JP2010044412A - Wide band voice encoding method, and wide band voice encoding device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a wide band voice encoding device capable of attaining excellent sound quality even when a narrow band voice signal is encoded. <P>SOLUTION: A band detecting section 11 detects that the narrow band voice signal is input. When the narrow band voice signal is input, a voice encoding section 14 performs encoding suitable for the narrow band voice signal, while it is a device for encoding the wide band voice, and thereby, the narrow band voice signal is encoded in excellent sound quality. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a wideband speech coding method and a wideband speech coding apparatus, and relates to a wideband speech coding process that encodes a narrowband speech signal with high quality.

  Also in the transmission of sound using a mobile communication system, it is required to transmit a wideband sound signal with good sound quality. 722.2 describes a wideband speech coding scheme. This ITU-T Recommendation G. The wideband speech coding method described in 722.2 is called an AMR-WB (Adaptive Multi Rate Wide Band) method, which is intended to encode a wideband speech signal of 16 kHz sampling with high quality. The rate is available. In general, in audio coding, the higher the bit rate, the better the sound quality, and the lower the bit rate, the lower the sound quality.

  This ITU-T Recommendation G. Since the wideband speech coding method described in 722.2 assumes the coding of a wideband speech signal having a bandwidth of about 50 Hz to 7 kHz, the sampling rate of the input signal is set to 16 kHz. However, when a voice signal having a sampling rate of 8 kHz that does not have a frequency of 4 kHz or higher, such as normal telephone voice, is input, it is necessary to first upsample to 16 kHz in order to encode the narrowband voice signal. is there.

  The audio signal up-sampled from 8 kHz to 16 kHz in this way is treated as the same as a normal wideband audio signal, although it is a narrowband audio signal having no frequency of 4 kHz or more, and is specialized for a wideband audio signal. It is encoded by the encoding method. Therefore, since an audio signal that is upsampled from 8 kHz to 16 kHz that does not have a frequency of 4 kHz or more is encoded specifically for a normal wideband audio signal, the audio encoding method and the bandwidth of the input signal Due to the mismatch, there is a problem that the encoding efficiency is lowered and the encoding with good sound quality cannot be performed.

  For this reason, when a wideband audio codec is used for a narrowband audio signal that is band-limited through a narrowband communication path or a narrowband codec, the medium to low level is about 6 to 10 kbit / s. At bit rates, the sound quality is much worse than when using a narrowband audio codec. The same problem occurs when an audio signal having a very low frequency of 4 kHz or more is input.

  In addition, as a method for improving sound quality by speech coding, a plurality of sets of pulse positions are held, and a distortion between the speech signal is calculated for each set of pulse positions by the sound source quantization unit. A process for selecting a set of pulse positions for reducing the distortion is known (see, for example, Patent Document 1).

JP 2001-318698 A (page 2-4, FIG. 1)

  However, in the conventional method described above, since it is necessary to calculate the distortion for each set of pulse positions held, the problem is that the amount of calculation required to select the set of pulse positions is enormous. There is a point. In addition, the conventional method does not take into consideration the problem of mismatch between the speech coding method and the bandwidth of the input signal, and the sound quality is extremely poor when the narrowband speech signal described above is subjected to wideband speech coding. The problem of becoming is not solved.

  The present invention has been made to solve the above-described problems, and it is an object of the present invention to provide a wideband speech coding method and a wideband speech coding apparatus capable of obtaining sound quality that may encode a narrowband speech signal. And

  In order to achieve the above object, a wideband speech coding method of the present invention is provided in a wideband speech coding method in which an input speech signal is represented by a spectrum parameter and a sound source signal, and each of the spectrum parameter and the sound source signal is encoded. A plurality of pulses are selected from the pulse position candidates, and the sound source signal is encoded using the selected pulses, and whether the input audio signal is a wideband audio signal or a narrowband audio signal is determined. An identification process for identifying and, if the input speech signal is identified as a wideband speech signal by the identification process, the encoding to select a pulse position candidate having a predetermined resolution for the wideband speech signal If the input speech signal is identified as a narrowband speech signal by the identification process, the pulse position A control process for controlling the encoding process to lower the complementary resolution, a conversion process for converting a sampling rate of the input audio signal into a sampling rate for a wideband audio signal, and a code generated by the encoding process And an output process for outputting the conversion result.

