak of the spectrum near the cutoff frequency of the filter, so the effect of removing such spectral peak information is effective. is there.

  The first static feature extraction unit 602 for the quantized LSP parameter is configured by the above elements 614, 615, and 616.

  Further, in the second static feature extraction unit 603, the reflection coefficient calculation unit 617 converts the quantized LSP parameter into a reflection coefficient and outputs it to the voiced / unvoiced determination unit 620. At the same time, the linear prediction residual power calculation unit 618 calculates the linear prediction residual power from the quantized LSP parameter and outputs the linear prediction residual power to the voiced / unvoiced determination unit 620.

  Note that the linear prediction residual power calculation means 618 is the same as the linear prediction residual power calculation means 614, so that 614 and 618 can be shared.

  The second static feature extraction unit 603 for quantized LSP parameters is configured by the elements 617 and 618 described above.

  The outputs of the dynamic feature extraction unit 601 and the first static feature extraction unit 602 are given to the speech section detection means 619. The speech section detecting means 619 receives the variation amount of the smoothed quantized LSP parameter from the square sum calculating means 607 and receives the average quantized LSP parameter of the noise section and the current quantized LSP parameter from the square sum calculating means 613. , The quantized linear prediction residual power is input from the linear prediction residual power calculation means 614, and the variance information of the adjacent LSP interval data is input from the variance value calculation means 616. Then, using these pieces of information, it is determined whether or not the input signal (or decoded signal) in the current processing unit time is a speech section, and the determination result is output to the mode determination means 621. A more specific method for determining whether or not the speech section is present will be described later with reference to FIG.

  On the other hand, the output of the second static feature extraction unit 603 is given to the voiced / unvoiced determination means 620. Voiced / unvoiced determination means 620 receives the reflection coefficient input from reflection coefficient calculation means 617 and the quantized linear prediction residual power input from linear prediction residual power calculation means 618, respectively. Then, using these pieces of information, it is determined whether the input signal (or decoded signal) in the current processing unit time is a voiced interval or an unvoiced interval, and the determination result is output to the mode determination means 621. A more specific sound / silence determination method will be described later with reference to FIG.

  The mode determination unit 621 receives the determination result output from the voice section detection unit 619 and the determination result output from the voiced / unvoiced determination unit 620, and inputs the current processing unit time using these pieces of information. The mode of the signal (or decoded signal) is determined and output. A more specific mode classification method will be described later with reference to FIG.

  In the present embodiment, the AR type is used for the smoothing means and the average value calculating means, but it is also possible to perform smoothing and average value calculation using other methods.

  Next, the details of the speech segment determination method in the above embodiment will be described with reference to FIG.

First, in step 801, a first dynamic parameter (Para1) is calculated. The specific content of the first dynamic parameter is a variation amount of the quantization LSP parameter for each processing unit time,
(3) It is shown by Formula.

次に、ステップ８０２において、第１の動的パラメータが予め定めてある閾値Ｔｈ１より大きいかどうかをチェックする。閾値Ｔｈ１を越えている場合は、量子化ＬＳＰパラメータの変動量が大きいので、音声区間であると判定する。一方、閾値Ｔｈ１以下の場合は、量子化ＬＳＰパラメータの変動量が小さいので、ステップ８０３に進み、さらに別のパラメータを用いた判定処理のステップに進んでゆく。

Next, in step 802, it is checked whether or not the first dynamic parameter is larger than a predetermined threshold value Th1. When the threshold value Th1 is exceeded, the quantized LSP parameter has a large amount of variation, and therefore, it is determined that the voice section is being used. On the other hand, when the threshold value is less than or equal to Th1, the quantized LSP parameter variation is small, so the process proceeds to step 803, and further proceeds to a determination process step using another parameter.

  In step 802, when the first dynamic parameter is equal to or smaller than the threshold Th1, the process proceeds to step 803, and the number of counters indicating how much the stationary noise period has been determined in the past is checked. The counter has an initial value of 0 and is incremented by 1 for each processing unit time determined to be a stationary noise section by the mode determination method. In step 803, if the number of counters is equal to or smaller than a preset threshold value ThC, the process proceeds to step 804, where it is determined whether or not it is a voice segment using a static parameter. On the other hand, if the threshold ThC is exceeded, the process proceeds to step 806, where it is determined whether or not it is a voice segment using the second dynamic parameter.

