HK1069247B

HK1069247B - Speech bandwidth extension apparatus and speech bandwidth extension method

Info

Publication number: HK1069247B
Application number: HK05102460.2A
Authority: HK
Inventors: 小泽一范
Original assignee: 日本电气株式会社
Priority date: 2001-07-26
Filing date: 2002-07-26
Publication date: 2007-03-23

Description

Voice bandwidth extension device and voice bandwidth extension method

Technical Field

The present invention relates to a speech bandwidth extension apparatus, and more particularly, to a speech bandwidth extension apparatus that decodes a speech signal encoded at a low bit rate, and then extends a reproduction frequency bandwidth thereof to improve auditory sound quality.

Background

Conventionally, as a speech bandwidth extension method, a method is known in which a speech signal encoded at a low bit rate is extended in a reproduction frequency bandwidth on the receiving side without transmitting auxiliary information on bandwidth extension from the transmitting side. For example, a paper entitled "Wireless band extension of telephone using high marker markov model" by P.Jax and P.Vary et al (proc. IEEE spech Coding workshop. pp.133-135, 2000.).

In the conventional method, in order to model an HMM (markov model) based on a spectral envelope or filter coefficient of a wide-band speech, it is necessary to determine parameters of the HMM model in advance in an offline state based on an enormous speech database. In addition, in order to perform the frequency bandwidth expansion process on the receiving side in real time, a large amount of computation is required for the search using the HMM model.

The conventional speech bandwidth extension apparatus described above has a problem that an enormous speech database has to be referred to in order to determine the parameters of the HMM model. Furthermore, the following disadvantages are present: that is, in order to perform the frequency bandwidth expansion process on the receiving side in real time, a large amount of computation is required for the HMM model search.

Disclosure of Invention

An object of the present invention is to provide a speech bandwidth extension apparatus that can obtain speech with a good sound quality with a frequency bandwidth extended with a small amount of computation without receiving auxiliary information from a transmission side. To achieve the above object, the following steps are performed: an input reproduction speech signal is divided into frames, the frequency of a spectral parameter obtained for each frame is converted, a synthesis filter is composed of a plurality of bandwidth-extended linear prediction coefficients, and a speech signal with an extended bandwidth is reproduced by using a sound source signal passing through the synthesis filter.

The voice bandwidth extension device of the present invention is characterized by comprising: a spectral parameter calculation circuit that inputs the decoded reproduced speech signal and calculates a spectral parameter indicating a spectral characteristic; a coefficient calculation circuit that finds filter coefficients in which the frequency of the spectral parameter has been converted to a high frequency and the frequency bandwidth has been expanded; a sound/soundless judgment circuit which inputs the reproduced speech signal and outputs sound/soundless judgment information and a pitch period; a gain adjusting circuit for outputting a gain according to the voiced/unvoiced determination information; an adaptive codebook circuit for inputting the pitch period and generating an adaptive codevector from a previous sound source signal; a noise generation circuit that generates a noise signal whose bandwidth is limited; a gain circuit that inputs the adaptive codevector and the noise signal and applies an appropriate gain to at least one of them; a first adder that performs addition operation on the output of the gain circuit and outputs a sound source signal; a synthesis filter circuit that passes the sound source signal through a synthesis filter composed of a plurality of the filter coefficients, thereby outputting a sound source signal having an expanded frequency bandwidth; a sampling frequency conversion circuit that inputs the reproduced voice signal and outputs a signal converted at a predetermined sampling frequency; and a second adder for adding an output of the sampling frequency conversion circuit and an output of the synthesis filter circuit to output a reproduced speech signal having an expanded bandwidth.

Further, a speech bandwidth extension apparatus of the present invention is characterized in that the apparatus is constituted by: a spectral parameter calculation circuit that inputs the decoded reproduced speech signal and calculates a spectral parameter indicating a spectral characteristic; a coefficient calculation circuit that finds filter coefficients in which the frequency of the spectral parameter has been converted to a high frequency and the frequency bandwidth has been expanded; a sound/soundless judgment circuit which inputs the reproduced speech signal and outputs sound/soundless judgment information; a gain adjusting circuit for outputting a gain according to the voiced/unvoiced determination information; a noise generation circuit that generates a noise signal whose bandwidth is limited; a gain circuit that inputs the noise signal and outputs a sound source signal to which an appropriate gain is applied; a synthesis filter circuit that passes the sound source signal through a synthesis filter composed of a plurality of the filter coefficients, thereby outputting a sound source signal having an expanded frequency bandwidth; a sampling frequency conversion circuit that inputs the reproduced voice signal and outputs a signal converted at a predetermined sampling frequency; and an adder that adds an output of the sampling frequency conversion circuit and an output of the synthesis filter circuit and outputs a reproduced speech signal of an expanded bandwidth.