  Also, the wideband speech coding apparatus of the present invention represents an input speech signal by a spectral parameter and a sound source signal, and in the wideband speech coding device that encodes each of the spectral parameter and the sound source signal, a plurality of pulse position candidates are provided. Encoding means for encoding the sound source signal using the selected pulse, identification means for identifying whether the input audio signal is a wideband audio signal or a narrowband audio signal, When the identification unit identifies the input speech signal as a wideband speech signal, the encoding unit is controlled to select a pulse position candidate having a predetermined resolution for the wideband speech signal, and the identification unit When the input speech signal is identified as a narrow-band speech signal, the encoding means is controlled to lower the resolution of the pulse position candidate. And control means that is characterized by having conversion means for converting the sampling rate of the input audio signal to the sampling rate for wideband speech signals, and output means for outputting the encoded result of the encoding means.

  As described above, according to the present invention, by detecting that a narrowband speech signal is input and adapting wideband speech coding to the narrowband speech signal, the narrowband speech signal is also improved in sound quality. Can be encoded.

1 is a block diagram showing a wideband speech encoding apparatus according to an embodiment of the present invention. The block diagram which shows the audio | voice encoding part which concerns on the 1st Embodiment of this invention. The block diagram (1st example) which shows the pulse position candidate setting part and pulse position candidate which concern on the 1st Embodiment of this invention. The pulse position candidate of the integer sample position which concerns on embodiment of this invention. The pulse position candidate of the even sample position which concerns on the 1st Embodiment of this invention. The block diagram (2nd example) which shows the pulse position candidate setting part and pulse position candidate which concern on the 1st Embodiment of this invention. The pulse position candidate of the odd sample position which concerns on the 1st Embodiment of this invention. The flowchart which shows the control operation of the control part which concerns on embodiment of this invention. The block diagram which shows the audio | voice encoding part which concerns on the 2nd Embodiment of this invention. The block diagram which shows the audio | voice encoding part which concerns on embodiment of this invention.

  Hereinafter, embodiments of a wideband speech coding method and a wideband speech coding apparatus according to the present invention will be described with reference to the drawings.

(First embodiment)
FIG. 1 is a block diagram showing a wideband speech encoding apparatus according to an embodiment of the present invention. This device includes a band detection unit 11, a sampling rate conversion unit 12, a speech encoding unit 14, and a control unit 15 that controls the entire device, receives an input speech signal 10, and receives the input speech signal. This is an apparatus that outputs an output code 19 obtained by encoding 10.

  The band detector 11 detects the sampling rate of the input audio signal 10 and notifies the controller 15 of the detected sampling rate.

  At this time, the band detector 11 (1) detects and inputs the sampling rate information of the input audio signal 10 from the outside, (2) acquires attribute information (such as file header information) of the input audio signal 10 (3) Obtaining the identification information of the codec that generated the input audio signal 10 and detecting the sampling rate of the input audio signal depending on whether it is a narrowband codec or a wideband codec. Although rate detection is performed, the present invention is not limited to these. For example, FIG. 10 shows a band detection unit 11 a configured to acquire sampling rate information and information for identifying a wideband signal / narrowband signal from the input audio signal 10. This is because the sampling rate information and the information for identifying the wide band / narrow band, the attribute information of the input audio signal, or the identification of the codec that generated the input audio signal 10 is included in the bits of the predetermined part of the input audio signal sequence. This is the configuration when information or the like is embedded. As an embedding method, a method of embedding in the least significant bit of the PCM of the input audio signal sequence can be considered. By doing this, the sampling rate information, the information for identifying the wideband / narrowband, or the information of the input audio signal without affecting the upper bits of the PCM (that is, without affecting the sound quality of the input audio signal) It is possible to embed attribute information or identification information of the codec that generated the input audio signal 10.

  As described above, various embodiments are conceivable as the band detection unit, but any configuration may be used as long as it can identify the sampling rate information, wideband / narrowband, or codec. Needless to say. Also, the sampling rate information, wideband / narrowband identification information, and codec identification information may be information representative of them.

  The sampling rate conversion unit 12 converts the input speech signal 10 into a predetermined sampling rate, and transmits the obtained signal of the predetermined sampling rate to the speech encoding unit 14.

  The sampling rate converter 12 receives the 8 kHz sampling signal as an input, and obtains and outputs an upsampled 16 kHz sampling signal using an interpolation filter. The sampling rate conversion unit 12 receives a 16 kHz sampling signal and outputs it without converting the sampling rate, but is not limited to this.

