JPH0219899A - Voice accumulating and reproducing device - Google Patents

Abstract

PURPOSE:To input voices in frames and perform waveform thinning processes at every pitch in real time by accumulating the voices after encoding and reading out and decoding the accumulated encoded voices, and then, thinning waveforms at every pitch of the voices. CONSTITUTION:A power process circuit PER reads out the voice waveform of a frame waveform memory M3 and calculates the power of a frame and outputs a voice frame in the form of a flag (VUF) by discriminating that the frame to be outputted is the voice frame when the power exceeds a preset threshold. When the power does not reach the threshold, the circuit PWR outputs a soundless or voiceless frame in the form of the flag (VUH) by discriminating that the frame to be outputted is the soundless or voiceless frame. A thinning process circuit TDS thins the waveforms at every pitch NP by reading out voice waveforms from memories M4 and M3 and restores the remaining reside waveform in the memory M4. Thus the voices are accumulated after encoding and the accumulated encoded voices are read out, decoded, and outputted through a D/A converter circuit 5 after thinning waveforms at every pitch of the voices. Therefore, voices can be inputted in frames and waveform thinning can be performed at every pitch in real time.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a voice storage and playback device for voice memos such as comments in document retrieval systems, voice mails in communication systems, etc., and in particular, the present invention relates to a voice storage and playback device for voice memos such as comments in document retrieval systems and voice mails in communication systems. This invention relates to an easy-to-use audio storage and playback device that can be played back at high speed (quickly speaking) without changing anything.

[Conventional technology]

Conventionally, devices or methods for reproducing stored audio at high speed include (1) a device that converts the time axis based on the ratio of the sampling clock to the output clock, as disclosed in Japanese Patent Laid-Open No. 57-85099 (2) As disclosed in Japanese Patent Application Laid-Open No. 59-75295, a device for changing the playback speed of audio by adjusting the length of the pause section (3) IE, Transaction on Acoustics, Speech & Signal Processing, Anis Sp-27 (1979) No. 12
pp. 1-133 (IEE Trans.

Acoustics, 5peech and Sign
al Processing, ASSP-2
7 (1970), pp. 121-133), a method is known in which the waveform is thinned out and output in pitch units of speech.

[Problem to be solved by the invention]

Each of the above conventional techniques has the following problems. In other words, point (1) is that no consideration is given to the sound quality, and the voice changes to a high-pitched voice, as if playing a tape recorder at high speed, and the characteristics of the speaker are lost. By removing the pauses, the content of the utterance becomes unnatural and the meaning becomes unclear, and the high-speed performance cannot be achieved. (3) It is possible to play back at high speed without changing the sound quality. Therefore, it is a suitable method for audio storage and playback devices such as voice memos, but it is still in the experimental stage and no consideration has been given to making it into a device, so it cannot be applied to audio storage and playback devices that handle audio on a frame-by-frame basis. In this case, there were problems in how to thin out a waveform of a fixed time length with a pitch length that varies from frame to frame (timing problem), and in how to process fractional waveforms.

The present invention has been made in view of the above circumstances, and its purpose is to provide a relatively small audio storage and playback device capable of inputting audio in frame units and processing waveform thinning in pitch units in real time. Our goal is to provide the following.

[Means to solve the problem]

The above-mentioned object of the present invention is to encode and store audio, compare and decode the stored encoded audio, thin out the waveform in pitch units of the audio, and output it via a digital/analog conversion circuit. In an audio storage and playback device that handles audio on a frame-by-frame basis, voicedness of the decoded audio waveform frame;
A voice characterized by comprising: a means for determining whether there is no voice or no voice; and a means for successively handling the voice waveform remaining after the thinning process and the voice waveform of the next frame, and thinning the waveform in units of pitches. This is accomplished by a storage and playback device.

[Effect]

In the audio storage and playback device according to the present invention, the encoded audio of the i-th frame transferred to the second memory is decoded into a audio waveform (VF), and its pitch is calculated. next,
For example, during double speed playback, the thinning processing means reduces the residual waveform of the previous frame (i-1st frame) to ■2□-
1, the above v from the fourth memory and the third memory
2□-0 and vF□ are successively read out and thinned out in pitch units.