Further, the spectral parameter calculation circuit may be configured to perform calculation of a predetermined order for the spectral parameter indicating the spectral characteristic for each frame and output the spectral parameter after dividing the reproduced speech signal into frames.

Further, the coefficient calculation circuit is characterized in that the circuit converts out a filter coefficient (linear prediction coefficient) of a predetermined order in which the frequency of the spectral parameter has been converted into a high frequency and outputs the filter coefficient.

The adaptive codebook circuit is characterized in that the circuit inputs the pitch period and outputs an adaptive codebook vector of an adaptive codebook from a previous sound source signal for each frame.

Further, the noise generation circuit is characterized in that the circuit generates a noise signal of which the frequency bandwidth is limited, the average amplitude is normalized at a predetermined level, and the time length is equal to the frame length.

In addition, the speech bandwidth extension method of the present invention is a speech bandwidth extension method for extending a frequency bandwidth of a decoded reproduced speech signal, characterized in that an inputted reproduced speech signal is divided into frames and converted into filter coefficients (linear prediction coefficients), that is, filter coefficients in which the frequency of a spectrum parameter obtained for each frame has been converted into a high frequency and the frequency bandwidth has been extended, and a sound source signal obtained by adding a noise signal having a time length equal to a frame length and an adaptive codevector based on a past sound source signal is passed through a synthesis filter composed of a plurality of the filter coefficients to form a sound source signal having the extended frequency bandwidth. The extended sound source signal is added to a signal obtained by converting the reproduced speech signal at a sampling frequency of a high frequency component, thereby reproducing a speech signal of an extended frequency bandwidth.

Drawings

Fig. 1 is a block diagram showing an embodiment of a voice bandwidth extension apparatus of the present invention.

Fig. 2 is a block diagram showing another embodiment of the voice bandwidth extension apparatus of the present invention.

Fig. 3 is a block diagram showing still another embodiment of the voice bandwidth extension apparatus of the present invention.

Detailed Description

Hereinafter, embodiments of the present invention will be described with reference to the drawings. Fig. 1 is a block diagram showing an embodiment of a voice bandwidth extension apparatus of the present invention.

The present embodiment shown in fig. 1 is constituted by: a spectral parameter calculation circuit 100 that inputs the decoded reproduced speech signal and calculates spectral parameters representing spectral characteristics; a coefficient calculation circuit 130 that finds filter coefficients in which the frequency of the spectral parameter has been converted into a high frequency and the frequency bandwidth has been expanded; a sound/soundless judgment circuit 200 that inputs a reproduced speech signal and outputs sound/soundless judgment information and a pitch period; a gain adjustment circuit 210 that outputs a gain according to the voiced/unvoiced determination information; an adaptive codebook circuit 110 that inputs pitch periods and generates an adaptive codebook from a past sound source signal; a noise generation circuit 120 that generates a noise signal with a limited bandwidth; a gain circuit 140 that inputs the adaptive codevector and the noise signal and applies an appropriate gain to at least one of them; an adder 160 that adds the outputs of the gain circuit 140 and outputs a sound source signal; a synthesis filter circuit 170 that passes the sound source signal through a synthesis filter composed of a plurality of filter coefficients and outputs the sound source signal with the frequency bandwidth expanded; a sampling frequency conversion circuit 180 which inputs a reproduced voice signal and outputs a signal converted at a predetermined sampling frequency; and an adder 190 that adds the output of the sampling frequency conversion circuit 180 and the output of the synthesis filter circuit 170 to output a reproduced speech signal of an expanded bandwidth.

Next, the operation of the speech bandwidth extension apparatus according to the present embodiment will be described in detail with reference to fig. 1. In the following description, it is assumed that the frequency bandwidth is extended from 4kHz to 5kHz or 7kHz in the input reproduced voice signal.

Referring to fig. 1, a spectral parameter calculation circuit 100 inputs a decoded reproduced speech signal, divides it into frames (e.g., 10ms), then performs calculation of a predetermined order for spectral parameters representing spectral characteristics for each frame (e.g., P ═ 10 times), and outputs to a coefficient calculation circuit 130.

Here, for the calculation of the spectral parameters, a well-known LPC (linear predictive coding) analysis, a Burg analysis, or the like can be used. In this example, the Burg analysis was used. Details of the Burg analysis are described in pages 82 to 87 of a single file book entitled "signal analysis and system discrimination" (journal of 1998 by corona (コロナ)) written by the middle school (famous person), and therefore, a description thereof is omitted.

