JPH01312600A - Reproducing method for voice waveform by power adaptive window - Google Patents

Abstract

PURPOSE:To obtain a voice waveform being similar to an original sound by using a power adaptive window for varying a shape (weighting of a duplicate part) of its window in accordance with a ratio of voice power in frame sections before and behind the duplicate part at the time of synthesis. CONSTITUTION:With respect to a duplicate part at the time of synthesis, a shape (weighting) of a window of the duplicate part is varied adaptively in accordance with a ratio of voice power in analysis sections before and behind said part. That is, even when a sudden variation of power has been generated between frames by utilizing a variation of the voice power, this power adaptive window reproduces an encoded voice waveform in accordance with its power. In such a way, a transient part of a rise, a fall, etc., of the voice waveform becomes clear, and a waveform being similar to an original sound can be obtained.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to processing of overlapping portions between frames at the time of decoding in an audio encoding system that processes every frame of an audio signal.

``Prior art'' In speech encoding, when analyzing and synthesizing speech signals, the sample values of the input signal waveform are divided into short intervals that can be considered quasi-stationary (usually 20 to 39 m-5ee).
), and the analysis intervals (frames) are partially overlapped to maintain continuity of results. To reduce the influence of frame edges, Hamming windows, Hanning windows, square windows, etc. are used during analysis, and trapezoidal windows are often used during synthesis. However, when processing is performed by converting each frame into the frequency domain, using a trapezoidal window during synthesis increases distortion at transient parts such as the rising and falling parts of the audio waveform at frame boundaries. This may cause deterioration in encoding quality.

"Means for Solving the Problem" In order to solve the above problem (2. In this invention, a window of the superimposed part is set according to the ratio of the audio power of the analysis section before and after the superimposed part during synthesis. By adaptively changing the shape (weighting) of the voice waveform, transient parts such as rises and falls of the audio waveform can be clearly seen.
It becomes possible to improve the N ratio.

When analyzing and synthesizing audio signals, the sample values of the input signal waveform are divided into short intervals that can be considered quasi-stationary (usually 20 to 3 Q msee) and analyzed.
Analyzes are performed with some overlapping analysis intervals to ensure continuity of results. This method is shown in Figure 1 (g). At this time, in order to reduce the influence of the edges of the analysis frame, Figure 1 (;
As shown, a Hamming window, Hanning window, square window, etc. are used as windowed analysis waveforms. When synthesizing the analyzed signal waveforms, the present invention uses a power adaptive window that utilizes the audio power of each of two consecutive frames. The shape of the power adaptive window is the second most common (compared to the commonly used trapezoidal window).
Shown in Figure (al, (b)).

In FIG. 2, the time direction is taken as the X-axis and the amplitude direction is taken as the y-axis, and A, B, and C show the overlapping parts.

Let the x and y coordinates of point D be (0,0)t(091)1(N
, 0), (N, 1), Fig. 2(a) (:.

In the trapezoidal window shown, the overlapping portions between B-0 and A-0 are respectively B-C: y=-LX+i p, -D: y=3
−xN N . On the other hand, in the power adaptive window according to the present invention, the second
As shown in figure (b), in the time direction and amplitude direction,
Each frame period (α power of the audio power in two
Real number) P1a*P2"< D ratio 1: E1: t Determine the E point and 2 points. In other words, set the E point as (N - P2"/(
P,"+P2")s P2a/(P1a+P2
”) ) t F, #l'(N'P2a/CP, "+P
2”) I P,”/(P,” +P2a))
/, B-E, E-C, A-F, and F-0 respectively. However, the audio power P of the frame is defined as follows. Xi is the value of the audio signal in the sampled frame, and L is the number of samples in l frame.

This power adaptive window uses changes in audio power to reproduce the encoded audio waveform according to the power even when a sudden power change occurs between frames.

"Embodiments" Examples of the present invention will be described below with reference to the drawings. Third
The figure shows this invention as a voice coding system (Patent Application No. 63-3019).
This is an example in which 8) is applied.

The input audio signal from the input terminal 1 is subjected to a discrete Fourier transform (DF
T) (= converted into a frequency domain signal, that is, a spectrum.Next &: In the acquisition information extraction section 3, the voice power and the spectrum band (for example, 0~1, 1~2, 2~
After the power per 4 KHz is extracted as auxiliary information, it is quantized, and the auxiliary information is further decoded in advance in the auxiliary information decoding section 4. The converted spectrum is divided into an amplitude component and a phase component by an amplitude/phase component extractor 6, and a peak extractor 7 extracts each peak of the amplitude component having a harmonic structure.

