JP3266920B2 - Audio encoding device, audio decoding device, and audio encoding / decoding device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech encoding system used for, for example, digitally transmitting or storing speech.
The present invention relates to a device, a speech decoding device, and a speech encoding / decoding device.

[0002]

2. Description of the Related Art A description of a basic technique of this kind in the prior art is given in "Multiband Excitati" by DWGriffin and JSLim.
on Vocoder ”(IEEE trans.on ASSP, Vol.36, No.8, pp.12
23-1235, Aug 1988) (hereinafter referred to as [Literature 1]), and further described by PCMeuse, "A 2400 bpsMulti-ba"
nd Excitation Vocoder ”(ICASSP 90, pp. 9-12, 1990)
(Hereinafter referred to as [Reference 2]). FIG. 5 is a configuration diagram of the conventional speech encoding device shown in the above [Reference 1].
1 is an encoding unit, 2 is a decoding unit, 3 is a transmission path, 4 is an audio input terminal, and 5 is an audio output terminal. 6 is pitch / harmonic component analysis means, 7 is voiced / unvoiced determination means, 8 is voiced / unvoiced coding means, 9 is pitch coding means, 10 is harmonic component coding means, 11 is voiced / unvoiced decoding. Means, 12 is pitch decoding means. 13 is a harmonic component decoding means, 14 is a voiced synthesis means, and 15 is an unvoiced synthesis means.

[0003] The operation of the conventional speech encoding / decoding apparatus will be described below with reference to FIG. First, the encoding unit 1 will be described. The pitch / harmonic component analysis means 6 analyzes the input voice from the voice input terminal 4 in units of a fixed length analysis frame,
The amplitude value Am and the phase θm (m is a harmonic number) of the harmonic component at the pitch frequency interval appearing on the pitch and frequency spectrum are extracted and output to the harmonic component encoding means 10.

[0004] The voiced / unvoiced determination means 7 sets the frequency domain to 1
It divides into a plurality of bands including three harmonics, performs voiced / unvoiced / judgment for each band, and outputs voiced / unvoiced information to voiced / unvoiced judgment coding means 8. In the voiced / unvoiced determination, a band having a harmonic structure is determined to be voiced, and a band having a distorted harmonic structure is determined to be unvoiced. The voiced / unvoiced determination encoding means 8 performs the voiced / unvoiced
The unvoiced determination information is encoded, and the code is output to the decoding unit 2 via the transmission path 3. The pitch encoding unit 9 encodes the pitch frequency and outputs encoded data to the decoding unit 2 via the transmission path 3. Harmonic component encoding means 1
0 encodes the amplitude values Am and θm, and outputs the encoded data to the decoding unit 2 via the transmission path 3.

FIG. 6 is a block diagram showing the internal structure of the harmonic component encoding means 10. The harmonic component encoding means of the 2.4 Kbps speech encoding / decoding apparatus described in [Reference 2] is described. It is. Hereinafter, the details of the harmonic component encoding means 10 will be described with reference to FIG. The fixed sample point spectrum extracting means 32 interpolates the value of the amplitude Am of the harmonic component and extracts a frequency spectrum represented by the samples of the fixed points (48). The frequency band dividing means 33 partially overlaps the spectrum and divides the spectrum into 18 samples in the low frequency band (0 to 1.5 kHz) and 36 samples in the high frequency band (1 to 4 kHz). The 18 samples in the low frequency band are M1-dimensional vector quantization means 34, and the 36 samples in the high frequency band are M
By the two-dimensional vector quantization means 35, M1
(18) Encode by performing dimensional vector quantization and M2 dimensional (36) vector quantization. At this time, the number of quantization bits in the vector quantization means is 10
Is a bit.

Next, the decoding section will be described. The voiced / unvoiced decoding unit 11 decodes the coded data of the voiced / unvoiced judgment information input from the transmission path 3, obtains a voiced / unvoiced judgment, and outputs it to the voiced synthesis unit 14 and the unvoiced synthesis unit 15. The pitch decoding unit 12 decodes the pitch-encoded data input from the transmission path 3, obtains a pitch frequency, and outputs the pitch frequency to the voiced synthesis unit 14 and the unvoiced synthesis unit 15.