The spectrum parameter calculation circuit 100 converts the linear prediction coefficient α i (i ═ 1, … P) calculated by the Burg method into LSP parameters suitable for quantization or interpolation, and outputs the LSP parameters.

Here, the conversion from the linear prediction coefficient to the LSP parameter can be referred to a paper entitled "compression of speech information by Line Spectral Pair (LSP) speech analysis synthesis method" written in "village (trade name) or the like (journal of the society of electronic communications, J64-a.pp.599-606.1981).

The coefficient calculation circuit 130 receives the LSP parameters output from the spectral parameter calculation circuit 100, converts the LSP parameters into coefficients of a signal having an expanded frequency bandwidth, and outputs the coefficients to the synthesis filter circuit 170. In this conversion, for example, a known method such as a method of converting the frequency of the LSP parameter only at a high frequency, a nonlinear conversion method, or a linear conversion method may be used. Also, all or a part of the LSP parameters is used here, and after the frequency of the LSP parameters is converted to a high frequency, it is converted to linear prediction coefficients (filter coefficients) of a predetermined order M.

The sound/soundless decision circuit 200 inputs the decoded reproduced speech signal and decides whether the signal of each frame is sound or soundless. Next, a specific determination method is described. The signal of each frame is judged to be a voiced part if the maximum value of the normalized autocorrelation function d (t) is larger than a predetermined threshold value, and judged to be an unvoiced part if it is small. The normalized autocorrelation function d (t) of the reproduced speech signal x (n) up to the predetermined delay time m can be calculated by the following formula (1). The judged voiced/unvoiced judgment information is input to the gain adjustment circuit 210. In addition, the signal of each frame of the voiced part outputs the value T that maximizes the normalized autocorrelation function d (T) to the adaptive codebook circuit 110 as the pitch period T. In the above equation (1), N is the number of samples for calculating the normalized autocorrelation.

The gain adjustment circuit 210 receives the sound/soundless decision information from the sound/soundless decision circuit 200, and outputs the gain of the adaptive codebook signal and the gain of the noise signal to the gain circuit 140 according to whether the sound portion or the soundless portion is present.

The adaptive codebook circuit 110 receives the pitch period of the adaptive codebook from the voiced sound/unvoiced sound judging circuit 200, generates an adaptive codevector, and outputs the adaptive codevector. The adaptive codebook circuit 110 also generates an adaptive codebook component from a previous sound source signal.

The noise generation circuit 120 generates a noise signal whose average amplitude is normalized at a predetermined level and for a time length equal to the frame length on the basis that the frequency bandwidth has been limited, and then outputs it to the gain circuit 140. Here, white noise is used as an example of the noise signal, but a noise signal having another statistical distribution may be used.

The gain circuit 140 inputs the gain of the adaptive codebook signal output from the gain adjustment circuit 210 and the gain of the noise signal, multiplies at least one of the adaptive codevector output from the adaptive codebook circuit 110 and the noise signal output from the noise generation circuit 120 by an appropriate gain, and outputs the respective signals to the adder 160.

The adder 160 outputs the sound source signal obtained by adding the two signals output from the gain circuit 140 to the filter circuit 170 and the adaptive codebook circuit 110.

The synthesis filter circuit 170 inputs the linear prediction coefficient (filter coefficient) of the order M output from the coefficient calculation circuit 130, and composes a synthesis filter. The synthesis filter circuit 170 inputs the sound source signal output from the adder 160 and outputs the sound source signal of which the frequency bandwidth has been extended.

The sampling frequency conversion circuit 180 inputs a reproduced speech signal and outputs a signal converted at a sampling frequency of a predetermined integral multiple. The converted signal retains the components before frequency spreading.

The adder 190 adds the sound source signal output from the synthesis filter circuit 170 to the signal output from the sampling frequency conversion circuit 180, thereby forming and outputting a reproduced speech signal having an expanded frequency bandwidth.

According to the present embodiment, an input reproduced speech signal is divided into frames, converted into filter coefficients (linear prediction coefficients), that is, the frequency of the spectral parameter or LSP parameter obtained for each frame is converted into a high frequency and the filter coefficient of the frequency bandwidth is expanded, and adds a noise signal of a time length equal to the frame length to an adaptive codevector based on the past sound source signal, and the obtained sound source signal is passed through a synthesis filter composed of the synthesis coefficients to form a sound source signal having an expanded frequency bandwidth, and then the expanded sound source signal is added with a signal, that is, a signal obtained by converting an input reproduced speech signal at a sampling frequency of a high frequency component, thereby reproducing a voice signal of which frequency bandwidth has been expanded, and thus it is not necessary to receive information for bandwidth expansion from a transmitting side, and it is not necessary to perform a large number of operations based on HMM as in the existing method. In addition, since white noise or the like is used as the sound source information, it can be very easily processed.

