JP2853266B2 - Audio encoding device and audio decoding device - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech encoding device and a speech decoding device for encoding or decoding speech.

2. Description of the Related Art Conventionally, to realize sound encoding at a low bit rate (about 4.8 kbps), frequency characteristics of speech are extracted by using spectral analysis such as linear prediction analysis, and encoded together with sound source information. There is a speech analysis / synthesis coding method that performs speech analysis, and a speech waveform coding method that encodes a waveform itself using speech redundancy.

SUMMARY OF THE INVENTION However, of the above-mentioned conventional examples, the former speech analysis / synthesis coding method can realize a low bit rate, but can encode a driving sound source for synthesizing high quality speech. Is difficult. On the other hand, the latter speech waveform encoding method can decode high quality speech, but it is difficult to realize a low bit rate.

As described above, the conventional audio encoding / decoding system cannot simultaneously satisfy high-quality sound quality and low-bit-rate encoding. Further, a problem common to both of them is that the amount of calculation increases due to complicated processing.

The present invention has been made in view of the above problems, and has as its object to realize low-bit-rate voice encoding by simple data processing while taking a form of voice waveform coding.

Means for Solving the Problems To achieve this object, a first invention of the present invention is
The shape of the one-pitch basic waveform obtained by analyzing the pitch of the audio signal is obtained, a point at which the absolute value of each of the positive and negative is maximized from the basic waveform is searched, and a vertical line lowered from the point to the time axis is used as a skeleton. The number of frames determining means for determining the optimal number of sets of the plurality of types of frames according to the waveform of the audio signal, and the number of frames determined by the number of frames determining means, the time position and amplitude of the point Skeleton encoding means for encoding using the skeleton encoding means, a first interskeletal waveform codebook storing a plurality of numbered interskeletal waveform samples, and a skeleton obtained by the skeleton encoding means. And an interosseous waveform encoding means for encoding the interosseous waveform to be obtained using the first interosseous waveform codebook.

According to a second aspect of the present invention, a vertical line lowered on a time axis from a point determined by information encoded using a temporal position and an amplitude regarding an audio signal is set as a skeleton, and extracted from the information regarding the audio signal. A skeleton decoding means for creating a plurality of types of skeletons by using the number of skeletons; a second interbone waveform codebook storing a plurality of numbered interbone waveform samples; And an interosseous waveform combining means for decoding an interosseous waveform stretched between the frames using the interosseous waveform codebook.

According to a third aspect of the present invention, the first interstitial waveform codebook normalizes temporally and power each of a plurality of interosseous waveforms obtained in advance by voice signal analysis with fixed end points. Stores interstitial waveform samples with waveform identification numbers,
An interstitial waveform encoding unit configured to perform the encoding by comparing the interosseous waveform of the audio signal with the interosseous waveform sample and selecting an identification number of the closest interosseous waveform sample; A second interstitial waveform codebook, wherein a plurality of interosseous waveforms obtained in advance by voice signal analysis are fixed at end points to temporally and power normalized. The interosseous waveform sample in which the identification number is added to the obtained waveform is stored, and the interosseous waveform decoding means extracts the identification number from the obtained information and selects the interosseous waveform sample corresponding to the identification number. Thus, the present invention provides an audio decoding device configured to perform decoding.

Operation According to the present invention, pitch information is obtained by first performing pitch analysis on a voice, assuming that the voice has a waveform having a fundamental frequency. Next, a basic waveform of one pitch is obtained based on the pitch information, and a plurality of types of frames representing the shape of the basic waveform are searched to obtain frame information. Further, information of a waveform (inter-bone waveform) stretched between the frames is compressed and encoded. Then, by transmitting the skeleton information and the interosseous waveform information together, voice encoding and decoding at a low bit rate can be realized by simple data processing.

Embodiment FIG. 1 is a functional block diagram showing one embodiment of a speech encoding device and a speech decoding device according to the present invention.

First, a description will be given below of each block constituting the speech encoding device and the speech decoding device.

The encoder 1 is supplied with an input audio signal 3 which is sampled, converted into a digital signal, and divided for each predetermined time length (one frame).

The pitch in the section of the input speech signal 3 is obtained by the pitch analysis unit 4 of the encoder 1, and this is used as pitch information. Based on the pitch information, the pitch analysis unit 4 calculates an average waveform of one pitch from the waveform in the section, and sends the average waveform to the skeleton search unit 5 as a basic waveform.