  Further, the method of converting the sampling rate is not limited to the interpolation filter, and can be realized by using frequency conversion such as FFT, DFT, MDCT, and the conversion. For example, in the case of upsampling, the data is converted to the frequency conversion domain by FFT, DFT, MDCT, etc., and the data is expanded by adding zero data to the high frequency side of the frequency domain data obtained by the conversion (virtual) A method of obtaining an upsampled input signal by inversely transforming the expanded data is also effective. By doing so, since high-speed calculations such as FFT and MDCT can be used, the sampling rate can be converted with a smaller amount of calculation than the interpolation filter.

  The voice encoding unit 14 receives a 16 kHz sampling signal from the sampling rate conversion unit 12 and outputs an output code 19 obtained by encoding the signal.

  The speech coding method used by the speech coding unit 14 will be described by taking the CELP (Code Excited Linear Prediction) method as an example, but the speech coding method is not limited to this. The CELP method is described in, for example, MR Schroeder and BS Atal: "Code-Excited Linear Prediction (CELP): High-quality Speech at Very Low Bit Rates", Proc. ICASSP-85, pp.937-940, 1985. Has been.

  FIG. 2 is a block diagram showing a configuration of the speech encoding unit 14. The speech encoding unit 14 includes a spectral parameter encoding unit 21, a target signal generation unit 22, an impulse response calculation unit 23, an adaptive codebook search unit 24, a noise codebook search unit 25, and a gain codebook search unit. 26, a pulse position candidate setting unit 27, a wideband pulse position candidate 27a, a narrowband pulse position candidate 27b, and a sound source signal generation unit 28.

  The operation of speech encoding according to the embodiment of the present invention configured as described above will be described. The speech encoding unit 14 is a device that receives the speech signal 20 and outputs an output code 19 obtained by encoding the speech signal 20, and performs an operation described below.

  The spectrum parameter encoding unit 21 extracts a spectrum parameter by analyzing the received audio signal 20. Next, using the extracted spectrum parameters, a spectrum parameter codebook stored in advance in the spectrum parameter encoding unit 21 is searched, and a code that can better represent the spectrum envelope of the input speech signal A book index is selected, and the selected index is output as a spectrum parameter code (A). The spectrum parameter code (A) is a part of the output code 19.

  The spectral parameter encoding unit 21 outputs an unquantized LPC coefficient and a quantized LPC coefficient corresponding to the extracted spectral parameter. Hereinafter, in order to simplify the description, LPC coefficients that are not quantized and LPC coefficients that are quantized are also simply referred to as spectrum parameters.

  In the CELP system described here, an LSP (Line Spectrum Pair) parameter is used as a spectrum parameter used when encoding a spectrum envelope. However, the present invention is not limited to this. Predictive Coding) coefficient, K parameter, G. Other parameters such as the ISF parameter used in 722.2 may be used.

  The target signal generation unit 22 receives the audio signal 20, the spectrum parameter output from the spectrum parameter encoding unit 21, and the sound source signal from the sound source signal generation unit 28, and uses these input signals to generate a target signal. Calculate the signal X (n). As the target signal, a signal obtained by synthesizing an ideal sound source signal excluding the influence of past coding by a synthesis filter with auditory weight is used, but the target signal is not limited thereto. It is known that an auditory weighted synthesis filter can be realized by using spectral parameters.

The impulse response calculation unit 23 obtains an impulse response h (n) from the spectrum parameter output from the spectrum parameter encoding unit 21 and outputs it. This impulse response can be typically calculated using an auditory weighted synthesis filter H (z) having the following characteristics, which is a combination of a synthesis filter using an LPC coefficient and an auditory weight filter, but is not limited thereto.

Where 1 / Aq (z) is the quantized LPC coefficient

Represents a synthesis filter consisting of

It is. On the other hand, W (z) is an auditory weighting filter and is an LPC coefficient that is not quantized.

Consisting of

It is. p is the order of LPC, and it is known that about p = 16 to 20 is used in wideband speech coding assuming a speech signal having a bandwidth of about 0 to about 7 kHz.

  The adaptive codebook search unit 24 receives the spectrum parameter output from the spectrum parameter encoding unit 21 and the target signal X (n) output from the target signal generation unit 22, and inputs the input signal, the adaptive code A pitch period included in the speech signal is extracted from the adaptive codebook stored in the book search unit 24, and is encoded to obtain an index corresponding to the pitch period, and an adaptive code (L) is output. The adaptive code (L) forms part of the output code 19.