  In step 804, two types of parameters are calculated. One is the linear prediction residual power calculated from the quantized LSP parameter (Para3), and the other is the variance of the difference information of the adjacent order of the quantized LSP parameter (Para4). The linear prediction residual power can be obtained by converting the quantized LSP parameter into a linear prediction coefficient and using a relational expression in the Levinson-Durbin algorithm. Since it is known that the linear prediction residual power tends to be larger in the unvoiced part than in the voiced part, it can be used as a criterion for voiced / unvoiced. The difference information of the adjacent order of the quantized LSP parameter is shown in the equation (2), and the variance of these data is obtained. However, depending on the type of noise and how the band is limited, there is a peak (peak) of the spectrum in the low band, so the difference information of the adjacent order at the low band end (i = 1 in equation (2)) is not used. In addition, in equation (2), it is better to obtain the variance using data from i = 2 to M-1 (M is the analysis order). Since an audio signal has about three formants in the telephone band (200 Hz to 3.4 kHz), there are some portions where the LSP interval is narrow and wide, and there is a tendency that the dispersion of the interval data increases. . On the other hand, since stationary noise does not have a formant structure, the LSP interval is often relatively equal and the variance tends to be small. Using this property, it is possible to determine whether or not it is a voice section. However, as described above, depending on the type of noise or the like, there may be a peak (peak) of the spectrum in the low band. In such a case, the LSP interval on the lowest band side becomes narrow, so all adjacent LSP differences When the variance is obtained using data, the difference due to the presence or absence of the formant structure is reduced, and the determination accuracy is lowered. Accordingly, by obtaining the variance by excluding the adjacent LSP difference information at the low band end, such accuracy deterioration is avoided. However, since such a static parameter has a lower determination ability than a dynamic parameter, it is preferably used as auxiliary information. The two types of parameters calculated in step 804 are used in step 805.

  Next, in step 805, threshold processing using the two types of parameters calculated in step 804 is performed. Specifically, when the linear prediction residual power (Para3) is smaller than the threshold value Th3 and the variance (Para4) of adjacent LSP interval data is larger than the threshold value Th4, it is determined as a speech section. In other cases, it is determined as a stationary noise section (non-voice section). If it is determined that it is a stationary noise section, the counter value is increased by one.

  In step 806, the second dynamic parameter (Para2) is calculated. The second dynamic parameter is a parameter indicating the degree of similarity between the average quantized LSP parameter in the past stationary noise interval and the quantized LSP parameter in the current processing unit time. As shown, a difference value is obtained for each order using the two types of quantized LSP parameters, and a sum of squares is obtained. The obtained second dynamic parameter is used for threshold processing in step 807.

次に、ステップ８０７において、第２の動的パラメータが閾値Th2を越えているかどうかの判定が行われる。閾値Th2を越えていれば、過去の定常雑音区間における平均的な量子化ＬＳＰパラメータとの類似度が低いので、音声区間と判定し、閾値Th2以下であれば、過去の定常雑音区間における平均的な量子化ＬＳＰパラメータとの類似度が高いので、定常雑音区間と判定する。定常雑音区間と判定された場合は、カウンターの値を１増やす。

Next, in step 807, it is determined whether or not the second dynamic parameter exceeds the threshold value Th2. If the threshold Th2 is exceeded, the similarity to the average quantized LSP parameter in the past stationary noise section is low, so it is determined as a speech section. If the threshold Th2 or less, the average in the past stationary noise section is averaged. Since it is highly similar to a quantized LSP parameter, it is determined as a stationary noise section. If it is determined that it is a stationary noise section, the counter value is increased by one.

  Next, the details of the voiced / unvoiced section determination method in the above embodiment will be described with reference to FIG.

  First, in step 901, a primary reflection coefficient is calculated from the quantized LSP parameter in the current processing unit time. The reflection coefficient is calculated by converting the LSP parameter into a linear prediction coefficient.