The harmonic component decoding means 13 decodes the encoded harmonic component data inputted from the transmission line 3 to determine the amplitude Am and the phase θm of the harmonic component, and outputs a voiced synthesizing means 14 and an unvoiced synthesizing means. 15 is output. The voiced synthesizing unit 14 decodes the pitch frequency obtained by the pitch decoding unit 12 and the harmonic component decoding unit 13 in a frequency band in which the voiced / unvoiced determination obtained by the voiced / unvoiced decoding unit 11 is voiced. Based on the amplitude Am and the phase θm, a decoded voice of a voiced part is synthesized by superimposing a cosine wave represented by the equation (1) and output.

[0008]

(Equation 1)

The unvoiced synthesizing unit 15 converts the harmonic component decoding unit 1 into white noise in a frequency band in which the voiced / unvoiced determination determined by the voiced / unvoiced decoding unit 11 is unvoiced.
3. A decoded speech is synthesized by giving a spectrum amplitude obtained from the amplitude Am obtained in step 3.

[0010]

Since the encoding section of the amplitude of the harmonic component of the conventional speech encoding / decoding apparatus is configured as described above, the correlation is inherently strong and vector quantization must be performed collectively. Since the low-frequency and high-frequency regions of the audio signal spectrum are separated and the amplitude values of the harmonic components are separately vector-quantized, the quantization efficiency is poor, and the number of dimensions in vector quantization is large. When quantizing with a small number of vector quantization bits, there is a problem that quantization distortion increases and the quality of synthesized speech deteriorates.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-described problem, and it is possible to efficiently quantize the amplitude value of the harmonic component with a small bit amount with little quantization distortion and to obtain high quality decoding. It is intended to synthesize speech.

[0012]

A speech code according to the present invention
The digitizer analyzes the pitch analyzed for each frame of a certain length of time.
After converting the amplitude value of the harmonic at the frequency interval to the logarithmic domain,
Fourier transform and output low-order inverse Fourier transform coefficients
Harmonic amplitude envelope analysis means and the harmonic amplitude envelope analysis means
Quantizes the low-order inverse Fourier transform coefficients output from
Component quantization means for encoding by means of
The low-order inverse Fourier transform coefficient quantized by the
Fourier transform after inverse quantization, and Fourier transform coefficients
And the Fourier transform coefficient and the pitch frequency interval
Quantizes the difference value of each harmonic with the amplitude value of the harmonic component of
Differential quantizing means for performing encoding. In addition,
The audio decoding device according to the present invention performs one of the encoded harmonic components.
Decode the low-order inverse Fourier transform coefficients that are the
Envelope component inverse quantization means for outputting inverse Fourier transform coefficients
And the Fourier transform of this low-order inverse Fourier transform coefficient.
And harmonic amplitude envelope decoding means for obtaining a Fourier transform coefficient
And decode the difference that is part of the encoded harmonic component
Difference inverse quantization means, the decoded difference,
Add the Fourier transform coefficients obtained by the wave amplitude envelope decoding means.
The amplitude value of the harmonic component at the pitch frequency interval.
And addition means for performing the addition. Furthermore, the sound according to the present invention
The voice encoding / decoding device analyzes each frame of a certain time length.
The amplitude value of the harmonic at the pitch frequency interval
After performing the inverse Fourier transform, the low-order inverse Fourier transform
Harmonic amplitude envelope analyzing means for outputting a number,
The low-order inverse Fourier transform section output from the fault analysis means
An envelope component quantization means for quantizing and encoding the number,
The low-order inverse Fourier quantized by the envelope component quantizing means.
Fourier transform after inverse quantization of the
The Fourier transform coefficient is obtained, and the Fourier transform coefficient
Difference on each harmonic from the amplitude value of the harmonic component at the first frequency interval
Differential quantization means for quantizing and encoding values
And the encoded low-order transmitted from the encoding unit.
Decode the inverse Fourier transform coefficients and perform a low-order inverse Fourier transform
Envelope component inverse quantization means for outputting coefficients, and
Fourier transform of the Fourier transform coefficients
From the harmonic amplitude envelope decoding means for obtaining
Difference inverse quantization means for decoding received encoded difference
And the decoded difference and the harmonic amplitude envelope decoding
By adding the Fourier transform coefficients obtained in the step,
Adding means for decoding the amplitude value of the harmonic component at the switch frequency interval
Is provided in the decoding unit.