  Note that the adaptive codebook search unit 24 has a structure in which the excitation signal generated by the excitation signal generation unit 28 is input before the search of the adaptive codebook, and the adaptive codebook is updated with the input excitation signal. In the adaptive codebook, past sound source signals are stored.

  The adaptive codebook search unit 24 outputs an adaptive code vector from the adaptive codebook corresponding to the pitch period to the excitation signal generation unit 28. Furthermore, using this adaptive code vector and the auditory weighted synthesis filter, a signal corresponding to the contribution from the adaptive codebook (an adaptive code vector synthesized with auditory weight) is generated, and this is generated as a gain codebook search unit 26. Further, a signal component corresponding to the contribution of the adaptive codebook is subtracted from the target signal X (n) to generate a second target signal X2 (n) (hereinafter also referred to as target vector X2), which is noise. Output to the codebook search unit 25.

  The pulse position candidate setting unit 27 designates the position of the pulse searched by the noise codebook search unit 25 based on the notification from the control unit 15. In response to the notification from the control unit 15 of whether the sampling rate of the input signal is 16 kHz or 8 kHz (or whether the input signal is a wideband signal or a narrowband signal), the pulse position candidate setting unit 27 One of the wideband pulse position candidate 27a and the narrowband pulse position candidate 27b is selected, and the selected pulse position candidate is output.

  That is, when the pulse position candidate setting unit 27 receives a notification that the sampling rate of the input signal is 16 kHz, the pulse position candidate setting unit 27 selects the wideband pulse position candidate 27a and receives a notification that the sampling rate of the input signal is 8 kHz. The narrow band pulse position candidate 27b is selected.

  As described above, when the sampling rate of the input signal is 8 kHz, control is performed so that the noise codebook search unit 25 searches for an exceptional narrowband pulse position candidate 27b, which is different from normal wideband speech coding processing. By doing so, the operation of the speech encoding unit 14 is controlled.

  In the conventional wideband speech coding, since only a sampling rate of 16 kHz is assumed as an input signal, when the input signal before coding is a signal having only narrowband information of a sampling rate of 8 kHz, the signal is In order to encode, the only method is to first up-sample an input signal having a sampling rate of 8 kHz to 16 kHz and encode it as a normal wideband audio signal.

  In the conventional wideband excitation coding, pulse position candidates for representing the excitation signal are prepared at high sampling rate positions corresponding to the wideband.

  In such a case, when the encoding bit rate is, for example, about 10 kbit / s or less, it is impossible to allocate a large number of bits to a pulse for representing a sound source signal, and in particular, bits are used inefficiently at the pulse position. For this reason, it becomes difficult to generate a pulse for sufficiently expressing the sound source signal. As a result, the sound quality of the audio signal encoded and reproduced tends to deteriorate.

  On the other hand, the wideband speech encoding apparatus according to the present embodiment determines whether the input signal is wideband or narrowband even when the input signal of 8 kHz sampling rate is converted to the sampling rate of 16 kHz and input to the speech encoding unit 14. Therefore, the speech encoding unit 14 can be adapted to either wideband or narrowband using this detection result.

  In this way, when the input signal is a narrow-band signal, the pulse position candidate for representing the sound source signal is a candidate for a pulse position with an unnecessarily fine resolution by reducing the sampling rate to 8 kHz, for example. It is possible to prevent the use of a bit.

  In addition, since the resolution of pulse position candidates can be appropriately reduced, the surplus bits can be used for other information. For example, the number of pulses can be increased. Thus, the sound source signal can be expressed more efficiently. Therefore, even if the bit rate is as low as about 10 to 6 kbit / s, there is an effect that an audio signal can be encoded with higher quality with respect to an input signal having an 8 kHz sampling rate.

  FIG. 3 shows a pulse position candidate 27c of an integer sample position composed of integer sample positions as the wideband pulse position candidate 27a, and an even sample position composed of even sample positions as the narrowband pulse position candidate 27b. A block diagram when the pulse position candidate 27d is used is shown.

  FIG. 4 shows an example of a pulse position candidate 27c at an integer sample position when an algebraic codebook is used. Here, the sound source signal is represented by four pulses, and each pulse has an amplitude of +1 or -1. The section for encoding the excitation signal is called a subframe. Here, the subframe length is 64 samples, and each pulse is selected from 0 to 63 sample positions in the subframe.

