JP3240832B2 - Packet voice decoding method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a packet speech decoding method for encoding a speech signal, receiving and decoding a packet in which the encoded information is packetized and transmitted, and outputting a speech signal.

[0002]

2. Description of the Related Art First, transmission and reception of voice by packet will be described. As shown in FIG. 10, an audio signal input from a terminal 1 is stored in an input buffer 2 and then audio-encoded by an encoding unit 3. The encoded voice is temporarily stored in the transmission buffer 4 and then transmitted from the terminal 5 to the packet network 11 as a packet. The outgoing packet is
The signal is received at the input terminal 6 of the receiver, temporarily stored in the reception buffer 7, and then decoded by the decoding unit 8. The decoded sound is sent to the output buffer 9 and output from the terminal 10 as sound. The speech coding method used here may be either a coding method for each sample or a block coding method for a plurality of samples. For example, linear PCM, I
TU-T Recommendation G. 711 (μ-law PCM) and G. 726
(ADPCM), G.M. 728 (LD-CELP) and CE
An LP (Code Excited Linear Prediction) coding scheme may be used.

FIG. 11 shows the timing of packet transmission / reception. The received packet (b) received at the terminal 6 is delayed in time from the ten transmitted packets P _{1 to} P ₁₀ (a) transmitted from the terminal 5 in FIG. This shows a case where P _{1 to} P ₁₀ arrive with the same delay, that is, arrive at the expected arrival time from the arrival time of the first received packet P ₁ without delay. The audio output from the terminal 10 obtained by decoding the received packet is as shown in FIG. When all the received packets arrive without delay, the output audio signal (c)
No deterioration such as cutting occurs.

However, as shown in FIG. 12A, when the received packets P ₄ and P ₈ arrive later than the estimated arrival times t ₄ and t ₈ , the decoded output audio signal is output as shown in FIG.
(B), the decoded voice signal V ₃ of the packet P ₃
And cutting occurs between the decoded speech signal V ₄ of the packet P _4, is cut between the decoded speech signal V ₇ and V ₈ similarly occurs.

[0005] This conventional packet reception / decoding process is shown in FIG.
As shown in ( ₁ ), a voice packet is received (S ₁ ), each voice packet is decoded (S ₂ ), the decoded voice signal is buffered (S ₃ ), and it is determined whether there is a decoded voice signal in the buffer. It examined (S _4), if there is audio signal to output the audio signal (S _5), silence is output if there is no audio signal (S _6). As described above, the voice is decoded from the received packet and output. If there is no voice to be output at the time of outputting the voice, the output is zero (silence) until the delayed packet is output. ) And the delay of the packets P ₄ and P ₈ causes a cut section in the output voice, and the cumulative time of the cut section becomes the cumulative delay time of the output voice as it is.

Conventionally, in order to prevent such audio disconnection, an output buffer large enough to delay the initial audio output time, absorb an expected packet delay, and continuously output audio is prepared. It has been proposed to. In this case, the time for outputting the initial decoded voice signal V ₁ as example shown in FIG. 12 (c) is delayed for sufficient time T _3, the cutting is eliminated. However, the delay of the voice output becomes large, which is unsuitable for low-latency voice communication assuming conversation.

Conventionally, a processing shown in FIG. 14 has been proposed as a decoding method without output delay. That is, a voice packet is received (S ₁ ), and it is determined whether or not the voice packet is a delay packet delayed from the scheduled time (S ₂ ). If not, the packet is decoded (S ₃ ). , Buffering (S ₄ ) and outputting the decoded voice (S ₅ ). If the packet is a delay packet, the packet is regarded as being lost, and silence is output (S ₆ ). As shown in this case FIG. 12 (d), the the output delay is not a voice signal V ₄ of the packet P ₄ can not delayed decoding, a section corresponding with the voice signal V ₈ of the packet P ₈ are each cut Sections T ₄ and T ₅ will result.

[0008] Therefore, conventionally, a block configuration is shown in FIG. 15, and the processing procedure is treated as shown in FIG. 16, where a packet that cannot be decoded in time for the current frame is treated as a lost packet, and a packet that arrives earlier than the current frame is processed. A method has been proposed in which the current frame is decoded by extrapolation from the frame voice. That is, the control unit 20 monitors the arrival of the information packet to be decoded in the current frame with respect to the reception buffer 7, and when a necessary packet is delayed, the control unit 20 switches the input of the output buffer 9 for switching the switch 22. The side is connected from the output side of the decoding unit 8 to the output side of the interpolation unit 21, and the interpolation unit 21 generates an interpolated speech using the decoded speech information already obtained. FIG.
As shown in FIG. 6, a voice packet is received (S ₁ ), and it is checked whether or not it is a delayed packet that is later than the scheduled time (S ₂ ). If it is not a delayed packet, voice decoding is performed (S ₃ ) and buffering is performed. and (S _4), and then outputs the audio signal (S _5). Performing interpolation processing by the decoded speech signal that has already been received if the delay packet (S _7), buffers (S _4).