[0013]

The harmonic amplitude envelope analyzing means of the speech coding apparatus and the speech coding / decoding apparatus according to the present invention converts the amplitude value of the harmonic component at the pitch frequency interval analyzed for each fixed length frame into a logarithmic domain. After that, inverse Fourier transform is performed to obtain and output a low-order inverse Fourier transform coefficient. Envelope component quantizing means you encode inverse Fourier transform coefficients outputted in the low order of harmonic amplitude envelope analysis means and quantization. Differential quantization
Means dequantizing the low-order inverse Fourier transform coefficient quantized by the envelope component quantizing means and then performing Fourier transform to obtain a Fourier transform coefficient; and a harmonic component between the Fourier transform coefficient and the pitch frequency interval. On each harmonic with the amplitude value of
Is quantized and encoded. In addition, the present invention
The envelope component inverse quantization means of the audio decoding device and the audio encoding / decoding device performs inverse quantization of a low-order inverse Fourier transform coefficient which is a part of the encoded harmonic component . The harmonic amplitude envelope decoding means Fourier-transforms the low-order inverse Fourier transform coefficients to obtain Fourier transform coefficients. Difference inverse quantization means
Inverse quantization of the difference that is a part of the encoded harmonic component is performed. The adding means adds the difference output from the difference dequantizing means and the Fourier transform coefficient obtained by the harmonic amplitude envelope decoding means to obtain the harmonic component of the pitch frequency interval.
Decode the amplitude value.

[0014]

【Example】

Embodiment 1 FIG. FIG. 1 is a configuration diagram of an input speech encoding / decoding apparatus to which an embodiment of the present invention is applied. In FIG. 1, the same portions as those in FIG. 5 are denoted by the same reference numerals, and description thereof will be omitted. In FIG. 1, reference numeral 16 denotes harmonic component encoding means,
7 is a harmonic component decoding means. FIG. 2 is an internal configuration diagram of the harmonic component encoding unit 16 of the audio encoding / decoding apparatus embodying the present invention, FIG. 3 is an internal configuration diagram of the harmonic component decoding unit 17, and FIG. It is explanatory drawing which showed the example of harmonic amplitude and harmonic amplitude envelope.

The operation of one embodiment of the present invention will be described below. First, the operation of the harmonic component encoding means 16 will be described with reference to FIG. The harmonic amplitude envelope analyzing means 18 in the harmonic component coding means 16 is composed of a logarithmic amplitude means 19, an M-point discrete inverse Fourier transform means 20, and a low-order component extracting means 21, and is analyzed for each frame. After the amplitude value Am of the harmonic component is converted to a logarithmic amplitude by the logarithmic amplitude means 19, a discrete inverse Fourier transform is performed by the M-point discrete Fourier transform means 20, and a low-order inverse Fourier transform coefficient ( For example, 20th order) is output as a parameter expressing the envelope of the amplitude value Am of the harmonic. At this time, the number M of points of the discrete Fourier transform is twice the number of all harmonics in the frame.

Next, the low-order inverse Fourier transform component output from the harmonic amplitude envelope analysis means 18 is vector-quantized and coded by, for example, 10 bits by the envelope component quantization means 22, and the coded data is transmitted to the transmission path. Output.

The difference extracting means 23 comprises an envelope component inverse quantizing means 24, an N-point discrete Fourier transform means 25, and an adder 26, and the quantization output from the envelope component quantizing means 22 The obtained low-order inverse Fourier transform coefficient is inversely quantized by the envelope component inverse quantization means 24 to output a low-order inverse Fourier transform coefficient, and the N-point discrete Fourier transform means 25 decodes the Fourier transform coefficient.
Output to the adder 26. The value of N at this time is an arbitrary value of M or more.

The adder 26 inputs a signal obtained by sampling the decoded Fourier transform coefficients at pitch frequency intervals, inverting the sign, and a logarithmic harmonic amplitude Am output from the logarithmic amplitude means 19. And output the difference. The difference quantization unit 27 performs, for example, scalar quantization on the harmonic amplitude envelope difference output by the difference extraction unit 23, and outputs encoded data to a transmission path.