Number of sample points (NF) and pitch (Nl'□) of the above waveform
Generally speaking, the ratio between
A waveform (Vzi) of Np□ sample points remains unprocessed.

This is transferred to the fourth memory, and the same process as above is repeated together with the audio V F i + of the next frame. The output audio (VO□) for frame i, which is written to the output waveform memory (fifth memory described later) that stores the output audio and has two pages so that writing and reading are possible independently of each other, is as follows: nN P i sample points.

〔Example〕

Embodiments of the present invention will be described in detail below with reference to the drawings.

FIG. 1 is a configuration diagram showing an embodiment of the present invention. In the figure, 1 is a microphone, 2 is an amplification and AID conversion circuit,
3 is ADPCM compliant with CCITT G, 721 recommendation
Encoder, Ml inputs from microphone 1 and amplifies it, 8K
A relatively large-capacity memory that stores audio encoded at 32 Kb/s (4 bits/sampling) by the ADPCM encoder 3 after A/D conversion using Hz sampling;
M2 is a memory that stores encoded audio for one frame (for example, 40 m5 (320 sample points)), 4 is an ADPCM decoder that decodes the encoded audio into audio waveforms in frame units, and is compliant with the CCITT recommendations. M3 is a decoding device. It shows a waveform memory that stores one frame of audio waveform (320 sample points).

During storage, the MCU captures the output of the ADPCM encoder 3 into Ml, and during playback, the MCU captures the encoded audio in Ml one frame at a time each time a transfer request signal TREQ is input.
This transfer control circuit is mainly composed of a microcomputer that transfers the data to ADPCM decoder 2 and activates (FRUN) the ADPCM decoder 3 when the transfer is completed. PWR reads the audio waveform of M3 and calculates the power of the frame, and when the power is above a preset threshold, it is determined to be an audio frame, and when it is less than the threshold, it is determined to be a silent or unvoiced frame (V/U
The power processing circuit, PTH, reads the audio waveform of M3 and calculates the frame pitch, and when it is a voiced frame (VUF=1), outputs the calculated value in the form of a flag (VUF). =0 indicates a pitch processing circuit that outputs a predetermined value, for example, 160 (unit is the number of sample points).

In addition, M4 is a memory for storing residual waveforms that could not be thinned out because the frame length was not an integral multiple of the pitch, and TD
S reads the audio waveform from the memories M4 and M3,
The waveform is thinned out in units of pitch NP, and the remaining waveform is
A thinning processing circuit M5 is a two-page output waveform memory OC in which the output waveform is written by the thinning processing circuit TDS and can be written/read in parallel.
T is a 1-thinning processing circuit TD of the output waveform memory M5 written in L.
The output waveform is read out at every sampling period (8 KHz) from a page different from that written by S, and outputted to the D/A conversion circuit 5. When the set number N0UT has been read out, TREQ is This is an output control circuit that outputs to the MCU, switches the write/read page of M5, and reads the output waveform.

In FIG. 1, the area after M2 separated by a broken line indicates the main parts of the apparatus of this embodiment. In addition, AID conversion circuit 2
The output is 16 bits.

The detailed structure of the memory is shown in FIGS. 2(a) to 2(e).

In the figure, M1 to M5 correspond to those in FIG. 1, respectively.

M14. - is 17 L/-A320 sample points (
ADPCM encoded audio AVJ of j=0 to 319) is stored in frames (F=1.2...N).

Since the encoded audio is backed up by 4 sample points into 1 word, it occupies 80 words per frame.

In addition to one frame of encoded audio A V j (F = j) transferred from Ml by the MCU, M2 has P
Threshold value pwRTH read by WR to perform V/U judgment
, PTH is the waveform correlation threshold P T used in pitch calculation.
T H and pitch search range PTMIN to PTMAX,
The pattern CPAT of the thinning speed used by TDS is stored.

The CPAT shown in the figure is a value for double speed playback, and C2
is read out periodically as "101010..." for each pitch (thinned out at 111 II). For 1.5x playback, MCU is CT=0011. By setting CP=OO1, CP becomes “100100100...
・It is read out as follows.