The skeleton search unit 5 analyzes the shape of the basic waveform created by the pitch analysis unit 4 and takes the maximum number of positive and negative absolute values in accordance with the number of steps of the skeleton, while considering the number of stages of the skeleton. A point is searched, and the position of the signal and the amplitude of the signal are used as skeleton information.

Here, the skeleton search method in the skeleton search unit 5 will be described in detail.

The one-pitch basic waveform has an impulse response shape, but the shape varies depending on the voice and the occurrence situation. Therefore, in order to represent the general shape by a frame, it is necessary to determine the number of steps according to the shape of the waveform. That is, it is necessary to set a small number of steps for a waveform having a gentle mountain shape, and to set a large number of steps for a waveform which vibrates strongly in positive and negative directions.
Therefore, an algorithm for performing a frame search in consideration of the number of frame steps will be described below.

(1) Set the initial value.

Xi (i = 1, L): One pitch basic waveform. L is the length.

D: Maximum number of frame steps.

K: A search prohibited area set having elements from positions 1 to L as elements.
Let K = φ (empty set) as an initial value.

M: Number of search stages. Initial value M = 0 Hi = (Ax, An, Ix, In): Frame information. MAX signal values Ax, M
It is composed of four values of IN signal value An, MAX position Ix, and MIN position In.

(2) M = M + 1 (3) Xmax = max {Xi | i = 1, LiK} = Xi1 Xmin = min {Xi | i = 1, LiK} = Xi2 HM = (Xmax, Xmin, i1, i2) (4) With respect to i1 and i2, all the positions of the preceding and succeeding sections in which the sign of Xi does not change are added to the K element as a prohibited area.

(5) If M = D or K has all 1 to L as elements, go to (6). Otherwise go to (2).

(6) Only the position information portion of Hj (j = 1, M) is extracted and arranged in the order of size.

(7) From the smaller one, the position is the MAX position.
Check if the position is IN. And either one is 2
If it continues one after another, set M = M−1 and go to (6).
If MAX and min are all alternately arranged, go to (8).

(8) The search is terminated using M as the frame number and Hj (j = 1, M) as the frame information.

FIG. 2 shows an example of a set of basic waveforms classified by the above-described algorithm. In FIG. 2, the solid line indicates the basic waveform of one pitch, and the broken line indicates the position of the skeleton. FIG. 2 (a) shows the case where the skeleton has one stage, and FIG. 2 (b) shows the case where the skeleton has two stages.
FIG. 2C shows an example of the case of three stages, and FIG. 3 shows the relationship between the basic waveform and the skeleton information. In FIG. 3, A1
1, A12, A21, A22 are skeleton position information, B11, B12, B21, B
22 is signal value information.

Next, the function of the interosseous waveform selector 6 will be described with reference to FIG. However, FIG. 4 shows a case where the frame is one stage.

First, based on the skeleton information supplied from the skeleton search unit 5,
Within 1 pitch, MIN signal from MAX signal 1 which is the skeleton
The waveform stretched between C2 and MIN signal C2 to MAX signal C1
(See FIG. 4 (a)), and these are defined as basic interosseous waveforms D1 and D2 (see FIG. 4 (b)). Next, the respective basic interosseous waveforms d1 and
Waveforms E1 and E2 normalized with respect to time and power are obtained with d2 fixed at the end points. (See FIG. 4 (c)).

The waveforms E1 and E2 are compared with the numbered interosseous waveform samples stored in the interosseous waveform codebook 7 (see FIG. 1) as shown in FIG. The numbers N and M attached to the waveform samples are output as inter-bone waveform information. The pitch information (output of the pitch analysis unit 4), skeleton information (output of the skeleton search unit 5), and interskeletal waveform information (output of the interskeletal waveform selection unit 6) obtained as described above are used It is transmitted to the decoder 2 as a code.

The interosseous waveform codebook 7 used at this time collects basic interosseous waveforms obtained by analyzing the speech in advance for a large amount of speech data, and fixes the end points to normalize in terms of time and power to obtain a number. Stored information with.

Here, a method of creating information stored in the interbone waveform codebook 7 will be described in detail.