The harmonic amplitude envelope obtained by Fourier-transforming the low-order inverse Fourier transform coefficient output by the harmonic amplitude envelope analysis performed by the harmonic amplitude envelope analyzing means 18 and the amplitude value Am of the harmonic component. FIG. 4 shows an example of the relationship. In FIG. 4, the frequency is divided into a frequency band centered on the pitch frequency and its harmonic position, and the amplitude value Am of the harmonic component analyzed for each band is pointed to the center frequency position of the corresponding band. .

In this example, the number of harmonics is 26. Performing a 52-point discrete inverse Fourier transform on the harmonic amplitude Am,
The solid line represents the harmonic amplitude envelope obtained by quantizing the low-order inverse Fourier transform coefficient 20th order, performing inverse quantization, and performing discrete Fourier transform of 256 points (0-order except for the 20th order).
By sampling the harmonic amplitude envelope at pitch frequency intervals, a value of the harmonic amplitude Am approximated from the harmonic amplitude envelope is obtained. The harmonic amplitude envelope difference is a difference between Am obtained from the harmonic amplitude envelope and the amplitude value Am of the harmonic component.

Next, the harmonic component decoding means 1 will be described with reference to FIG.
7 will be described. The envelope component inverse quantization means 28 decodes the envelope component encoded data transmitted from the transmission path by an inverse quantization process, and outputs a low-order inverse Fourier transform coefficient. The harmonic amplitude envelope decoding means 29 converts the low-order inverse Fourier transform coefficient output from the envelope component inverse quantization means 28 into N
Perform a Fourier transform of the points, decode the Fourier transform coefficients,
Output to the adder 31. Here, N is the same value as the N-point Fourier transform used in the difference extracting means of the encoding device.

The differential inverse quantization means 30 inversely quantizes the differentially encoded data transmitted from the transmission path, and outputs the harmonic envelope difference to the adder 31. The adder 31 adds the value obtained by sampling the Fourier transform coefficient output from the harmonic amplitude envelope decoding means 29 at pitch frequency intervals and the difference output from the difference inverse quantization means 30 to obtain a logarithmic amplitude harmonic. Outputs the amplitude value.

In FIG. 2, the difference quantization means 27
Has been described using the scalar quantization method as an example, but vector quantization may be used instead of scalar quantization.

[0024]

As described above, according to the present invention, a parameter expressing the envelope of the amplitude value of the harmonic component of the input voice over the entire frequency band is obtained with a relatively small number of dimensions, and the parameter is vector-quantized. Since the difference between the amplitude value and the envelope is quantized, the audio encoding device has a small number of quantization bits.
Can efficiently quantize the amplitude value of the harmonic component
Harmonics quantized with fewer quantization bits in the decoder
High-quality decoded speech can be synthesized from the amplitude values of the components.
In addition, the speech encoding / decoding device has an effect that the amplitude value of the harmonic component can be efficiently quantized even with a small number of quantization bits, and a high-quality decoded speech can be synthesized.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of an encoding / decoding device showing an embodiment of the present invention.

FIG. 2 is a configuration diagram of harmonic component encoding means according to one embodiment of the present invention.

FIG. 3 is a configuration diagram of harmonic component decoding means according to an embodiment of the present invention.

FIG. 4 is an explanatory diagram showing an example of a harmonic amplitude envelope characteristic in the encoding / decoding device according to an embodiment of the present invention.

FIG. 5 is a configuration diagram of a conventional speech encoding / decoding device.

