JP3305338B2 - Pitch frequency codec - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice pitch frequency code decoder in a voice codec used for digitally transmitting or storing a voice signal, and more particularly to an improvement in a coding unit. is there.

[0002]

2. Description of the Related Art When encoding a pitch frequency obtained by analyzing an input speech signal for each analysis frame of a fixed time with a small amount of information, a segment (a group of a plurality of continuous frames) is subjected to phonological continuity. And a method of approximating a plurality of pitch frequencies in the segment by a function such as a segment length (the number of frames in the segment) is known. For example, the document “A SEGMENT VOCODER AT 150 B / S”
S.Roucos, R.Schwarts, J.Makhoul Proc.ICASSP-83,
pp. 61-64, (1983).

FIG. 3 is a block diagram showing a configuration of a conventional pitch frequency encoding / decoding device based on the above-mentioned document. In FIG. 3, reference numeral 51 denotes an encoding unit that encodes a pitch frequency vector obtained by analyzing a plurality of pitch frequencies obtained by analyzing an input audio signal at regular intervals as one unit and outputs an encoding result, and 52 denotes an encoding unit. The decoding unit decodes the pitch frequency from the encoding result output from the unit 51. The encoding unit 51 includes the pitch frequency storage unit 2, the adaptive quantization unit 26, and the difference inverse quantization unit 28. The decoding unit 52 has the adaptive inverse quantization means 30. The details of each of the above means will be described in the following operation description.

Next, the operation will be described. The pitch frequency accumulating means 2 receives the segment length 25 (segment length is L) of the segment determined by the continuity of phonemes, accumulates L pitch frequencies, and is composed of the L pitch frequencies. The pitch frequency vector 5 is output. As shown in FIG. 4, the adaptive quantization means 26 converts the input pitch frequency vector 5 into the pitch frequency 29 of the last frame of the segment one before the current segment.
(Hereinafter referred to as the previous pitch frequency) and the pitch frequency on the final frame at which the error is minimized when approximated by a straight line connecting the pitch frequency assumed on the final frame of the current segment. The difference of 29 is quantized, and this is output as a difference quantization result 27. The difference inverse quantization means 28 obtains a difference from the difference quantization result 27, and adds the difference and the previous pitch frequency held in the difference inverse quantization means 28 to obtain the pitch of the L-th frame after the inverse quantization. A frequency is obtained, and this is output as a new front pitch frequency 29, and this value is held.

[0005] The adaptive inverse quantization means 30 obtains a difference from the difference quantization result 27 output from the encoding section 51, and calculates the difference and the segment length 25 output from the encoding section 51.
(Length L) and L from the previously held pre-pitch frequency
The pitch frequencies are obtained and output as an output pitch frequency sequence 24, and the pitch frequency of the last frame is held as a new previous pitch frequency.

[0006]

As described above, in the conventional pitch frequency encoding / decoding apparatus, when encoding a pitch frequency, a plurality of pitch frequencies in a segment determined by the continuity of phonemes are represented by straight lines. Due to the approximation, the decoded pitch frequency sequence cannot follow the complicated time change of the pitch frequency of the input voice that appears independently of the phoneme change, and the intonation of the input voice is greatly impaired. There was a point.

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and is directed to a pitch frequency system for an input voice signal.
By vector quantizing the columns, the pitch frequency sequence
And to provide a pitch frequency coding decoder capable and rows of TURMERIC at very low bit rate coding of the pitch frequency stored complex time variation of.

[0008]

A pitch frequency coding / decoding device according to the present invention accumulates a codebook 8 composed of finite M pitch frequency vector codewords and a pitch frequency sequence of an input speech signal. Pitch frequency storage means 2,
Pitch frequency vector constructing means 4 for constructing a pitch frequency vector with reference to the minimum point of the pitch frequency sequence accumulated in the pitch frequency accumulating means 2, a pitch frequency vector constituted by the pitch frequency vector means 4, Distance calculating means 6 for calculating a distance from the pitch frequency vector codeword in the book 8, and the distance calculated by the distance calculating means 6 is used to optimize the M pitch frequency vector codewords in the codebook 8. der one provided the best codeword selecting means 11 outputs the optimal pitch frequency vector code word select pitch frequency vector code word as coding result of the pitch frequency vector to the encoding unit 41
You. Further, the pitch frequency from the beginning of the pitch frequency of the stored pitch frequency sequence to the pitch frequency accumulation hand stage 2 to the first point characteristic in the time variation of the pitch frequency as the initial candidate pitch frequency vector, the pitch frequency A function of searching for and configuring a pitch frequency vector having a maximum pitch frequency vector length within a condition that distortion between the vector and the optimal pitch frequency vector codeword output from the optimal codeword selecting means 11 does not exceed a predetermined value. The pitch frequency vector constructing means 4 having the above.

