JPH04219797A - Time base compressing and elongating method - Google Patents

Abstract

PURPOSE:To obtain combination between frames with less distortion and to enable time base compression and elongation by a frame unit by making a position of a peak between connected frames integral times of a pitch. CONSTITUTION:A compression and elongation adjustment part 1 chooses per frame whether input voice is sent to a buffer 2 for work or a buffer 3 for output or deleted according to a compression rate or an elongation rate which was put in. For example, in the case where it is to be compressed to 2/3, a first frame is sent to the buffer 3 for output, the next frame is deleted and the frame next to it is sent to the buffer 2 for work. The frame next to the deleted one is sent to the buffer 2 for work without fail. And cross-correlation between the buffer 2 for work and the buffer 3 for output is calculated at a cross-correlation calculation part 4, and an interval which becomes the maximum is set as a pitch. Thus, as time base is compressed or elongated with composing the voice which is compressed and encoded, the frames can be connected with each other without dislocation of the pitch.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a time axis compression/expansion method used when reproducing recorded audio, such as English conversation or lectures, at a speed different from that at which it was recorded.

0002

[Prior art] As a technology for converting the time axis of audio, T
There are DHS and PICOLA methods, but these methods perform processing based on the pitch of the audio, so they cannot process natural audio whose pitch constantly changes on a fixed frame basis. There wasn't.

0003

[Problem to be solved by the invention] However, PIC
Processing that requires pitch analysis, such as the OLA method, can produce high-quality early/slow playback sound, but the problem is that the processing for performing pitch analysis is complex and the amount of calculations is extremely large. there were.

[0004] With the development of digital technology, it has become common for each packet to be compressed and encoded in audio recording, playback, transmission, and the like. Therefore, when converting the audio time axis, the audio must be decoded, stored once in a buffer, and then compressed and expanded on the time axis. When trying to perform the above processing with a single CPU, the frame for audio decoding has a fixed length, but the frame for time axis compression/decompression changes in size depending on the pitch, making the processing very complicated. There was a problem with becoming.

[0005] Furthermore, the frames for audio processing are made of a fixed length that is large enough to accommodate two or more maximum pitches, and the time axis is compressed by deleting every few frames. If the time axis is expanded by inserting a frame, there is a problem in that the pitch will shift at the joint between frames.

[0006]

[Means for Solving the Problems] The present invention is intended to solve the above problems, and is to make the frame for audio processing a fixed length large enough to accommodate two or more maximum pitches, and to divide the frame into a number of frames. When compressing the time axis by deleting every other frame and expanding the time axis by inserting the same frame every few frames, the cross-correlation between the two frames to be connected is calculated, and the The interval where the correlation is maximum is defined as the pitch, the position where the amplitude is maximum at the last pitch length of the frame before splicing is found, the position where the amplitude is maximum at the first pitch length of the frame after splicing is found, We propose an audio time axis compression/expansion method characterized by connecting these two positions so that the interval is an integral multiple of the pitch, and shaping the output waveform at the joint of frames.

[0007]

[Operation] With the above configuration, when performing time-axis compression/expansion while decoding compression-encoded audio,
Frames can be connected without pitch deviation.

[0008]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. In FIG. 1, reference numeral 1 denotes a compression/expansion adjustment unit that selects whether to send the input audio frame by frame to the work buffer 2 or the output buffer 3, or to delete it, according to the input compression rate or expansion rate. For example, when compressing by 2/3 times, deletion is performed once every three times. In other words, the first frame is sent to the output buffer 3, the next frame is deleted, and the next frame is sent to the work buffer 2.The frame after deletion is always sent to the work buffer 2.
sent to. Also, when decompressing to 3/2 times, the same frame is repeated every two times, so the first frame is sent to output buffer 3, and the next frame is sent to work buffer 2 and output buffer 3. It will be done. The working buffer 2 is used to store waveforms to be connected to the next frame outputted from the output buffer 3. The output buffer 3 outputs one old frame when two frames are accumulated. 4 is a cross-correlation calculation unit that calculates the cross-correlation between the working buffer 2 and the output buffer 3, and sets the interval at which the correlation is maximum as the pitch. 5 is an output waveform shaping section, and its operation will be explained with reference to FIGS. 2 and 3. In FIG. 2, t1 is the time from the end of frame F1 output from the output buffer 3 to the position where the amplitude is maximum in pitch length, and t2 is the pitch from the beginning of frame F2 output from the work buffer 2. Waveform 1 is a waveform from t2 to the pitch length. Furthermore, a window is created by drawing a straight line from 1 to 0 as shown by the broken line in the t1 interval. In the t2 interval, a window is drawn by drawing a straight line from 0 to 1 as shown by the broken line. Furthermore, for waveform 1 shown in part (A) in the figure, 0 as shown by the broken line
Create a window so that it becomes ~1 and then returns to 0. and,
The waveforms of the upper and lower rows are added together to form a waveform as shown in FIG. In FIG. 3, t1+t2 is longer than the pitch length, and waveform 2 is a waveform corresponding to t2+pitch length from the beginning of frame F2 on the work buffer side, and the window applied to the rear of frame F1 on the output buffer side is t3 ( =t1+t2-
pitch length) section. The window placed ahead of the frame F2 on the work buffer side is placed in the t2 interval. Furthermore, for waveform 2 shown in part (B) in the figure, 0 as shown by the broken line
A window is created so that the value becomes ~1 and then returns to 0, and the waveforms of the upper and lower rows are added together to form a waveform as shown in FIG.

In both cases of FIGS. 2 and 3, after the waveforms are connected and shaped by the output waveform shaping section 5, two frames are extracted from the frame F1 and returned to the output buffer 3. 4. As shown in FIG. 5, when connecting frames, only the part that extends beyond the fixed frame length is discarded.

0010

[Effects of the Invention] In the present invention, by setting the peak position between connected frames to an integral multiple of the pitch, a connection between frames with less distortion can be obtained, and time axis compression/expansion can be performed on a frame-by-frame basis. Beneficial.

[Brief explanation of the drawing]

FIG. 1 is a schematic block diagram for explaining the present invention.

FIG. 2 is a diagram for explaining processing of an output waveform shaping section of the present invention.

FIG. 3 is a diagram for explaining the processing of the output waveform shaping section of the present invention.

FIG. 4 is a diagram for explaining the processing of the output waveform shaping section of the present invention.

FIG. 5 is a diagram for explaining the processing of the output waveform shaping section of the present invention.

[Explanation of symbols]

1 Compression/expansion adjustment section 2 Work buffer 3 Output buffer 4 Cross correlation calculation section 5 Output waveform shaping section

Claims (1)

[Claims] [Claim 1] The frames for which audio processing is to be performed are of a fixed length that is large enough to accommodate two or more maximum pitches, and the time axis is compressed by deleting every few frames, and the same frame is deleted every few frames. When extending the time axis by inserting a frame, calculate the cross-correlation between the two frames to be connected, set the interval with the maximum cross-correlation as the pitch, and set the pitch at the end of the frame before joining. Find the position where the amplitude is maximum at the length, find the position where the amplitude is maximum at the first pitch length of the frames after joining, and connect these two positions so that the interval is an integral multiple of the pitch, and An audio time axis compression/expansion method characterized by shaping the output waveform at the joints of frames.