FIG. 6 is a configuration diagram of a conventional harmonic component encoding means device.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 Encoding part 2 Decoding part 3 Transmission line 6 Pitch / harmonic component analysis means 9 Pitch encoding means 10 Harmonic component encoding means 12 Pitch decoding means 13 Harmonic component decoding means 14 Voiced synthesis means 15 Unvoiced synthesis Means 17 Harmonic component decoding means 18 Harmonic amplitude envelope analysis means 22 Envelope component quantization means 23 Difference extraction means 24 Envelope component inverse quantization means 27 Difference quantization means 28 Envelope component inverse quantization means 29 Harmonic amplitude envelope decoding Means 30 Differential inverse quantization means

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-57-62096 (JP, A) JP-A-57-161795 (JP, A) JP-A-60-35799 (JP, A) JP-A-2- 134699 (JP, A) (58) Field surveyed (Int. Cl. ⁷ , DB name) G10L 19/00-19/14 H03M 7/30

Claims (3)

(57) [Claims] An audio signal is transmitted for each analysis frame having a predetermined time length.
Signal's pitch frequency and amplitude value of harmonics at pitch frequency intervals
Audio encoding device that calculates and outputs
The amplitude value of the harmonic at the pitch frequency interval.
After transforming to the number domain, inverse Fourier transform
A harmonic amplitude envelope analyzing means for outputting a Rie transform coefficient,
The low-order inverse filter output from the harmonic amplitude envelope analysis means.
Envelope component quantizer that quantizes and encodes the Rie transform coefficients.
And the lower order quantized by the envelope component quantizing means.
Fourier transform after inverse quantization of inverse Fourier transform coefficients of
To obtain a Fourier transform coefficient, and this Fourier transform coefficient
And the amplitude value of the harmonic component of the pitch frequency interval
A difference quantization means for quantizing and encoding the difference value above
A speech encoding device comprising: 2. Encoding is performed for each analysis frame having a predetermined time length.
Pitch frequency of the encoded audio signal and encoded pitch
And decodes the harmonic components at the
Is based on the amplitude and phase of the decoded harmonic
Speech decoding that synthesizes decoded speech by generating and superimposing
In the encoding device, it is a part of the encoded harmonic component
The low-order inverse Fourier transform coefficients are decoded, and the low-order inverse Fourier transform
An envelope component inverse quantization means for outputting a transform coefficient.
Fourier transform the following inverse Fourier transform coefficients
A harmonic amplitude envelope decoding means for obtaining a transform coefficient;
Inverse quantizer that decodes the difference that is part of the harmonic component
And the decoded difference and the harmonic amplitude envelope decoding
By adding the Fourier transform coefficients obtained by
Adding means for decoding the amplitude value of the harmonic component at the switch frequency interval
A speech decoding device comprising: 3. An audio signal for each analysis frame having a predetermined time length.
Signal's pitch frequency and amplitude value of harmonics at pitch frequency intervals
And an encoding unit that calculates and outputs a phase, and encodes and outputs each.
The encoded pitch frequency and the pitch frequency interval
The harmonic components are decoded, and the cosine wave having the frequency of each harmonic is
Generate and superimpose based on the amplitude and phase of the decoded harmonic
And a decoding unit that synthesizes the decoded speech by
In the audio encoding and decoding apparatus, Convert the amplitude value of harmonics at the above pitch frequency interval to logarithmic domain
After the inverse Fourier transformation And the low-order inverse Fourier transform coefficients
And a harmonic amplitude envelope analyzing means for outputting the harmonic amplitude envelope.
The low-order inverse Fourier transform coefficient output from the analysis means is
Envelope component quantization means for quantizing and encoding, and this envelope
The above low-order inverse Fourier transform quantized by the component quantization means
Fourier transform after inverse quantization of the permutation coefficient
The transform coefficient is obtained, and the Fourier transform coefficient and the pitch
The difference value on each harmonic from the amplitude value of the harmonic component at the wave number interval
The encoding unit is provided with a differential quantization means for quantizing and encoding.
The encoded low-order inverse transmitted from the encoding unit.
Decodes Fourier transform coefficients and generates a low-order inverse Fourier transform
An inverse component quantization means for outputting a number,
Fourier transforms the Fourier transform coefficients, and calculates the Fourier transform coefficients.
The harmonic amplitude envelope decoding means to be sought and transmitted from the encoding unit
Inverse quantization means for decoding the encoded encoded difference
And the decoded difference and the harmonic amplitude envelope decoding
By adding the Fourier transform coefficients obtained in the step,
Adding means for decoding the amplitude value of the harmonic component at the switch frequency interval
Encoding / decoding characterized in that the decoding unit is provided with
apparatus.