[0009]

[Action] pitch frequency vector forming means 4, the pitch frequency from e.g. <br/> beginning of the pitch frequency of the pitch frequency sequences stored in the pitch frequency storage unit 2 to the first point characteristic in the time variation of the pitch frequency Is used as an initial candidate of the pitch frequency vector, and the pitch frequency vector length is maximized within a condition that the distortion between the pitch frequency vector and the optimal pitch frequency vector codeword output from the optimal codeword selecting means 11 does not exceed a predetermined value. A pitch frequency vector is searched for and configured. The distance calculating means 6 calculates the distance between the pitch frequency vector formed by the pitch frequency vector forming means 4 and the pitch frequency vector codeword in the codebook 8. The optimal codeword selecting means 11 selects an optimal pitch frequency vector codeword from the M pitch frequency vector codewords in the codebook 8 using the distance calculated by the distance calculating means 6, and selects the optimal pitch frequency vector code. The word is output as an encoding result of the pitch frequency vector.

[0010]

FIG. 1 is a block diagram showing a configuration of a pitch frequency coder / decoder according to an embodiment of the present invention. FIG.
, 41 is an encoding unit that encodes a pitch frequency vector obtained by analyzing a plurality of pitch frequencies obtained by analyzing the input audio signal at regular intervals as one unit and outputs an encoding result, and 42 denotes an encoding unit 41 The decoding unit decodes the pitch frequency from the encoding result output from the decoding unit. In the encoding unit 41, reference numeral 8 denotes a codebook composed of finite M pitch frequency vector codewords, reference numeral 2 denotes a pitch frequency storage unit for storing a pitch frequency sequence of an input speech signal, and reference numeral 4 denotes a pitch frequency storage unit Pitch frequency vector constructing means 6 for constructing a pitch frequency vector from the pitch frequency sequence obtained, calculates the distance between the pitch frequency vector constructed by the pitch frequency vector constructing means 4 and the pitch frequency vector codeword in the codebook 8. Distance calculation means,
Numeral 11 selects the optimum pitch frequency vector codeword from the M pitch frequency vector codewords in the codebook 8 using the distance calculated by the distance calculation means 6, and converts the optimum pitch frequency vector codeword into a code of the pitch frequency vector. The optimum codeword selecting means 17 for outputting as a result of quantization is an average pitch frequency quantizing means for quantizing the average pitch frequency from the pitch frequency vector constructing means 4. In the decoding unit 42, reference numeral 21 denotes a codebook composed of finite M pitch frequency vector codewords, 11 denotes inverse quantization means for inversely quantizing the output of the optimal codeword selection means 11, and 22 denotes the average pitch frequency quantization code. This is an average pitch frequency inverse quantization means for inversely quantizing the output of the quantization means 17.

Next, the operation of this embodiment will be described.
The pitch frequency storage means 2 has finite K pitch frequencies 1
Is entered. A pitch frequency sequence can be generally represented by a chain of a cross shape having a peak at the accent position of a word. By extracting a characteristic shape of such a pitch frequency sequence and performing vector quantization on the extracted shape, the quantization efficiency can be increased. For this purpose, the pitch frequency vector constructing means 4 reads the pitch frequency sequence 3 from the pitch frequency accumulating means 2 and performs segmentation at the minimum point of the pitch frequency sequence to form an N-dimensional pitch frequency vector. Output as FIG. 2 is an explanatory diagram for explaining the operation of the pitch frequency vector constructing means 4. In order to segment the K pitch frequencies accumulated in the pitch frequency accumulating means 2, as shown in FIG. , The first minimum point N of the pitch frequency sequence is searched, and from the first frame to the Nth frame of this minimum point N is defined as one segment, and N pitch frequencies normalized by the average pitch frequency in the segment , And output them as an average pitch frequency 16 and a pitch frequency vector 5. Next, the distance calculation means 6 linearly compresses and expands the pitch frequency vector 5 so that the dimension number of the pitch frequency vector 5 matches the dimension number of the code word, and obtains a pitch frequency vector in which the dimension number is normalized. Next, the distance calculating means 6 sequentially reads out the pitch frequency vector codewords from the codebook 8 composed of finite M pitch frequency vector codewords via the changeover switch 7 and normalizes the pitch frequency vector codewords and the number of dimensions. The distance (for example, the Euclidean distance) between the pitch frequency vectors calculated is calculated, and the distance 9 and the corresponding pitch frequency vector codeword number 10 are output. Optimal codeword selection means 1
In step 1, the minimum distance among the distances 9 is obtained, and the minimum distance is determined as the minimum distance 12. The pitch frequency vector constructing means 4 together with the vector dimension length 13 of the pitch frequency vector 5 at this time is used.
Output to Further, the pitch frequency vector codeword number 10 and the vector dimension length 13 at this time are held.