  Next, in step 902, it is determined whether the reflection coefficient exceeds a first threshold Th1. If the threshold value Th1 is exceeded, it is determined that the current processing unit time is an unvoiced section and the voiced / unvoiced determination process is terminated. If the threshold value Th1 or less, the voiced / unvoiced determination process is continued.

  If it is not determined in step 902 that there is no voice, it is determined in step 903 whether the reflection coefficient exceeds the second threshold Th2. If it exceeds the threshold Th2, the process proceeds to Step 905, and if it is equal to or less than the threshold Th2, the process proceeds to Step 904.

  If it is determined in step 903 that the reflection coefficient is equal to or smaller than the second threshold value Th2, it is determined in step 904 whether the reflection coefficient exceeds the third threshold value Th3. If it exceeds the threshold value Th3, the process proceeds to step 907. If it is equal to or less than the threshold value Th3, it is determined as a voiced section, and the voiced / unvoiced determination process is terminated.

  If the reflection coefficient exceeds the second threshold value Th2 in step 903, linear prediction residual power is calculated in step 905. The linear prediction residual power is calculated after converting the quantized LSP into a linear prediction coefficient.

  Following step 905, in step 906, a determination is made whether the linear prediction residual power exceeds a threshold Th4. If the threshold value Th4 is exceeded, it is determined as an unvoiced section and the voiced / unvoiced determination process is terminated, and if it is equal to or less than the threshold value Th4, it is determined as a voiced section and the voiced / unvoiced determination process is terminated.

  If the reflection coefficient exceeds the third threshold Th3 in step 904, linear prediction residual power is calculated in step 907.

  Following step 907, in step 908, a determination is made whether the linear prediction residual power exceeds a threshold Th5. If it exceeds the threshold value Th5, it is determined as a voiced section and the voiced / unvoiced determination process is terminated. If it is equal to or less than the threshold value Th5, it is determined as a voiced section and the voiced / unvoiced determination process is terminated.

  Next, with reference to FIG. 10, the mode determination method used for the mode determination means 621 will be described.

  First, in step 1001, a speech segment detection result is input. This step may be the block itself that performs the voice section detection process.

  Next, in step 1002, it is determined whether or not to determine the stationary noise mode based on the determination result of whether or not it is a speech section. If it is a speech section, the process proceeds to step 1003. If it is not a speech section (a stationary noise section), a mode determination result indicating that it is a stationary noise mode is output, and the mode determination process is terminated.

  If it is determined in step 1002 that the mode is not the stationary noise section mode, then in step 1003, a voiced / unvoiced determination result is input. This step may be a block itself that performs voiced / unvoiced determination processing.

  Subsequent to step 1003, in step 1004, based on the voiced / unvoiced determination result, it is determined whether the mode is the voiced section mode or the unvoiced section mode. If it is a voiced section, a mode determination result indicating that it is a voiced section mode is output and the mode determination process is terminated, and if it is a voiceless section, a mode determination result indicating that it is a voiceless section mode is output and a mode is output. The determination process ends. As described above, the mode of the input signal (or decoded signal) in the current processing unit block is classified into three modes using the speech segment detection result and the voiced / unvoiced determination result.

(Embodiment 5)
FIG. 7 shows the configuration of the post-processor according to the fifth embodiment of the present invention. This post-processor is used in the speech signal decoding apparatus shown in the second embodiment in combination with the mode determiner shown in the fourth embodiment. The post-processor shown in the figure includes mode changeover switches 705, 708, 707, and 711, amplitude spectrum smoothing means 706, phase spectrum randomizing means 709 and 710, and threshold setting means 703 and 716, respectively.

  The weighting synthesis filter 701 inputs the decoded LPC output from the LPC decoder 201 of the speech decoding apparatus to construct a perceptual weighting synthesis filter, and outputs it from the synthesis filter 209 or the post filter 210 of the speech decoding apparatus. A weighting filter process is performed on the synthesized speech signal and output to the FFT processing means 702.

  The FFT processing unit 702 performs FFT processing on the decoded signal after the weighting process output from the weighting synthesis filter 701, and converts the amplitude spectrum WSAi into the first threshold value setting unit 703, the first amplitude spectrum smoothing unit 706, and the first. To the phase spectrum randomizing means 709.