M5 is the output waveform V of the first frame subjected to the thinning process by the TIS. When (F=i) is written on page 1, OCT shows the waveform ■ of the i-1th frame on page 2. (
F=i-1) and outputs it to the D/A conversion circuit 5. After reading it, TDS shows the processed waveform on page 2.
. Write (F=i+1), OCT is V from page 1. (F=
i) Read out.

The operation of this embodiment configured as described above is shown in Figures 3 to 3 below.
This will be explained using FIG. 5 as well. Figure 3 shows the overall operation as a flowchart, Figure 4 shows the details of the waveform thinning <TDS operation in pitch units, and Figure 5 shows the overall operation timing. .

In FIG. 3, PLAY indicates the entry point for starting the playback of the volumetric audio. First, the transfer control circuit M CU is the number of remaining waveforms I2
is set to O, the encoded audio of the first frame (A VJ (F = 1)), the pattern CPAT of the above-mentioned thinning rate, etc. are transferred from Ml to M2, and A is transmitted by the signal FRU N.
Activate DPCM decoder 4.

The ADPCM decoder 4 reads the encoded audio of M2, decodes it, and writes the audio waveform VF (F=1) to M3. When the decoding is completed, the power processing circuit pWR:
Read V, (F=1.) from M3, calculate power, and determine V/U. Next, the pitch processing circuit engineer] TH is
When the frame is voiced (VUF = 1), calculate the pitch NP, and when VUF = O, calculate the NP to 160.
Set to .

The thinning processing circuit TDS uses the I2 residual waveforms v2 stored in M4 as a "waveform pack" and 1 frame 3 of M3.
The 20 audio waveforms vF are edited into one memory (for example, M3) so that they can be easily accessed consecutively. At this time, M3 has ■2 at addresses O~■2-1, ■2~I2+
vF is stored at address 319. These will be collectively referred to as (1) below. Note that when 2=0, VT=VF, and M3(7) 0-319 addresses nivT are stored (see FIGS. 2(d) and 2(d')).

Subsequently, the number of remaining waveforms (12) is set to I 24320 (the number of waveforms in V). In the first frame, since the number of residual waveforms in the previous frame is O, it is set to Iz''320 here.

Next, the thinning processing circuit TDS executes thinning pattern C,
If is II I II and I z > 2 N p, the waveform is thinned out (TDH31 in FIG. 2). This operation will be explained using FIG. 4.

Figure 4 shows double speed playback (CP = 1010...repetition), pitch NP of the first frame = 12N (unit: 66 Hz in units of 2 sample points and frequency), pitch NP of the second frame = 114 (70Hz).

In the first frame, thinning is started with ■2=320 as the initial value. Since Iz>2Np, it can be thinned out in units of pitches. Therefore, after reading the waveform of 2NF□ from vT (addresses O to 2N, !-1 of M3) and multiplying each waveform by the window function W, the left half waveform and the right half waveform separated by NPl are created. The output waveform V is obtained by adding . , get . Expressing this as a formula, Do t=1. NP□ NP□-1 END D.

Therefore, ■7, which is two pitches long, is V, which is one pitch long. are thinned out. After thinning out, the number of remaining waveforms is updated to 12=I2-2NP□=78, and the next next pattern is read out from M2 as CP corresponding to the pitch.

This completes the first thinning of the first frame, and returns to label r in the flowchart of FIG.

Referring to CP again, C,=1 in double speed playback. However, this time, it does not satisfy 2〉2NP□ (I2
-78.2Np, = 242), so in the flowchart ■
Then, the human waveform vT of two words (78 words) remaining in M3 is transferred to M4 as v2, and the first frame thinning process is completed.

Output waveform V for the audio of the first frame. The number of sample points N0UT□ of □ becomes NP□ points (121 points).

Processing for the encoded audio of the second frame is started when TREQ is output from the output control circuit OCT and, as a result, FRUN is output from the transfer control circuit MCU. In other words, the process returns to label f in the flowchart.

In the second frame, there are 78 points in the human waveform V2, so in the waveform pack, VT (M3Nm is V2°78 from o to address 77).
VF (F=2) is stored at address -397, and I 2=
It is set to 78+320=398.