  In the algebraic codebook shown in FIG. 4, integer sample positions 0 to 63 in a subframe are divided into four tracks, and each track has only one pulse. For example, pulse i0 is one of the pulse position candidates {0, 4, 8, 12, 16, 20, 24, 28, 32 36, 40, 44, 48, 52, 56, 60} included in track 1 Indicates that one position is selected. In this example, for each track, 16 bits are required for 16 pulse position candidates and 1 bit is required for the pulse amplitude to encode the pulse, so 4 pulses require (4 + 1) × 4 = 20 bits. .

  The configuration of the algebraic codebook shown in FIG. 4 is an example, and the configuration is not limited to this. In any case, four pulses are selected from integer sample position candidates in the subframe.

  FIG. 5 shows a pulse position candidate 27d for even sample positions. Here, each pulse is configured to be selected from pulse position candidates arranged only at even sample positions among 0 to 63 sample positions in the subframe. However, even if there are several odd-numbered sample position candidates in addition to even-numbered sample positions, the essence of the pulse-position candidates is not impaired, and it goes without saying that this case is also included in the present invention.

  In the pulse position candidate 27d of the even sample position, the sound source signal is represented by five pulses, and each pulse has an amplitude of +1 or -1. In the algebraic codebook of FIG. 5, pulse position candidates capable of generating each pulse are arranged only at even sample positions among 0 to 63 sample positions in the subframe.

  In addition, even sample positions are divided into five tracks in the subframe, and each track has only one pulse. For example, the pulse i0 is selected from any one of the pulse position candidates {0, 8, 16, 24, 32, 40, 48, 56} included in the track 1.

  In the pulse position candidate 27d of the even sample position, for each track, 5 bits are set up if 3 bits are given to 8 kinds of pulse position candidates and 1 bit is given to the pulse amplitude when 20 bits are given. It becomes possible. That is, (3 + 1) × 5 = 20 bits.

  The configuration of the pulse position candidate 27d at the even-numbered sample position shown here is an example, and various track configurations are conceivable. In any case, the pulse for the narrow band is generated from the even-numbered sample position in the subframe. It is selected from the position candidates to be configured.

  FIG. 6 shows a pulse position candidate 27c of an integer sample position composed of integer sample positions as the wideband pulse position candidate 27a, and an odd sample position composed of odd sample positions as the narrowband pulse position candidate 27b. The block diagram at the time of using the pulse position candidate 27e is shown.

  FIG. 7 shows a pulse position candidate 27e at an odd sample position. The pulse position candidate 27e at the odd sample position has a configuration in which a pulse is selected from pulse position candidates arranged only at the odd sample position. Even in this case, the same effect can be obtained.

  In the pulse position candidate 27e at the odd sample position, the sound source signal is represented by five pulses, and each pulse has an amplitude of +1 or -1. In the algebraic codebook shown in FIG. 7, pulse position candidates capable of generating each pulse are arranged only at odd sample positions among 0 to 63 sample positions in the subframe. Further, the odd-numbered sample position is divided into five tracks in the subframe, and each track has only one pulse.

  For example, the pulse i0 is selected from any one of the pulse position candidates {1, 9, 17, 25, 33, 41, 49, 57} included in the track 1. In this example, if 3 bits are given to 8 kinds of pulse position candidates and 1 bit is given to the pulse amplitude for each pulse, 5 pulses can be set up if 20 bits are given. That is, (3 + 1) × 5 = 20 bits.

  The configuration of the algebraic codebook shown here is merely an example, and various configurations of the track are conceivable. In any case, the narrowband pulse is selected from the position candidates of the odd sample positions. .

  The narrow-band pulse position candidate 27b can be configured in another way, and the even-numbered sample position and the odd-numbered sample position are switched every subframe, or the even-numbered sample position and the odd-numbered sample position are switched every subframe. good.

  The point is that the narrow-band pulse position candidate is located at the sample position thinned out from the wide-band pulse position candidate, and it corresponds to the ratio of the narrow-band bandwidth to the wide-band bandwidth. If the structure is such that pulse position candidates are given at a thinning rate, the pulse position candidates used for the narrowband sound source function sufficiently. In that case, it goes without saying that any configuration is included in the present invention.