  The first threshold value setting means 703 calculates the average value of the amplitude spectrum calculated by the FFT processing means 702 using all frequency components, and uses the average value as a reference to set the threshold value Th1 as the first amplitude spectrum smoothing. It outputs to the means 706 and the 1st phase spectrum randomization means 709, respectively.

  The FFT processing means 704 performs FFT processing on the synthesized speech signal output from the synthesis filter 209 or the post filter 210 of the speech decoding apparatus, and converts the amplitude spectrum into mode changeover switches 705 and 712, an adder 715, and a second phase. The phase spectrum is output to the spectrum randomizing means 710 to the mode switch 708, respectively.

  The mode changeover switch 705 receives the mode information (Mode) output from the mode selector 202 of the speech decoding apparatus and the difference information (Diff) output from the adder 715 and inputs the current processing unit. It is determined whether the decoded signal in time is a speech interval or a stationary noise interval. If it is determined to be a speech interval, it is connected to the mode changeover switch 707. If it is determined to be a stationary noise interval, the first amplitude spectrum smoothing means. Connect to 706.

  The first amplitude spectrum smoothing means 706 inputs the amplitude spectrum SAi from the FFT processing means 704 via the mode changeover switch 705, and the frequency determined by the separately input first threshold Th1 and weighted amplitude spectrum WSAi. The component is smoothed and output to the mode changeover switch 707. The method of determining the frequency component to be smoothed is determined depending on whether the weighted amplitude spectrum WSAi is equal to or less than the first threshold value Th1. That is, the smoothing process is performed only for the frequency component i whose WSAi is equal to or less than Th1. By this smoothing process, the temporal discontinuity of the amplitude spectrum due to coding distortion in the stationary noise section is alleviated. For example, the coefficient α when the smoothing process is performed in the AR type as shown in the equation (1) can be set to about 0.1 when the number of FFT points is 128 points and the processing unit time is 10 ms.

  The mode changeover switch 707 is similar to the mode changeover switch 705, and mode information (Mode) output from the mode selector 202 of the speech decoding apparatus, difference information (Diff) output from the adder 715, To determine whether the decoded signal in the current processing unit time is a speech section or a stationary noise section. If it is determined to be a speech section, it is connected to the mode changeover switch 705 and if it is determined to be a stationary noise section. , Connected to the first amplitude spectrum smoothing means 706. The determination result is the same as the determination result of the mode switch 705. The other end of the mode switch 707 is connected to the IFFT processing means 720.

  The mode changeover switch 708 is a switch that is switched in conjunction with the mode changeover switch 705, and mode information (Mode) output from the mode selector 202 of the speech decoding apparatus and difference information (from the adder 715) ( Diff) is input to determine whether the decoded signal in the current processing unit time is a speech section or a stationary noise section. If it is determined to be a speech section, it is connected to the second phase spectrum randomizing means 710. If the stationary noise section is determined, the first phase spectrum randomizing means 709 is connected. The determination result is the same as the determination result of the mode switch 705. That is, when the mode changeover switch 705 is connected to the first amplitude spectrum smoothing means 706, the mode changeover switch 708 is connected to the first phase spectrum randomizing means 709, and the mode changeover switch 705 is switched to the mode. When connected to the changeover switch 707, the mode changeover switch 708 is connected to the second phase spectrum randomizing means 710.

  The first phase randomizing means 709 receives the phase spectrum SPi output from the FFT processing means 704 via the mode changeover switch 708, and is determined by the first threshold Th1 and the weighted amplitude spectrum WSAi separately input. The frequency component is randomized and output to the mode switch 711. The method of determining the frequency component to be randomized is the same as the method of determining the frequency component to be smoothed by the smoothing means 706 of the first amplitude spectrum. That is, the randomization process of the phase spectrum SPi is performed only for the frequency component i whose WSAi is equal to or less than Th1.

