JPH0378639B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Prior Art) This invention modulates word-by-word original speech using adaptive differential pulse code modulation.
This invention relates to a speech encoding method that uses code modulation (hereinafter abbreviated as ADPCM).

(Background Art) A block diagram of a conventional audio editing system is shown in FIG. In Fig. 1, 1 is an audio input terminal;
2 is an analysis section that analyzes audio and performs ADPCM encoding; 3 is a sub-memory section that stores the encoded data encoded by the analysis section 2; and 4 stores the necessary encoded data from the sub-memory section. 5 is a synthesis unit that converts encoded data into original speech; 6 is a central unit that has functions for controlling, analyzing, and synthesizing the analysis unit 2, submemory unit 3, main memory unit 4, and synthesis unit 5; Processing unit CPU.

In the conventional audio editing method, as shown in the flowchart shown in Figure 2, a word-by-word audio (hereinafter referred to as a audio piece) is input to the input terminal 1 (step 8), analyzed by the analysis unit 2, and encoded into ADPCM ( Step 9), the result is temporarily stored in the sub-memory section 3 as audio encoded data (step 10), and the encoded data of the required word unit (for example, "A") is read out from the sub-memory section 3. (Step 11), stores the data in the main memory 4 (Step 12),
The input speech including the next speech piece (for example, "B") is similarly analyzed and ADPCM encoded, and the encoded data of the necessary speech piece "B" is read out and stored in the main memory 4. Thereafter, the audio pieces necessary for editing are read out and stored in the same manner.

Next, the desired order when editing (e.g. “B”,
The encoded data of each voice piece is read out from the main memory unit 4 in the order of “A” and “C” (step 14),
The voice synthesizer 5 reproduces the original voice (step
15), obtain the desired audio output “BAC” (step
16).

However, as is well known, the ADPCM method
Since the delta parameter Δp changes in the range from the minimum value "0" to the maximum value "MAX (Δp)" depending on the amount of change in the input audio (that is, the difference value), the delta parameter Δp of each voice piece does not end at "0". Sometimes there isn't. For example, when cutting out a speech segment from continuous speech, the end of the speech segment (for example, "B") is
As shown in Figure 2, the amount of change may end up being large. In this case, as mentioned above, the delta parameter Δp(B) at the end of the speech segment is "0" as shown in Figure 3b.
The value "α" is set to a certain value above. However, the next speech piece shown in Figure 4a (e.g. "A")
As shown in Figure 4b, the delta parameter Δp(A)=α has been initialized, analyzed, and encoded at the tip of , so if the speech pieces "B, A" are combined and synthesized as they are, the tip of If Δp(A) is “0”, the original audio will be reproduced, but since it is actually “α”, the waveform will change as shown in Figure 5.
Inconveniences such as (a) and (b) that the zero level shifts occur.

(Purpose of the Invention) The present invention has been made by focusing on the above-mentioned conventional problems. Silence data is forcibly added to the end of each piece of speech that has been cut off (speech that connects words or its encoded data; in the case of word-based speech, it is the silent part between words or its encoded data). The aim is to solve the above problem by subsequently performing ADPCM encoding.

(Structure and operation of the invention) This invention will be explained using the flowchart shown in FIG.

The apparatus for carrying out this invention has a similar structure to the conventional apparatus shown in FIG. First, input continuous audio into input terminal 1 (step 17), and then
is analyzed and ADPCM encoded (step 18),
The result is temporarily stored in the sub memory unit 3 as encoded data.
(step 19), and the central processing unit 6 reproduces the encoded data into PCM encoded data.
Then, the encoded data of the necessary voice piece is cut out from the PCM encoded data, the redundant part of the data is cut and removed (step 20), and silent data (for example, analysis is added to the end of the encoded data of the remaining voice piece). When the internal A/D converter is used in a 12-bit bipolar system, a certain number or more samples of 7FF(H) are added (step 21), analyzed again by the central processing unit, and ADPCM encoded (step 22).
The ADPCM encoded data is stored in the main memory section (step 23). Thereafter, extraction and storage are performed word by word in the same manner.