  In the present embodiment, the bandwidth of the narrowband signal is about 4 kHz (in the case of a signal obtained by upsampling an input signal of 8 kHz sampling to 16 kHz originally), and the bandwidth of the wideband signal is about 8 kHz (in the case of a normal 16 kHz sampling signal). ), The thinning-out sample position thinning method seems to have lowered the sampling rate to 1/2 (of course, it may be a thinning rate of 1/2 or more, such as 2/3). Any configuration may be used as long as the pulse position candidate is located at the correct position. Therefore, the narrow-band pulse position candidate is 27b, and the position is thinned out by half compared to the wide-band pulse position candidate 27a.

  If no consideration is given to the case where a signal that is a narrowband speech signal is encoded by the wideband speech encoder, the narrowband speech signal is also represented by, for example, a wideband pulse position candidate 27a shown in FIG. The same high-resolution time position pulse position candidate as that of a normal wideband signal is used.

  Using such position candidates with high temporal resolution, only a few pulses can be generated with a limited number of bits, but due to unnecessarily fine resolution, several pulses are overly concentrated in adjacent integer samples. In other words, the pulse is not distributed to other necessary positions, which is insufficient as a sound source signal. As a result, the reproduced sound is deteriorated.

  In this embodiment, it is detected that the original input signal is a narrowband signal, and low-resolution pulse position candidates suitable for the narrowband signal are used. Therefore, bits for representing the pulse position are wasted in the highband signal. It can be prevented from being used. Furthermore, since the pulse is limited so that the pulse only appears at a position having a low time resolution, the plurality of pulses for representing the sound source signal is not unnecessarily concentrated, and more pulses are generated. It is possible to stand. Therefore, it is possible to provide higher quality reproduced audio.

  Returning to FIG. 2, the noise codebook search unit 25 uses the algebraic codebook composed of pulse position candidates output from the pulse position candidate setting unit 27, that is, the code of the code vector that minimizes distortion, that is, Search for noise code (K). The algebraic codebook limits the possible values of the amplitudes of predetermined Np pulses to +1 and -1, and outputs a pulse vector according to pulse position information and amplitude information (ie, polarity information) as a code vector It is a codebook of the structure to do.

  As a feature of the algebraic codebook, the code vector itself is not stored directly, but only the arrangement information about the pulse position candidates and the pulse polarity is stored. Although the amount is small and the amount of calculation for selecting the code vector is small, the noise component included in the sound source information can be expressed with relatively high quality.

  Such a method using an algebraic codebook for encoding a sound source signal is called an ACELP (Algebraic Code Excited Linear Prediction) method, and it is known that synthesized speech with relatively little distortion can be obtained.

  Under such a structure, the noise codebook search unit 25 includes the pulse position candidate output from the pulse position candidate setting unit 27, the second target signal X2 output from the adaptive codebook search unit 24, and the impulse. The impulse response h (n) output from the response calculation unit 23 is input, and the auditory weighted synthesized code generated using the output signal (code vector) from the algebraic codebook according to the pulse position candidate. The distortion between the vector and the second target vector X2 is evaluated, and an index that reduces the distortion, that is, a noise code (K) is searched.

The evaluation value used at this time is

Searching for the code of the code vector that maximizes this value is equivalent to selecting a code that minimizes distortion. Here, the superscript t represents the transposition of the matrix, H represents an impulse response matrix composed of the impulse response h (n), and ck represents the code vector from the codebook corresponding to the code k.

  The noise codebook search unit 25 outputs the searched noise code (K) and a code vector synthesized with auditory weights and a code vector corresponding to this code. The noise code (K) forms part of the output code 19.

When the noise codebook is realized by an algebraic codebook, it is composed of several (here, Np) non-zero pulses.

It can be expressed as. Here, mi is the position of the i-th pulse, θj is the amplitude of the i-th pulse, and f (n) is an element of the correlation vector X2tH. The denominator of the evaluation value is

It can be expressed as. The selection of pulse position information is completed by searching for a pulse position mj (i = 0 to Np) that maximizes the distortion evaluation value (X2tHkk) 2 / (cktHtHkk) based on these. Here, the pulse position mj to be searched is limited to the pulse position candidates set by the pulse position candidate setting unit 27. By doing so, it is possible to search for an algebraic codebook composed of pulse position candidates output from the pulse position candidate setting unit 27.

  At this time, by calculating in advance the necessary values of f (n) and ψ (i, j) used for the code search, the amount of calculation required for the code search becomes very small. The pulse position information thus selected is output as a noise code (K) together with the pulse amplitude information. The noise codebook search unit 25 outputs a code vector corresponding to the noise code and a code vector synthesized with auditory weights.