  The second phase spectrum randomizing means 710 receives the phase spectrum SPi output from the FFT processing means 704 via the mode changeover switch 708 and is determined by the separately input second threshold Th2i and amplitude spectrum SAi. The frequency component is randomized and output to the mode switch 711. The method for determining the frequency component to be randomized is the same as that of the first phase spectrum randomizing means 709. That is, the randomization process of the phase spectrum SPi is performed only for the frequency component i whose SAi is equal to or less than Th2i.

  The mode changeover switch 711 is linked to the mode changeover switch 707, and in the same manner as the mode changeover switch 707, mode information (Mode) output from the mode selector 202 of the speech decoding apparatus and the adder 715. The difference information (Diff) to be output is input, and it is determined whether the decoded signal in the current processing unit time is a speech section or a stationary noise section. If the decoded signal is determined to be a speech section, the second phase spectrum random If it is determined that it is a stationary noise section, the first phase spectrum randomizing unit 709 is connected. The determination result is the same as the determination result of the mode switch 708. The other end of the mode switch 711 is connected to the IFFT processing means 720.

  In the same manner as the mode changeover switch 705, the mode changeover switch 712 includes mode information (Mode) output from the mode selector 202 of the speech decoding apparatus, difference information (Diff) output from the adder 715, To determine whether the decoded signal in the current processing unit time is a speech section or a stationary noise section. If it is determined that the decoded signal is not a speech section (a stationary noise section), a second switch is connected to The amplitude spectrum SAi output from the FFT processing means 704 is output to the amplitude spectrum smoothing means 713. When it is determined that the voice section is selected, the mode switch 712 is opened and the amplitude spectrum SAi is not output to the second amplitude spectrum smoothing means 713.

  The second amplitude spectrum smoothing means 713 receives the amplitude spectrum SAi output from the FFT processing means 704 via the mode switch 712, and performs the smoothing process on all frequency band components. By this smoothing process, an average amplitude spectrum in the stationary noise section is obtained. This smoothing process is the same as the process performed by the first amplitude spectrum smoothing means 706. When the mode changeover switch 712 is opened, no processing is performed in this means, and the smoothed amplitude spectrum SSAi in the stationary noise section when the processing is performed last is output. The amplitude spectrum SSAi smoothed by the second amplitude spectrum smoothing processing unit 713 is output to the delay unit 714, the second threshold setting unit 716, and the mode switch 718, respectively.

  The delay means 714 receives the SSAi output from the second amplitude spectrum smoothing means 713, delays it by one processing unit time, and outputs it to the adder 715.

  The adder 715 calculates a distance Diff between the stationary noise section smoothed amplitude spectrum SSAi one processing unit time before and the amplitude spectrum SAi in the current processing unit time, and mode selector switches 705, 707, 708, 711, 712, 718 and 719, respectively.

  The second threshold value setting means 716 sets the threshold value Th2i with reference to the stationary noise section smoothed amplitude spectrum SSAi output from the second amplitude spectrum smoothing means 713, and the second phase spectrum randomizing means 710. Output to.

  The random phase spectrum generation means 717 outputs the randomly generated phase spectrum to the mode switch 719.

  In the same manner as the mode changeover switch 712, the mode changeover switch 718 includes mode information (Mode) output from the mode selector 202 of the speech decoding apparatus, difference information (Diff) output from the adder 715, To determine whether the decoded signal in the current processing unit time is a speech section or a stationary noise section. If it is determined that the decoded signal is a speech section, a second amplitude spectrum smoothing means is connected by connecting a switch. The output of 713 is output to the IFFT processing means 720. If it is determined that it is not a speech section (a stationary noise section), the mode switch 718 is opened, and the output of the second amplitude spectrum smoothing means 713 is not output to the IFFT processing means 720.

  The mode changeover switch 719 is switched in conjunction with the mode changeover switch 718. Similarly to the mode changeover switch 718, mode information (Mode) output from the mode selector 202 of the speech decoding apparatus and the adder 715 are used. The difference information (Diff) to be output is input, and it is determined whether the decoded signal in the current processing unit time is a speech section or a stationary noise section. If it is determined that it is a speech section, connect a switch. Then, the output of the random phase generation means 717 is output to the IFFT processing means 720. If it is determined that it is not a speech section (a steady noise section), the mode switch 719 is opened, and the output of the random phase generation means 717 is not output to the IFFT processing means 720.