The operation of the one-thinning process circuit TDS is the same as that for the first frame, and vT is thinned out in units of NP2. As a result, N0UT2=114. l2=170.

Thereafter, in the same manner, under the frame number management by the MCU, playback ends when all frames stored in Ml are processed.

Note that in step TDH82 in FIG. 3, when the thinning pattern CP is 0 and the number of remaining waveforms is one pitch length or more, the waveform for one pitch is directly transferred to the VT (M3
) to VO (M5). At this time, the number of remaining waveforms is set to l2=I2-Np, and the next CP is read from M2. For example, TDH32 is 1.5x playback (CP = 100100...), CP = 1, O (thinning out 2 pitches) and then < c
It operates when p=o.

Finally, the operation timing of the apparatus will be explained using the time chart of FIG. The output control circuit OCT is ADP
CM decoder 4. When the power processing circuit PWR, pitch processing circuit PTH, and thinning processing circuit TDS are processing the audio of the first frame and writing the output waveform to a certain page of M5, the i-1st frame is written from another page of M5. nl-0 output waveforms Vo (F=i-1) obtained by processing the audio are read out at every sampling period and output to the D/A conversion circuit 5. The above ni-□ is the same as the above-mentioned NOUT i-1, and is loaded into one of the counters in the output control circuit OCT when the thinning process of the i-1th frame is completed.

When the D/A conversion output of n knees is completed, a request TREQ to transfer the encoded audio of the j+1 frame to M2 is generated in the transfer control circuit MCU, and the readout page of M5 is switched to start from the first sampling timing. is n1 Vo(
D/A conversion output of F=i) is started. In this way, the device thins out the audio for each frame, dividing the frame into 125Xn (μ5ec) intervals.

n□ changes depending on the frame length (number of audio waveforms in one frame), pitch NP, number of residual waveforms, and thinning pattern, but with current hardware (for example, digital signal processing processor), transfer, decoding, and power processing circuits (pwR ), the pitch processing circuit (PTH), and the thinning processing circuit (TDS) require approximately 10 m5 of processing time, so the minimum value of nd needs to be 80 points. Therefore, the pitch is 50~400I]Z(NP=160~2
0 sample points), an appropriate frame length is 40 m5 (320 sample points).

In addition, the frame length is 32 as the thinning unit length (corresponding to pitch) for silent and unvoiced frames (VUF=O).
Since it is 0 sample point, 320/2n (n is a natural number)
is appropriate.

In the above embodiment, the explanation was mainly about the playback of the stored audio, but regarding the storage of the audio, the microphone 1
It is considered to be the same as the conventional method, as it is inputted from the computer, amplified, A/D converted at 8KHz sampling, encoded at 32Kb/s (4 bits/sampling) by ADPCM encoder 3, and stored in relatively large capacity memory M1. It's good.

In the above embodiment, CC is used as the audio encoding method.
Although an example using the ADPCM encoding method conforming to the ITTG 721 recommendation has been shown, it goes without saying that other methods may also be used.

For example, if the PARCOR method is used as the audio encoding method, the ADPCM encoder 3 in FIG.
In the analyzer, the ADPCM decoder 4 is configured with a PARCOR synthesizer, and in the PARCOR analysis, the spectral parameters of one frame of audio and the V/U flag and pitch are calculated as sound source information. This information is stored in M1 and M2. In this case, the power processing circuit PWR and pitch processing circuit PTH in FIG.
This is a circuit that reads out the flag and pitch.

Furthermore, when unvoiced frames (consonants, etc.) are thinned out, the clarity may decrease and the sound quality may deteriorate. In order to avoid this, the ADPCM encoder 3 in FIG.
, a circuit that encodes the audio and calculates the power of the frame to make a V/U judgment, a transfer control circuit MCU, and a V
Frames with UF “1” and VUF from 0” to 111”
The previous 4 frames and VUF changed to 1 (I I
A control circuit and a power processing circuit that create a silence flag that becomes "1" in frames other than the four frames after the change from + to IL Ojl, and transfer the encoded audio, V/U flag, and silence flag to M2. PWR is from M2 to above V/
The pitch processing circuit PTH, which is a circuit that reads the U flag and the silence flag, calculates the pitch NP when the silence flag is tr Otp and the V/U flag is tt 1 u, and calculates the pitch NP to 160 when the silence flag is 1''. The thinning processing circuit TDS thins out the waveform in units of NP and outputs the waveform when the silence flag is "0" and the V/U flag is "1", or when the silence flag is "1". is written to M5, and the silent flag is set to 0”.
When the /U flag is "Or+," the human waveform of M4 and the audio waveform of one frame of M3 are output as they are as output waveforms of M5.
It may also be used as a circuit to write to.