Next (three points including the peak point of the amplitude component extracted by the two-layer wave number information interpolation unit 8 and two points on both sides of it) (determine in detail the actual peak frequency by quadratic interpolation. The frequency position of the peak is quantized by the number of allocated pits by the adaptive information allocation unit 5 according to the power of each band by the adaptive information allocation unit 5 using the previously extracted auxiliary information.The frequency information quantization unit 9 quantizes the peak frequency position. The decoding unit 10 decodes only this frequency information and moves to the decoded frequency position.
Newly determine the amplitude and phase of . An amplitude information interpolation section 11 determines the amplitude corresponding to the frequency fi, and a phase information interpolation section 12 determines the phase corresponding to the frequency fi. Based on the previously determined auxiliary information (2), the adaptive information allocation unit 5 performs adaptive information allocation on each of the determined amplitude and phase parameters, and the amplitude information quantization unit 13 and the phase information quantization unit 14 quantize each.

On the receiving side, using the auxiliary information decoded by the auxiliary information decoding section 17, each decoding section 1
Frequency, amplitude, and phase information are decoded in steps 8 to 20,
Based on this information, the audio waveform reproducing section 21 obtains a reproduced waveform using a power adaptive window.

In general (two-speech coding), the power per frame is transmitted as auxiliary information, so the power adaptive window can be realized without increasing the number of transmission parameters, that is, without increasing the bit rate. can.

Next, in this embodiment, an experimental example comparing the case where a normal trapezoidal window is used and the case where a power adaptive window is used will be described. Sampling frequency is 8KHz, analysis frame length is 32m5
lIe, the length of the overlapping part is l (5 mse, the amount of information (bit rate) when α is set to 1/2 above, and the segmental S/N ratio for (2) is shown in FIG.

In the fourth factor, curve 23 becomes curve 2 when using a trapezoidal window.
4 is a case where the power adaptive window of the present invention is used, and the other conditions are exactly the same. From FIG. 4, it can be seen that the power adaptive window of this invention is superior to the conventional trapezoidal window.

Further, an example of the encoded speech waveform is shown in FIG. 5 together with the original speech waveform. When Figure 5(a) is the original speech waveform, Figure 5(b) is when a trapezoidal window is used, and Figure 5(C) is when a power adaptive window is applied, the encoding conditions are the same as above. The bit rate is F3 kbps. The example shown in Fig. 5 is an example in which the part where the audio waveform changes greatly is the frame boundary (two corners). As shown in Fig. 5, a large improvement effect has been obtained by the power adaptive window of the present invention. I understand that.

Here, as an embodiment of the present invention, an example in which it is applied to an audio encoding method (Japanese Patent Application No. 63-30198) is described. be able to.

"Effects of the Invention" As described above, according to the present invention, when audio analysis and synthesis processing is performed frame by frame, audio power information can be effectively used even when there is a steep change in the audio waveform. (Secondly, it is possible to obtain an audio waveform that is closer to the original sound.

[Brief explanation of the drawing]

Fig. 1 is an explanatory diagram of analysis using a Hamming window, Fig. 2 shows a synthesis window, (a) is a trapezoidal window, (bl is a power adaptive window of this invention, and Fig. 3 is an embodiment of this invention. Block diagram shown,
Figure 4 is a diagram showing the relationship between the amount of information -8N ratio when using a power adaptive window and when using a trapezoidal window, and Figure 5 is a diagram showing the effect of the power adaptive window. (a) is the original voice Waveforms, (b) are encoded speech waveforms when using a trapezoidal window, and (C) are encoded speech waveforms when using a power adaptive window. Patent Applicant Nippon Telegraph and Telephone Corporation Agent Takuhan Kusano 1 Yusuke Kuni 11 words of prayer, +172 g (a) (b) O 3 Figures 4 Word of mouth 5 Figure 2msec

Claims (1)

[Claims] (1) Audio in which the input signal is sampled at fixed time intervals, the sample values are taken out, a fixed number of them are stored as one frame, and the analysis and synthesis processing for each frame is performed by partially overlapping the analysis frame section. A power adaptive window in the encoding system, characterized in that a power adaptive window is used that changes the shape of the window (weighting of the overlapping part) according to the ratio of audio power of frame sections before and after the overlapping part during synthesis. Audio waveform playback method.