  The gain codebook search unit 26 receives the adaptive code vector synthesized with auditory weights output from the adaptive codebook search unit 24 and the code vector synthesized with auditory weights output from the noise codebook search unit 25. In order to express the gain component of the sound source, two types of gains, that is, a gain used for the adaptive code vector and a gain used for the code vector (for the sake of simplicity, the two types of gain may be simply referred to as gains hereinafter). Encode.

  The gain codebook search unit 26 has a small distortion between the auditory weighted synthesized speech signal reproduced using the gain candidates extracted from the gain codebook stored therein and the target signal (X (n) in this embodiment). A gain code (G) that is an index is searched. Then, the searched gain code (G) and the corresponding gain are output. The gain code (G) forms part of the output code 19.

  The excitation signal generator 28 uses the adaptive code vector output from the adaptive codebook search unit 24, the code vector output from the noise codebook search unit 25, and the gain output from the gain codebook search unit 26. Generate a sound source signal.

  The excitation signal is obtained by multiplying the adaptive code vector by the gain for the adaptive code vector, multiplying the code vector by the gain for the code vector, multiplying the gain by the adaptive code vector and the code vector after multiplying the gain. Although it is obtained by adding, it is not limited to this.

  The obtained excitation signal is stored in the adaptive codebook in the adaptive codebook search unit 24 so that it can be used by the adaptive codebook search unit 24 in the next coding section. Further, the generated excitation signal is used in the target signal generation unit 22 to calculate a target signal for encoding in the next section.

  Next, processing of the wideband speech encoding method of the present invention will be described with reference to the flowchart of FIG.

The band detector 11 identifies whether the input audio signal is a wideband signal (step 810). As a result of the identification, if the signal is a wideband signal, encoded data is generated by performing predetermined wideband encoding (step 850), and the process ends. On the other hand, if the signal is identified as a narrowband signal, the sampling rate conversion of the input signal is performed so as to match the sampling rate assumed by the wideband speech coding unit (usually 16 kHz) as an exceptional process. (Step 820). Next, using narrowband parameters for performing exceptional wideband speech coding, encoded data is generated by performing wideband speech coding processing modified for narrowband processing (step 840). The process is terminated. In step 840, the portion where the processing is corrected for narrowband is at least a part of the wideband speech encoding processing, and an example is the pulse position used in the noise code search unit. Is to correct the candidate.

  This is the end of the description of the wideband speech coding method of the present invention using the flowchart of FIG.

(Second Embodiment)
Hereinafter, a second embodiment of the wideband speech encoding apparatus according to the present invention will be described with reference to the drawings with a focus on differences from the first embodiment. FIG. 7 shows a block diagram of the speech encoding unit 14 according to the second embodiment. Here, compared with the speech encoding unit 14 according to the first embodiment shown in FIG. 2, the same components are denoted by the same reference numerals, and description thereof is omitted.

  The speech encoding unit 14 according to the second embodiment shown in FIG. 9 has a parameter order setting unit 31 as compared with the speech encoding unit 14 according to the first embodiment shown in FIG. 31 outputs the parameter order. The spectrum parameter encoding unit 21a operates in the same manner as the spectrum parameter encoding unit 21 according to the first embodiment except that the parameter order is variable and the parameter order output by the parameter order setting unit 31 is changed. Enter and use.

  Further, there is no pulse position candidate setting unit 27 and narrow band pulse position candidate 27b, and a wide band pulse position candidate 27a is always set in the noise codebook search unit 25. The broadband pulse position candidate 27a is omitted in FIG.

  The parameter order setting unit 31 sets the order of the LSP parameters used by the spectrum parameter encoding unit 21 a based on the notification from the control unit 15. That is, upon receiving notification that the sampling rate of the input signal is 16 kHz, the parameter order setting unit 31 selects and outputs the broadband LSP order. In addition, upon receiving a notification that the frequency is 8 kHz, the narrowband LSP order is selected and output.

  As the LSP order p, when the input signal is a wideband signal in the 7 to 8 kHz band, about p = 16 to 20 is used. However, when the input signal is a narrowband signal, the exception is about p = 10. The value of is used. Thus, since the LSP order can be limited to an appropriate level for a narrowband signal, there is an effect that the number of bits required for encoding spectral parameters can be reduced accordingly.