〔Effect of the invention〕

As described above, according to the present invention, audio is encoded and stored, the stored encoded audio is read out and decoded, the waveform is thinned out in pitch units of the audio, and the waveform is output via a digital/analog conversion circuit. , in an audio storage and playback device that handles audio on a frame-by-frame basis, means for determining whether the decoded audio waveform frame is voiced, unvoiced, or silent; Since we have provided a means to thin out the waveform in pitch units, we have realized a relatively small audio storage and playback device that can input audio in frame units and process waveform thinning in pitch units in real time. This has the remarkable effect that it can be done.

[Brief explanation of the drawing]

Fig. 1 is a configuration diagram showing an embodiment of the present invention, Fig. 2(a)
- (e) are diagrams showing the detailed configuration of the memory, Figure 3 is a flowchart showing the overall operation, Figure 4 is a diagram showing details of the operation of thinning out the waveform in pitch units <"ros," 5
The figure shows the overall operation timing. 2: Amplification, A/D conversion circuit, 3: ADPCM encoder, 4: ADPCM decoder, Ml: Memory, M
2: Frame encoded audio memory, M3: Frame waveform memory, MCU: Transfer control circuit, PWR: Power processing circuit, PTH: Pitch processing circuit, M4: Human waveform memory, TD
S: Thinning processing circuit, M5: Output waveform memory, OCT
: Output control circuit, 5: D/A conversion circuit. H- Figure (Part 2) Figure (Part 3 --> CRsad) m- Medium RAad D/A

Claims (1)

[Claims] 1. A frame unit that encodes and stores audio, reads and decodes the stored encoded audio, thins out the waveform in pitch units of the audio, and outputs it via a digital/analog conversion circuit. In an audio storage and playback device that handles audio, there is a means for determining whether an audio waveform frame is voiced, unvoiced, or silent, and a means for continuously handling the audio waveform remaining after the thinning process and the audio waveform of the next frame in pitch units. An audio storage and playback device characterized in that it is provided with means for thinning out a waveform. 2. A first memory that stores encoded audio, a second memory that stores encoded audio in frames, and reads the encoded audio in frames from the first memory and stores it in the second memory. a control circuit for transferring, and a third memory for storing the audio waveform read out from the second memory and decoded; A fourth memory that determines whether there is no voice or no sound and stores the audio waveform remaining after the thinning process by the thinning processing means, and the audio waveform is continuously read from the fourth memory and the third memory. 2. The audio storage and playback device according to claim 1, wherein said thinning processing means thins out the waveform in pitch units. 3. As a result of the determination of whether the frame is voiced, unvoiced, or silent by the determination means, if the frame is silent or voiced, the thinning processing means performs a thinning process, and if the frame is a voiceless frame, the thinning processing means performs a thinning process. Claim 1 or 2 characterized in that
The audio storage and playback device described in . 4. The third memory that stores the decoded audio waveform is:
4. The audio storage and playback device according to claim 1, wherein the audio storage and playback device has a capacity to store audio waveforms of sample points of at least twice the pitch period that the device can handle. 5. A patent claim characterized in that, when the judgment result by the judgment means is a silent or silent frame, the pitch of the thinning process performed by the thinning process means is set to be equal to or less than the longest pitch period that can be handled by the apparatus. range 1
4. The audio storage and playback device according to items 4 to 4. 6. The method according to claim 1 to 5, characterized in that a memory is provided to store a pattern of the thinning process performed by the thinning process means, and the thinning process is performed based on the contents of the memory. The audio storage and playback device according to any one of the above.