  The IFFT processing means 720 outputs the amplitude spectrum output from the mode changeover switch 707, the phase spectrum output from the mode changeover switch 711, the amplitude spectrum output from the mode changeover switch 718, and the mode changeover switch 719. Each of the phase spectra is input, inverse FFT processing is performed, and a post-processed signal is output. When the mode changeover switches 718 and 719 are opened, the amplitude spectrum input from the mode changeover switch 707 and the phase spectrum input from the mode changeover switch 711 are converted into an FFT real part spectrum and an imaginary part spectrum. Convert, perform inverse FFT processing, and output the real part of the result as a time signal. On the other hand, when the mode changeover switches 718 and 717 are connected, the amplitude spectrum input from the mode changeover switch 707 and the phase spectrum input from the mode changeover switch 711 are converted into the first real part spectrum and the first spectrum. In addition to the one converted into the imaginary part spectrum, the amplitude spectrum input from the mode changeover switch 718 and the phase spectrum input from the mode changeover switch 719 are converted into the second real part spectrum and the second imaginary part spectrum. Then, the inverse FFT process is performed by adding the converted data. That is, the third real part spectrum is obtained by adding the first real part spectrum and the second real part spectrum, and the third real part spectrum is obtained by adding the first imaginary part spectrum and the second imaginary part spectrum. , The inverse FFT processing is performed using the third real part spectrum and the third imaginary part spectrum. At the time of the addition of the spectrum, the second real part spectrum and the second imaginary part spectrum are attenuated by a variable that is controlled by a constant multiple or adaptively. For example, in the addition of the spectrum, the second real part spectrum is multiplied by 0.25 and then added to the first real part spectrum, and the second imaginary part spectrum is multiplied by 0.25 and then the first imaginary part spectrum is added. Are added to obtain a third real part spectrum and a third imaginary part spectrum, respectively.

  Next, the post-processing method will be described with reference to FIGS. FIG. 11 is a flowchart showing specific processing of the post-processing method in the present embodiment.

  First, in step 1101, the FFT logarithmic amplitude spectrum (WSAi) of the input signal (decoded speech signal) subjected to auditory weighting is calculated.

  Next, in step 1102, a first threshold Th1 is calculated. Th1 is obtained by adding a constant k1 to the average value of WSAi. The value of k1 is determined empirically and is, for example, about 0.4 in the common logarithm region. Assuming that the number of FFT points is N and the FFT amplitude spectrum is WSAi (i = 1, 2,... N), WSAi is symmetric about i = N / 2 and i = N / 2 + 1. If the average value of WSAi of a book is calculated, the average value of WSAi can be obtained.

  Next, in step 1103, the FFT logarithmic amplitude spectrum (SAi) and the FFT phase spectrum (SPi) of the input signal (decoded speech signal) without auditory weighting are calculated.

  Next, in step 1104, the spectral variation (Diff) is calculated. The spectrum fluctuation is the sum of the residual spectra obtained by subtracting the average FFT logarithmic amplitude spectrum (SSAi) from the current FFT logarithmic amplitude spectrum (SAi) in the section determined as the stationary noise section in the past. The spectral fluctuation (Diff) obtained in this step is a parameter for determining whether the current power is not large compared to the average power in the stationary noise section. It is a section where a signal different from the component exists and can be determined not to be a stationary noise section.

  Next, in step 1105, a counter indicating the number of times determined to be a stationary noise section in the past is checked. If the number of counters is greater than or equal to a certain value, that is, if it has been determined that the steady noise interval is stable to some extent in the past, the process proceeds to step 1107. Otherwise, it is determined that the counter is a stationary noise interval in the past. If there is not much, go to Step 1106. The difference between step 1106 and step 1107 is whether or not spectrum variation (Diff) is used as a criterion. The spectrum variation (Diff) is calculated by using an average FFT logarithmic amplitude spectrum (SSAi) in a section determined as a stationary noise section in the past. In order to obtain such an average FFT logarithmic amplitude spectrum (SSAi), since a steady noise section having a sufficiently long time length is required in the past, step 1105 is provided, and a steady time period having a sufficient time length in the past is provided. If there is no static noise section, it is considered that the average FFT logarithmic amplitude spectrum (SSAi) of the noise section is not sufficiently averaged, and therefore, the process proceeds to step 1106 without using the spectrum variation (Diff). The initial value of the counter is zero.