Next, another embodiment of the present invention will be explained. Fig. 2 is a block diagram showing another embodiment of the voice bandwidth extension apparatus of the present invention. Since the constituent elements denoted by the same reference numerals as those in fig. 1 perform the same operations as those in fig. 1, the description thereof will be omitted.

In fig. 2, the gain adjustment circuit 310 inputs sound/soundless determination information from the sound/soundless determination circuit 200, and outputs a signal for adjusting the gain of the noise signal to the gain circuit 300 according to whether it is a sound part or a soundless part.

The gain circuit 300 inputs the gain of the noise signal output from the gain adjustment circuit 310, multiplies the noise signal output from the noise generation circuit 120 by the gain, and outputs the resultant signal to the synthesis filter circuit 170.

Here, the adaptive codebook circuit 110 shown in fig. 1 is used to generate a periodic component included in a vowel or the like of a speech signal. In addition, since the vowel signal does not generally reach a high frequency, it may be omitted in the voice bandwidth extension apparatus. Therefore, since the adaptive codebook circuit 110 is eliminated, the data processing amount can be reduced.

Next, still another embodiment of the present invention will be explained. Fig. 3 is a block diagram showing another embodiment of the voice bandwidth extension apparatus of the present invention.

In the speech bandwidth extension apparatus in another embodiment, as shown in fig. 3, the speech decoder is configured in the front section, wherein the speech decoder is composed of: signal separator 505, gain decoding circuit 510, adaptive codebook circuit 520, sound source signal recovery circuit 540, spectral parameter decoding circuit 570, adder 550, synthesis filter circuit 550, gain codebook 380, and sound source codebook 351.

Here, the spectral parameter decoding circuit 570 also has the operation of the spectral parameter calculation circuit 100 shown in fig. 1. Thereby, the structure is simplified. Since the constituent elements denoted by the same reference numerals as those in fig. 1 perform the same operations as those in fig. 1, the description thereof will be omitted.

In fig. 3, the demultiplexer 505 demultiplexes and outputs, as speech information, an index indicating a gain code vector, an index indicating an adaptive codebook delay, information of a sound source signal, an index of a sound source code vector, an index of a spectral parameter, and the like, which are multiplexed, from a received signal.

The gain decoding circuit 510 inputs an index indicating a gain code vector, reads the gain code vector from the gain codebook 380 based on the index, and outputs the read gain code vector.

The adaptive codebook circuit 520 inputs an index indicating a delay of an adaptive codebook, generates an adaptive codevector whose gain is composed of a gain codevector output from the gain decoding circuit 510, multiplies the adaptive codevector by the gain of the adaptive codebook, and outputs the resultant adaptive codevector. Then, an adaptive codebook component is generated from the previous drive sound source signal.

The acoustic source signal recovery circuit 540 generates an acoustic source pulse using the index of the acoustic source code vector received from the demultiplexer 505, the information of the acoustic source signal, and the polar code vector read out from the acoustic source codebook 351, and outputs the acoustic source pulse to the adder 550.

The adder 550 generates a driving sound source signal v (n) according to the following expression (2) represented by the numeral 2 using the adaptive code vector output from the adaptive codebook circuit 520 and the sound source pulse output from the sound source signal recovery circuit 540, and outputs the driving sound source signal v (n) to the adaptive codebook circuit 520 and the synthesis filter circuit 560.

The spectral parameter decoding circuit 570 decodes the spectral parameters after inputting the indices of the spectral parameters, converts the spectral parameters into linear prediction coefficients, and outputs the linear prediction coefficients to the synthesis filter circuit 560 and the coefficient calculation circuit 130.

The synthesis filter circuit 560 inputs the linear prediction coefficient α i output from the spectral parameter decoding circuit 570 and the drive sound source signal v (n) output from the adder 550, and calculates and outputs a reproduction signal x (n) according to the following formula (3) indicated by numeral 3.

Industrial applicability of the invention

As described above, according to the speech bandwidth extension apparatus and speech bandwidth extension method of the present invention, the decoded reproduced speech signal is divided into frames, the frequency of the spectral parameter obtained for each frame is converted into a high frequency, and the filter coefficient (linear prediction coefficient) of the extended frequency bandwidth is obtained, whereby when converting the spectral parameter into a parameter of which the frequency bandwidth is extended, the conventional method such as HMM is not used, and thus the amount of computation can be reduced.