As a method of interpolating speech, for example,
A method described in No. 1-7779 “Interpolation receiving device at the time of instantaneous interruption of voice” can be used. This method has a pitch period detecting means for measuring the period of the received voice signal. If interpolation is necessary, the signal before the pitch period is repeated for a necessary time from the start time of the interpolation based on the obtained pitch period. . In addition, R. V. Cox et al. “Robust CELP coders for noisy
backgrounds and noisy channels ", IEEE Proc. ICASSP
-89, pp. 739-742 (1989) describes an interpolation method in a CELP speech coding system. That is, in the CELP speech coding scheme, a linear prediction coefficient, a pitch period, a gain, and an excitation code are transmitted as speech coding information. When interpolating a certain frame, the parameters of the previous frame may be used repeatedly. Further, when the interpolation section is long, the gain may be gradually reduced.

FIG. 12 (e) shows the state of the output sound when the interpolation shown in FIGS. 15 and 16 is performed. That is, FIG.
(D) in the cutting zone T ₄ based on the delay of the packet P ₄ of the interpolated sound signal V using a decoded audio signal V ₃ and the preceding
'Is interpolated, the cutting interval T ₅ based on the same delay of the packet P ₈ immediately before the decoded speech signal V ₇ interpolation signal V ₇ made of' ₃ are interpolated. In this method, a missing packet is interpolated only from a previously arrived frame. Therefore, there is no problem if the same phoneme continues without changing the voice content of the current frame from the content of the immediately preceding frame. However, if there is a phoneme included only in the missing packet, its contents cannot be restored by interpolation.

Conventionally, in the case of voice interpolation, FIG.
As shown in FIG._KVoice signal V
_KTime t ends₀Has a packet P_{K + 1}Of delay
The delay T and the corresponding section T_iIs the packet P_KOr
Interpolation sound signal V_K′ Are interpolated, and the interpolation section T_i
End time t₁Delay packet P_{K + 1}Of the decoded audio signal V
_{K + 1}If (FIG. 17B) is directly connected, t₀And t₁
Has no restrictions, the connected signal is
As shown in FIG.₁Before and after
In addition, the pitch periodicity is disturbed. The interpolation time T_i
The audio output is delayed only by. Already decrypted already received
To use the interpolated voice V_K′ To delay
Total packet P_{K + 1}Will be decoded, and
It is not possible to encode and transmit in consideration of inter-voice
The transmission side (encoder side) and the reception side (decoder
Side), the decoding process is different, and the same audio as the transmitting side is decoded.
Can not be done.

[0012]

As described above, as described above,
In the conventional packet voice decoding method, when a decoded voice having a cut section in which voice interpolation is not performed is used, the voice is degraded in terms of audibility. In addition, when the disconnection is reduced by the buffering method, a large time delay is required, and real-time speech dialogue becomes inconvenient. Further, when a delayed packet is interpolated as a missing packet, phonemes that exist only in the delayed packet cannot be interpolated, and correct voice content cannot be restored.

An object of the present invention is to solve the above-mentioned drawbacks. If a packet is delayed within a certain time limit, a lost frame phoneme is eliminated, a smooth interpolated voice is output, and a cut section is output. Another object of the present invention is to provide a packet voice decoding method which eliminates the problem and does not increase the time delay.

[0014]

According to the present invention, when a packet is delayed, an interpolated audio signal obtained by interpolation is first output following the previous audio signal, and thereafter, within a predetermined time limit. If a packet arrives late to
The decoded audio signal is connected after the interpolated audio signal. In the invention according to claim 1, when the delayed packet arrives without delay, the delayed audio packet is output until the end of the decoded audio signal. In the invention of claim 2, all the delayed packets are decoded, and a silent section after the decoded audio signal is
The time is adjusted by compressing only the time used for the interpolation sound,
According to the third aspect of the present invention, the voiced section of the packet which is further delayed in the second aspect is time-compressed, and the time adjustment for the interpolated voice section is performed by both the compression and the compression in the silent section.

In the packet speech decoding method according to any one of claims 1, 2 and 3, when the speech of the delay packet is connected when the interpolation speech has a pitch periodicity, the interpolation An interpolated voice section is defined from a voice start time to a time that is an integral multiple of the pitch period.
Invention), when a past decoded speech is required for decoding a packet, the speech information immediately before the interpolation speech signal is used for decoding the delayed packet (the invention of claim 5), and the interpolation speech signal and the delayed packet are decoded. The connection with the decoded audio signal is performed by multiplying both of these signals by the window function for interpolation and adding them (the invention of claim 6).

[0016]

According to the present invention, when a packet is delayed, the interpolated speech is output from the encoded speech information of the previously arrived packet until the packet arrives and is decoded. And quality degradation due to cutting can be prevented. Also, by connecting the decoded audio signal when a delayed audio packet arrives, the audio content can be reliably reproduced without losing the phoneme. Moreover, since the decoded audio signal of the delayed packet is compressed on the time axis, there is no accumulation of audio delay.