Next, when the minimum distance 12 is smaller than a predetermined threshold value, the pitch frequency vector constructing means 4 sets the start of the section from the Nth frame to the last Kth frame as shown in FIG. The minimum point N ′ of
After constructing a pitch frequency vector with the pitch frequency from the frame to the N'th frame and normalizing this with the average pitch frequency from the first frame to the N'th frame,
Output as pitch frequency vector 5. Next, the minimum distance 12 in the pitch frequency vector of the newly configured vector dimension length 13 is obtained through the distance calculation means 6, codebook 8, and optimum codeword selection means 11. If the minimum distance 12 is smaller than the threshold value, the optimal codeword selecting means 11 updates the values of the pitch frequency vector codeword number 10 and the vector dimension length 13 held by the optimal codeword selecting means 11, and the pitch frequency vector constructing means 4 updates the values after N '. The operation of searching for the minimum point in the frame and forming a new pitch frequency vector is repeated.

If the minimum distance 12 determined by the optimal codeword selecting means 11 is larger than a predetermined threshold value, or if the minimum point N 'is not detected by the pitch frequency vector constructing means 4, the optimal codeword selecting means 11 The pitch frequency vector code word number held in the optimum code word selecting means 11 is set as the optimum code word number 14, and is output to the outside together with the vector dimension length 15 at that time. The average pitch frequency quantization means 17 linearly quantizes the average pitch frequency 16, for example, and outputs an average pitch frequency codeword 18.

The average pitch frequency inverse quantization means 22 determines the average pitch frequency 23 from the average pitch frequency codeword 18. In the inverse quantization means 19, the pitch frequency vector code word specified by the optimum code word number 14 is read out from the codebook 21 via the switch 20, and the sample value of each dimension of the code word is converted into the read pitch frequency vector. Is multiplied by the average pitch frequency 23, and is linearly compressed and expanded so as to match the vector dimension book 15, and is output as an output pitch frequency sequence 24.

In the embodiment, the pitch frequency vector is formed by using the minimum point of the pitch frequency as the segment boundary. However, the maximum point of the pitch frequency, the minimum point of the change rate, the maximum point, a combination of these points, By increasing or decreasing one frame at a time, it is possible to determine an optimal segment on the basis that the segment length is longer and the quantization distortion for the input pitch frequency sequence is smaller.

[0016]

As described above, according to the present invention, the pitch circumference
Based on the minimum point of the pitch frequency series by the wave number configuration means
, The pitch frequency vector is generated, and the vector quantization is performed, so that the pitch frequency encoding that preserves the time change of the pitch frequency can be performed at an extremely low bit rate, and thereby, the pitch of the input voice is the most. It is possible to follow a complicated time change of the frequency, so that the intonation of the input voice is prevented from being impaired and the performance is improved. Further, according to the invention of claim 2,
If the coding distortion does not exceed the predetermined value,
Can be reduced to a minimum.
You.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a pitch frequency encoding / decoding device according to an embodiment of the present invention.

FIG. 2 is a diagram for explaining an operation of a pitch frequency vector forming unit in the embodiment.

FIG. 3 is a block diagram showing a configuration of a conventional pitch frequency encoding / decoding device.

FIG. 4 is a diagram for explaining the operation of an adaptive quantization means in this conventional example.

[Explanation of symbols]

 2 Pitch frequency storage means 4 Pitch frequency vector construction means 6 Distance calculation means 8 Codebook 11 Optimal codeword selection means 17 Average pitch frequency quantization means 41 Encoding unit 42 Decoding unit

Claims (2)

(57) [Claims] An encoding unit that encodes a pitch frequency vector obtained by analyzing a plurality of pitch frequencies obtained by analyzing an input audio signal at regular intervals as one unit, and outputs the encoding result; In a pitch frequency coding decoder comprising a decoding unit for decoding a pitch frequency from the coding result output from the coding unit, a code book comprising finite M pitch frequency vector codewords is inputted. Pitch frequency accumulating means for accumulating a pitch frequency sequence of the voice signal, pitch frequency vector constructing means for constructing a pitch frequency vector based on a minimum point of the pitch frequency sequence accumulated in the pitch frequency accumulating means, Calculate the distance between the pitch frequency vector configured by the frequency vector configuration means and the pitch frequency vector codeword in the codebook Using the distance calculated by the distance calculating means, and selecting the optimum pitch frequency vector codeword from the M pitch frequency vector codewords in the codebook, and selecting the optimum pitch frequency vector codeword by pitch. pitch circumference, characterized in that a and optimal codeword selecting means for outputting as a coded result of the frequency vectors
Wave number coder / decoder. 2. A pitch frequency vector constructing means comprising:
The first of the pitch frequency series stored in the frequency storage means
From pitch frequency to time variation of pitch frequency
The pitch frequency up to the characteristic first point
The pitch frequency vector and the maximum
Optimal pitch frequency vector output from suitable codeword selection means
Within a condition that the distortion between the
Pitch frequency vector that maximizes the maximum frequency vector length
2. The pitch circumference according to claim 1, wherein
Wave number coder / decoder.