  Even when the spectrum parameter used by the spectrum parameter encoding unit 21a is not an LSP parameter but an LPC parameter, a K parameter, an ISF parameter, or the like, the order is limited to an appropriate level for a narrowband signal, like the LSP parameter. Processing can be performed.

  The control operation of the control unit 15 in the second embodiment is the same as the control operation of the control unit 15 in the first embodiment whose flowchart is shown in FIG. However, the wideband encoding process of step 850 causes the parameter order setting unit 31 to set the wideband LSP order and causes the speech encoding unit 14 to perform the wideband speech encoding process.

  In the wideband coding process modified for narrowband in step 840, the parameter order setting unit 31 sets the narrowband LSP order, and the speech coding unit 14 performs the narrowband speech coding process. Become.

The present invention can be applied to various other applications. In a wideband speech encoding apparatus having sampling rate conversion means for an input speech signal, the input speech signal is converted according to the sampling rate conversion of the input speech signal. By using the identification information of wideband signal or narrowband signal,
・ Pre-processing part,
The number of parameters and the number of encoding candidates used in the adaptive codebook search unit or the pitch analysis unit / gain codebook search unit can be adaptively controlled.

  The present invention can also be applied to bit rate control of variable rate wideband speech coding. That is, by identifying whether the input speech signal is a wideband signal or a narrowband signal, it is possible to efficiently control the bit rate of the wideband speech encoding means. For example, if the input speech signal is a wideband signal, the input signal is suitable for the wideband speech coding unit, and therefore the coding bit rate can be lowered to some extent.

  On the other hand, when the input speech signal is a narrowband signal, as described above, since it is a signal that is not normally assumed by the wideband speech coding unit, the coding efficiency tends to be poor. In such a case, The bit rate is controlled so as to increase the bit rate of encoding. However, it is not necessary to control the bit rate to be increased in a section where the input audio signal is silent. In this way, when the input audio signal is detected as a narrow-band signal and when the voice activity is high, such as whether there is sound or no sound, control that increases the coding bit rate is performed. Working on the section has the effect of reducing the average bit rate because the bit rate can be kept low in a section where the voice activity is low.

  By doing in this way, even if an input signal is a wideband signal or a narrowband signal, there exists an effect which can provide the quality beyond a fixed level stably.

DESCRIPTION OF SYMBOLS 10 Input audio | voice signal 11, 11a Band detection part 12 Sampling rate conversion part 14 Speech encoding part 15 Control part 19 Output code 20 Speech signal 21, 21a Spectral parameter encoding part 22 Target signal generation part 23 Impulse response calculation part 24 Adaptive code Book search unit 25 Noise codebook search unit 26 Gain codebook search unit 27 Pulse position candidate setting unit 31 Parameter order setting unit

Claims (2)

In a wideband speech encoding method that represents an input speech signal by a spectrum parameter and a sound source signal, and that encodes the spectrum parameter and the sound source signal,
Selecting a plurality of pulses from given pulse position candidates, and encoding the sound source signal using the selected pulses; and
A process of identifying whether the input audio signal is a wideband audio signal or a narrowband audio signal;
If the input speech signal is identified as a wideband speech signal by the identification process, the encoding process is controlled to select a pulse position candidate having a predetermined resolution for the wideband speech signal, and When the input speech signal is identified as a narrowband speech signal by the identification process, a control process for controlling the encoding process to reduce the resolution of the pulse position candidate;
A process of converting the sampling rate of the input audio signal to a sampling rate for a wideband audio signal;
A wideband speech encoding method comprising: an output process for outputting an encoding result obtained by the encoding process. In a wideband speech coding apparatus that represents an input speech signal by a spectrum parameter and a sound source signal and encodes the spectrum parameter and the sound source signal,
Encoding means for selecting a plurality of pulses from given pulse position candidates and encoding the sound source signal using the selected pulses;
Identifying means for identifying whether the input audio signal is a wideband audio signal or a narrowband audio signal;
When the identifying means identifies the input speech signal as a wideband speech signal, the coding means is controlled to select a pulse position candidate having a predetermined resolution for the wideband speech signal, and the identifying means Control means for controlling the encoding means to lower the resolution of the pulse position candidate when the input audio signal is identified as a narrowband audio signal;
Conversion means for converting the sampling rate of the input audio signal into a sampling rate for a wideband audio signal;
A wideband speech encoding apparatus comprising: output means for outputting a result of encoding by the encoding means.

Images