  Next, in step 1106 or step 1107, it is determined whether or not it is a stationary noise section. In step 1106, a case where the sound source mode already determined in the speech decoding apparatus is the stationary noise interval mode is determined as a stationary noise interval, and in step 1107, the excitation mode already determined in the speech decoding apparatus is the stationary noise interval. When the mode and the amplitude spectrum fluctuation (Diff) calculated in step 1104 is less than or equal to the threshold k3, it is determined as a stationary noise interval. If it is determined in step 1106 or step 1107 that it is a stationary noise section, the process proceeds to step 1108, and if it is determined that it is not a stationary noise section, that is, a voice section, the process proceeds to step 1113.

  If it is determined that it is a stationary noise interval, then in step 1108, a smoothing process for obtaining an average FFT logarithmic spectrum (SSAi) in the stationary noise interval is performed. In the expression of step 1108, β is a constant indicating the strength of smoothing in the range of 0.0 to 1.0. In the case where the number of FFT points is 128 points and the processing unit time is 10 ms (8 points at 8 kHz sampling), β is about 0.1. good. This smoothing process is performed for all logarithmic amplitude spectra (SAi, i = 1,... N, N is the number of FFT points).

  Next, in step 1109, the FFT logarithmic amplitude spectrum is smoothed to smooth the fluctuation of the amplitude spectrum in the stationary noise section. This smoothing process is the same as the smoothing process in step 1108, but is not performed for all logarithmic amplitude spectra (SAi), but only for the frequency component i whose auditory weighted logarithmic amplitude spectrum (WSAi) is smaller than the threshold Th1. . Γ in the expression of Step 1109 is the same as β in Step 1108, and may be the same value. In step 1109, a partially smoothed logarithmic amplitude spectrum SSA2i is obtained.

  Next, in step 1110, the FFT phase spectrum is randomized. This randomization process is performed in a frequency-selective manner, similarly to the smoothing process in step 1109. That is, similar to step 1109, the processing is performed only for the frequency component i whose auditory weighting logarithmic amplitude spectrum (WSAi) is smaller than the threshold Th1. Here, Th1 may be the same value as in step 1109, but may be set to a different value adjusted so as to obtain better subjective quality. Also, random (i) in step 1110 is a numerical value in the range of −2π to + 2π generated randomly. For random (i) generation, a new random number may be generated each time, but to save the calculation amount, the previously generated random number is stored in a table, and the contents of the table are changed for each processing unit time. It is also possible to go around and use it. In this case, there are a case where the contents of the table are used as they are and a case where the contents of the table are added to the original FFT phase spectrum.

  Next, in step 1111, a complex FFT spectrum is generated from the FFT logarithmic amplitude spectrum and the FFT phase spectrum. The real part is obtained by returning the FFT logarithmic amplitude spectrum SSA2i from the logarithmic domain to the linear domain and then multiplying by the cosine of the phase spectrum RSP2i. The imaginary part is obtained by returning the FFT logarithmic amplitude spectrum SSA2i from the logarithmic domain to the linear domain and then multiplying by the sine of the phase spectrum RSP2i.

  Next, in step 1112, the counter of the section determined as the stationary noise section is incremented by one.

  On the other hand, if it is determined in step 1106 or 1107 that the speech section is not a stationary noise section, then in step 1113, the FFT logarithmic amplitude spectrum SAi is copied to the smoothed logarithmic spectrum SSA2i. That is, the logarithmic amplitude spectrum is not smoothed.

  Next, in step 1114, the FFT phase spectrum is randomized. This randomization process is performed in a frequency selective manner in the same manner as in step 1110. However, the threshold used for frequency selection is not Th1, but a value obtained by adding a constant k4 to SSAi previously obtained in step 1108 is used. This threshold corresponds to the second threshold Th2i in FIG. That is, the phase spectrum is randomized only for the frequency component having an amplitude spectrum smaller than the average amplitude spectrum in the stationary noise section.