It is obvious that the larger the size of the interstitial waveform codebook 7, the smaller the coding distortion. In order to achieve high sound quality, it is desirable that the size of the interstitial waveform codebook 7 is large. However, in order to realize a low bit rate, it is desirable that the number of bits of the interosseous waveform information is small,
In order to operate the encoder 1 in real time, it is desirable that the amount of calculation required for matching with the interbone waveform codebook 7 is small. Therefore, an efficient interskeletal waveform codebook 7 having a small size but small coding distortion is required.

In order to create the interstitial waveform codebook 7, clustering is performed on a sufficiently large set of interosseous waveform samples so that the Euclidean distance between the sample and the centroid (center of gravity) is minimized, and the interstitial waveform to be created is calculated. Waveform codebook 7
A technical means is used in which the interstitial waveform codebook 7 is created with the centroids (centroids) of the clusters divided into classes of the number of sizes. Clustering used in this embodiment
The algorithm is a cell division type algorithm. The algorithm is described below.

(1) K = 1 (2) A centroid of K clusters is obtained by a simple average. Then, calculate the Euclidean distance between all samples belonging to each cluster and the centroid,
The maximum value is defined as the distortion of the cluster.

(3) Two centroids are formed near the centroid of the cluster having the largest distortion among the K clusters.
(It becomes the nucleus of cell division.) (4) Clustering is performed based on K + 1 centroids, and the centroid is obtained again.

(5) If there is an empty cluster, delete the centroid and go to (3).

(6) The distortion of K + 1 clusters is obtained in the same manner as in (2). If the amount of change in the sum is equal to or smaller than a preset small threshold, go to (7), and if larger than the threshold, go to (4).

(7) If K + 1 does not reach the target number of clusters, set K = K + 1 and go to (2), otherwise go to (6).

(8) The centroids of all the clusters are obtained, and the interbone waveform codebook 7 is created.

Next, the function of the decoder 2 will be described with reference to FIGS. However, FIG. 5 shows a case where the frame has one stage.

First, in the skeleton forming unit 8, based on pitch information (output of the input audio signal 3) obtained by encoding by the encoder 1 and skeleton information (output of the interosseous waveform selecting unit 6), the audio skeletons C1 and C1 are output. Form C2. FIG. 5A is an example of this skeleton. It shows a state where a skeleton is formed based on skeleton information.

In the waveform synthesizing unit 9, based on the interosseous waveform information N and M supplied from the interosseous waveform selecting unit 6, the same interosseous waveform codebook 10 as that stored in the encoder 1 is used. E1 and E2 are selected, converted in terms of time and power according to the frame, and stretched between the bones, and this waveform F is used as an output audio signal 11 (see FIG. 1). FIG. 5 (b) to (d)
Is an example of this waveform synthesis. A state in which a basic interosseous waveform is stretched between frames by an interosseous waveform sample selected from the interosseous waveform codebook 10 based on the interosseous waveform information N and M is shown.

In order to show the effect of this speech coding method, a simulation experiment of this speech coding / decoding was performed. Hereinafter, the results of this experiment will be described.

The encoded voice data is the voice of the weather forecast uttered by one female announcer, "Weather forecast. The weather forecast announced at 1:30 pm, Japan Meteorological Agency Forecast Department. On the south coast of Japan, the east-west front Is stagnant, there is a low pressure near the east of Hachijojima on the front line and near the Goto Islands in Kitakyushu, and it is proceeding east-northeast. " 20 seconds. The audio data is analyzed every 20 msec (one frame). The interosseous waveform codebook 7 is created using the above clustering algorithm based on the interosseous waveform sample set obtained by analyzing the audio data of each of the 50 men and women without the audio data for about 10 seconds each. did. The number of samples is about 20,000.

In addition, the maximum number of frame stages was set to three. Then, in order to further reduce the bit rate, bit allocation was adaptively performed according to the number of stages. 2nd and 3rd skeleton position information and 3rd
The frame skeleton gain information of the stage is encoded by vector quantization to save the bit rate. The size of the interstitial waveform codebook 7 for obtaining the interosseous waveform information is:
The waveform is adaptively changed according to the number of stages and the length of the waveform, and the short waveform is encoded by the small interosseous waveform codebook 7 and the long waveform is encoded by the large interosseous waveform codebook 7. The bit allocation per unit of audio data (20 msec) is shown in Table 1 below.