Further, by using a sound source signal obtained by adding a noise signal (white noise) of a time length equivalent to a long frame and an adaptive code vector based on a past sound source signal, it is possible to perform processing with a small amount of information very easily.

Further, since the audio signal with the expanded frequency bandwidth is reproduced by passing the audio signal through the synthesis filter composed of the filter coefficient with the expanded frequency bandwidth and adding the signal obtained by converting the reproduced audio signal by the sampling frequency of the high frequency component to the obtained audio signal with the expanded spectral bandwidth, it is possible to improve the auditory sound quality without receiving the necessary information for performing the bandwidth expansion process from the transmission side.

Claims

1. A voice bandwidth extension apparatus, characterized in that the apparatus is composed of:

a spectral parameter calculation circuit that inputs the decoded reproduced speech signal and calculates a spectral parameter indicating a spectral characteristic;

a coefficient calculation circuit for converting the frequency of the spectrum parameter into a high frequency and then obtaining a filter coefficient with a frequency bandwidth expanded according to the spectrum parameter with the frequency converted into the high frequency;

a sound/soundless judgment circuit which inputs the reproduced speech signal and outputs sound/soundless judgment information and a pitch period;

a gain adjusting circuit for outputting a gain according to the voiced/unvoiced determination information;

an adaptive codebook circuit for inputting the pitch period and generating an adaptive codevector from a previous sound source signal;

a noise generation circuit that generates a noise signal whose bandwidth is limited;

a gain circuit that inputs the adaptive codevector and the noise signal and applies a gain output from the gain adjustment circuit to at least one of them;

a first adder that performs addition operation on the output of the gain circuit and outputs a sound source signal;

a synthesis filter circuit that passes the sound source signal through a synthesis filter composed of a plurality of the filter coefficients, thereby outputting a sound source signal having an expanded frequency bandwidth;

a sampling frequency conversion circuit that inputs the reproduced voice signal and outputs a signal converted at a predetermined sampling frequency;

and a second adder that adds an output of the sampling frequency conversion circuit and an output of the synthesis filter circuit and outputs a reproduced voice signal of the expanded bandwidth.

2. A voice bandwidth extension apparatus, characterized in that the apparatus is composed of:

a sound/soundless judgment circuit which inputs the reproduced speech signal and outputs sound/soundless judgment information;

a gain circuit that inputs the noise signal and outputs a sound source signal to which the gain output by the gain adjustment circuit is applied;

and an adder that adds an output of the sampling frequency conversion circuit and an output of the synthesis filter circuit and outputs a reproduced speech signal of an expanded bandwidth.

3. The voice bandwidth extension apparatus according to claim 1 or 2, wherein the spectral parameter calculation circuit performs calculation of a predetermined order for the spectral parameter representing the spectral characteristic for each frame and outputs it after dividing the reproduced voice signal into frames.

4. The speech bandwidth extension apparatus according to claim 1 or 2, wherein the coefficient calculation circuit finds a filter coefficient of a predetermined order, i.e., a linear prediction coefficient, from the spectral parameter frequency-converted to a high frequency after frequency-converting the spectral parameter to a high frequency, and outputs it.

5. The speech bandwidth extension device according to claim 3, wherein the adaptive codebook circuit inputs the pitch period and outputs an adaptive codevector of the adaptive codebook from a past sound source signal for each frame.

6. The speech bandwidth extension apparatus of claim 3, wherein the noise generation circuit generates a noise signal whose frequency bandwidth is limited, whose average amplitude is normalized by a predetermined level, and whose length in time is equal to a frame length.

7. A speech bandwidth extension method for extending a frequency bandwidth of a decoded reproduced speech signal,

the input reproduced voice signal is divided into frames,

the frequency of the spectral parameter obtained for each frame is converted into a high frequency, then a filter coefficient with an expanded frequency bandwidth, i.e., a linear prediction coefficient is obtained from the spectral parameter whose frequency is converted into a high frequency,

passing a sound source signal, which is obtained by adding a noise signal of a time length equal to the frame length and an adaptive code vector based on a past sound source signal, through a synthesis filter composed of a plurality of the filter coefficients to form a sound source signal of an extended frequency bandwidth,

the reproduced voice signal is input, a signal converted at a predetermined sampling frequency is output, and the converted signal is added to the sound source signal having the expanded frequency bandwidth to reproduce a voice signal having the expanded frequency bandwidth.