In Figure 7, silent data 7FF(H) is added to the end of the encoded data of voice piece "B" (step 21).
A waveform 29 of the delta parameter Δp when ADPCM encoding is performed (step 22) is shown. In the same figure,
When the end of the encoded data of speech piece “B”28 is delta parameter Δp(B)=α, the decrease in delta parameter Δp29 is 1 as a feature of the ADPCM method.
Therefore, it takes α samples for the delta parameter Δp to reach the minimum value and initial value “0”. Therefore, if the sampling period is _ts , the length T of the silent data to be added is T=α· _ts , and α pieces of silent data are required. Also, the delta parameter Δp has its maximum value at the end of the voice piece.
MAX(Δp). Therefore, in order to keep the additional silent period T at a constant value, it is necessary to set TMAX(Δp)· _ts .

FIG. 8 shows a synthesized speech waveform 30 and a waveform 31 of delta parameter Δp when speech pieces “B” and “A” are synthesized (steps 25 to 27) with additional silent interval T=MAX(Δp)· _ts . shows. As shown in the figure, the delta parameter Δp can be set to "0" by the start point 32 of the speech segment "A", so the next speech segment "A" is completely reproduced into the original speech waveform.
(a), and no zero shift phenomenon occurs (b).

As a specific example, if MAX (Δp) = 48, sampling period t _s is frequency f = 6KHz, and t _s = 1/6 = 167 (μs), the additional silent interval T is TMAX (Δp)・t _s = 48/6=8 (ms) However, since approximately 8 ms of silence cannot be perceived as silence, even if such processing is performed, unnatural sound will not be reproduced. Further, the length of the redundant section to be cut and removed is usually about 100 to 200 ms, and therefore the number of silent sections to be added is much smaller.

In addition, in the above embodiment, the original audio signal is
The ADPCM encoded data is regenerated into PCM encoded data, and the PCM encoded data is encoded using the ADPCM encoding method.
This is a case where a series of processes such as cutting and removal described above are performed on the end of encoded data, but as a second embodiment, the original audio is suddenly PCM encoded, and the PCM
A series of processes can also be performed on encoded data. Furthermore, although the above first embodiment deals with the case where the original speech is a continuous sound, it can be similarly implemented when the original speech is a speech on a word-by-word basis.

(Effects of the Invention) As explained above, according to the present invention, the structure is converted into ADPCM encoded data using the ADPCM method, and then the ADPCM encoded data is converted into PCM encoded data, or original audio is converted into PCM encoded data using the PCM method. The redundant parts at the end of these PCM encoded data are cut and removed to remain.
Because we decided to add a predetermined number of silence codes to the end of PCM encoded data and encode the PCM code using the ADPCM method, the delta parameter at the end of the encoded data can be smoothly adjusted no matter what point a voice piece is cut out. It can be initialized, defects such as waveform deformation in the audio synthesis section during editing can be removed, and the original audio can be perfectly reproduced.

[Brief explanation of drawings]

FIG. 1 is a block diagram of a conventional audio editing system or for implementing the present invention, FIG. 2 is a flowchart showing conventional audio editing, and FIG. 3 a
4 and b are diagrams showing the waveform at the end of a speech piece cut out from continuous speech and the waveform of delta parameter Δp, respectively, and FIGS. 4a and b show the waveform and delta parameter Δp at the tip of a speech piece cut out from continuous speech, respectively.
Figures 5a and 5b show the waveforms of the third
A diagram showing the synthesized waveform and the waveform of the delta parameter Δp when the speech pieces shown in FIGS.
FIG. 6 is a flowchart illustrating audio editing according to the present invention, and FIGS. 7a and 7b respectively show the audio waveform and the waveform of the delta parameter Δp when silence data is added to the end of a voice piece according to the present invention. FIGS. 8a and 8b are diagrams respectively showing the edited composite waveform and the waveform of the delta parameter Δp when the editing shown in FIG. 6 is performed. DESCRIPTION OF SYMBOLS 1...Audio input terminal, 2...Analysis section, 3...Sub memory section, 4...Main memory section, 5...Synthesizing section, 6...Central processing unit.

Claims (1)

[Claims] 1 In an audio encoding method that encodes the original audio using the applied differential pulse code modulation method (ADPCM method), the original audio is encoded using the ADPCM method.
After converting it into ADPCM encoded data, convert the ADPCM encoded data into PCM encoded data, or convert the original audio into PCM encoded data.
PCM coded data is generated using the PCM method, and the redundant portion at the end of the PCM coded data obtained by any of the above methods is cut off and removed to remain.
Delta parameter Δp at the end of PCM encoded data
The product of the maximum value Max (Δp) and the sample synchronization period ts
Add a silence code with a length determined by Max(Δp)・ts,
A voice encoding method characterized in that the PCM code is encoded using an ADPCM method.