As a result of a coding experiment under the above conditions, a smooth and natural speech could be synthesized at a low bit rate. Approximately 10 dB was also obtained in the S / N ratio. When a similar experiment was attempted with voices other than the voice data, an S / N ratio of 7 to 11 dB was obtained, and the sound quality was good.

The above-described speech coding experiment verified that the speech coding method according to the present invention was able to realize low bit rate speech coding with simple data processing.

As described above, according to the present invention, several types of skeletons are searched for by analyzing voice signals to obtain skeleton information, and information on waveforms (inter-bone waveforms) stretched between the skeletons is compressed. Inter-bone waveform information, and encodes and decodes the audio based on the skeleton information and the inter-bone waveform information. Therefore, it is possible to obtain a low-bit-rate and high-quality audio encoding device and an audio decoding device. Can be.

[Brief description of the drawings]

FIG. 1 is a functional block diagram showing one embodiment of a speech encoding device and a speech decoding device according to the present invention, and FIG. 2 is a skeleton of one embodiment of a speech encoding device and a speech decoding device according to the present invention. FIG. 3 is a waveform diagram showing a set of basic waveforms classified according to the number of skeleton stages by a search algorithm. FIG. 3 is a diagram showing the operation of an embodiment of a speech encoding device and a speech decoding device according to the present invention in the case where the skeleton has two stages. As an example, a waveform diagram showing the relationship between the basic waveform and the skeleton information, FIG. 4 is a waveform diagram for explaining the operation of one embodiment of the speech encoding device and the speech decoding device according to the present invention for the encoder 1, FIG. Figure 5
FIG. 5 is a waveform diagram illustrating the operation of the speech encoding device and the speech decoding device according to the embodiment of the present invention with respect to the decoder 2. 1 encoder 2 decoder 3 input audio signal 4
... Pitch analysis unit, 5... Skeleton search unit, 6... Interskeletal waveform selection unit, 7.
... Waveform synthesizing section, 10... Interosseous waveform codebook, 11... Output audio signal.

Claims (4)

(57) [Claims] A pitch analysis unit (4), a skeleton search unit (5),
A speech encoding apparatus comprising a first interbone waveform codebook (7) and an interbone waveform selection unit (6), and outputs pitch information, skeleton information, and an identification number for each frame, wherein a pitch analysis unit (4) ) Analyzes the pitch of the input frame audio signal, extracts the pitch information of the frame and the basic waveform for one pitch, and the frame search unit (5) determines the maximum absolute value of the basic waveform based on a predetermined rule. And outputs amplitude information and position information of the maximum absolute value point as skeleton information. The first interosseous waveform codebook (7) has a plurality of interosseous waveform samples corresponding to the identification numbers. The interosseous waveform selecting unit (6) selects and selects an interosseous waveform sample matching the waveform between the absolute value maximum points obtained by the skeleton search unit (5) from the first interosseous waveform codebook (7). Speech coder that outputs identification number of interstitial waveform sample . 2. The interosseous waveform sample of the first interosseous waveform codebook (7) is a waveform normalized in terms of time and power. The speech encoding apparatus according to claim 1, wherein a bone-to-bone waveform sample to be matched is selected after performing temporal and power conversion on the speech signal. 3. A frame forming unit (8), a second interbone waveform codebook (10), and a waveform synthesizing unit (9). For each frame, pitch information, position information and amplitude information of the absolute value maximum point, An audio decoding device for inputting an identification number and decoding an audio signal, wherein a frame forming unit (8) calculates an absolute value maximum point based on pitch information, position information of the absolute value maximum point, and amplitude information. The second interosseous waveform codebook (10) has a plurality of interosseous waveform samples corresponding to the identification numbers, and the waveform synthesizing unit (9) compares the interosseous waveform samples corresponding to the identification numbers between the absolute value maximum points. To form a one-pitch basic waveform, and restore a speech waveform between frames based on the basic waveform. 4. The interosseous waveform sample of the second interosseous waveform codebook (10) is a waveform normalized in terms of time and power, and the waveform synthesizing unit (9) outputs a waveform corresponding to the identification number. The audio decoding apparatus according to claim 3, wherein the inter-waveform sample is temporally and power-converted, and is then set between the absolute value maximum points.