  Next, in step 1115, a complex FFT spectrum is generated from the FFT logarithmic amplitude spectrum and the FFT phase spectrum. The real part returns the FFT logarithmic amplitude spectrum SSA2i from the logarithmic domain to the linear domain, then multiplies the cosine of the phase spectrum RSP2i, and after the FFT logarithmic amplitude spectrum SSAi returns from the logarithmic domain to the linear domain, the phase spectrum random2 ( It is obtained by adding the product of the cosine of i) and the product of the constant k5. The imaginary part is obtained by returning the FFT logarithmic amplitude spectrum SSA2i from the logarithmic domain to the linear domain and then multiplying by the sine of the phase spectrum RSP2i, and after returning the FFT logarithmic amplitude spectrum SSAi from the logarithmic domain to the linear domain, the phase spectrum random2 ( This is obtained by adding the product of the sine of i) and the product of the constant k5. The constant k5 is set in the range of 0.0 to 1.0, more specifically, about 0.25. Note that k5 may be an adaptively controlled variable. By superimposing the average stationary noise multiplied by k5, the subjective quality of the background stationary noise in the speech section can be improved. random2 (i) is a random number similar to random (i).

  Next, in step 1116, the inverse FFT of the complex FFT spectrum (Re (S2) i, Im (S2) i) generated in step 1111 or 1115 is performed, and complex numbers (Re (s2) i, Im (s2)) are obtained. i) get.

  Finally, in step 1117, the complex real part Re (s2) i obtained by inverse FFT is output as an output signal.

DESCRIPTION OF SYMBOLS 103 LPC quantizer 104 Synthesis filter 105 Mode selector 109 Noise codebook 110 Adaptive codebook 111 Gain codebook 201 LPC decoder 202 Mode selector 209 Synthesis filter 210 Post filter 501 Speech input device 503 Speech encoder 509 Speech decoding 511 Voice output device 601 Dynamic feature extraction unit 602 Static feature extraction unit 604 AR type smoothing unit 609 Switch 611 AR type average value calculation unit 614 Linear prediction residual power calculation unit 615 Adjacent LSP interval calculation unit 616 Variance value Calculation means 617 Reflection coefficient calculation means 618 Linear prediction residual power calculation means 619 Voice segment detection means 620 Voiced / unvoiced determination means 621 Mode determination means 702 FFT processing means 703 First threshold setting means 705 Mode changeover switch 706 First 1 amplitude spectrum smoothing means 707, 708 mode changeover switch 709 first phase spectrum randomization means 710 second phase spectrum randomization means 711,712 mode changeover switch 713 second amplitude spectrum smoothing means 716 second Threshold setting means 717 Random phase spectrum generation means 718, 719 Mode changeover switch 720 Inverse FFT processing means

Claims (2)

Means for smoothing the quantized LSP parameters;
The inter-frame variation of the smoothed quantized LSP parameter is calculated by the sum of squares of the difference in each order between the smoothed quantized LSP parameter in the current processing unit time and the smoothed quantized LSP parameter in the previous processing unit time. Means for calculating;
Means for calculating an average quantized LSP parameter in a frame in which the quantized LSP parameter is stationary;
Means for calculating a distance between the average quantized LSP parameter and the current quantized LSP parameter based on a sum of squares of differences of the average quantized LSP parameter and the current quantized LSP parameter for each order. ,
Quantized LSP parameter dynamic feature extractor. Performs smoothing processing of quantized LSP parameters,
The inter-frame variation of the smoothed quantized LSP parameter is calculated by the sum of squares of the difference in each order between the smoothed quantized LSP parameter in the current processing unit time and the smoothed quantized LSP parameter in the previous processing unit time. Calculate
Calculating an average quantized LSP parameter in a frame in which the quantized LSP parameter is stationary;
A distance between the average quantized LSP parameter and the current quantized LSP parameter is calculated by a sum of squares of the difference in each order between the average quantized LSP parameter and the current quantized LSP parameter;
Quantized LSP parameter dynamic feature extraction method.

Images