According to the second aspect of the present invention, if a packet is delayed, the interpolated speech is output from the encoded speech information of the previously arrived packet until the packet arrives and is decoded. In addition, quality degradation due to cutting can be prevented. Further, since a delayed voice packet arrives and a decoded voice signal is connected, the voice content existing in the packet can be reliably reproduced. Furthermore, since the silent section of the audio signal after the decoded audio signal is compressed on the time axis, the delay of the audio does not accumulate.

According to the third aspect of the present invention, if a packet is delayed, the interpolated speech is output from the encoded speech information of the previously arrived packet until the packet arrives and is decoded. In addition, quality degradation due to cutting can be prevented. When the delayed voice packet arrives and the decoded voice signal is connected, the voice content existing in the packet can be reliably reproduced. Further, by performing time axis compression in a silent section and a sound section of the audio signal after the decoded audio signal,
Voice delays do not accumulate.

In the packet speech decoding method according to any one of the first, second and third aspects, according to the fourth aspect of the invention, when the interpolated speech has a pitch periodicity, the speech of the delayed packet is reproduced. In the case of connection, since the interpolation sound is from the start time of the interpolation sound to the time that is an integral multiple of the pitch period, the waveform of the interpolation sound start time and the waveform of the interpolation end time are at the same position of one pitch. Therefore, even if the decoded voice of the delayed packet is connected after that, no discontinuity occurs at the connection boundary. According to the fifth aspect of the present invention, when a past decoded voice is required for decoding a delayed packet, the decoded voice is decoded by using the voice information immediately before the interpolation. Therefore, the voice interpolation process is performed only on the receiving side. The same waveform as that on the transmitting side is generated as the subsequent decoded voice without affecting the decoding process. According to the invention of claim 6, by interpolating the interpolated voice and the delayed packet decoded voice by multiplying them by a window function for interpolation, even if the voice changes during the interpolation, the weights are continuously added. Discontinuity weakens.

[0020]

Example Figure 1 EXAMPLES claim 1, shows claims reception block configuration of a voice packet communication according to the embodiments of the first aspect of the invention, with the parts corresponding to those in FIG. 15. In FIG. 1, packets received from a terminal 6 are stored in a reception buffer 7 and rearranged in the order of transmission packets. As shown in the flow chart of FIG. 2, when the control unit 30 receives a voice packet (S ₁ ), it determines whether or not the voice packet to be decoded is delayed (S ₂ ). Is a decoding unit 8 for sequentially decoding the encoded information to generate a decoded audio signal. If the packet is not a delay packet, the decoded audio signal is sent to the output buffer 9 through the contacts N of the switches 33 and 34, and the audio signal is output from the terminal 10 via the output buffer 9. In the flowchart of FIG.
A packet that has not been delayed is subjected to voice decoding processing (S ₃ ), and after being further buffered (S ₄ ), voice is output (S ₅ ).

If the control unit 30 determines that the packet is a delay packet, as shown in FIG. 2, the interpolation unit 31 performs voice interpolation processing until the delay packet arrives and is decoded (S ₆ ). The interpolation process in this case is a repetition process based on the extraction of the pitch period of the waveform described in the section of [Prior Art],
Alternatively, in the case of the CELP system, the previous transmission parameters are used repeatedly.

This voice interpolation is performed until a delay packet arrives (S ₇ ). When the delay packet arrives, the delay packet is subjected to voice decoding processing (S ₈ ), and the decoded voice signal is processed by the time axis compression unit 32. The time axis compressed and the compressed signal is output to the switch 3 until the end time of the decoded audio signal when the delayed packet arrives without delay.
The signals are output to the output buffer 9 through the contacts A of the terminals 3 and 34 (S ₄ ), and output from the terminal 10 following the interpolation sound (S ₅ ).

The time axis compression method used here is as follows.
For example, D. Malah's dissertation: "Time-Domain Algorithms
 for Harmonic Bandwidth Reduction and Time Scaling
 ofSpeech Signals ", IEEE Trans. on Asouctics, Spee
ch, and Signal Processing, vol.ASSP-27, No.2, pp.1
21-133, (1979) Time domain harmonic structure stretching (TDHS: Tim
e Domain Harmonic Scaling) algorithm, or
Morita and Itakura's Study Group Materials: “Autocorrelation
Method of speech expansion and contraction on the time axis by the method ",
IEICE Technical Report EA86-5 (1
986). These are pitch
A weight window is applied to the preceding and following waveforms in cycle units, and the section
The time axis is compressed by overlapping. FIG. 3 shows TD
The state of 2: 1 compression by the HS algorithm is shown. Ma
Instead, the pitch period Tp is obtained from the audio signal shown in FIG.
Next, for example, as shown in FIG.₁Karaso
Each time t before and after one pitch period Tp₀, T₁To a straight line
The weighted window function that becomes 1 from 0 is shown in FIG.
At the time t₀~ T₁, T ₁~
t_Two(C) of FIG.
The shape is superimposed and added, and a one-pitch circumference shown in FIG.
A time axis compressed audio signal of the period Tp is obtained. Also, the pitch period
It is necessary to perform superposition at appropriate intervals
Then, the time axis is compressed.

As shown in FIG. 4A, when the packets P ₄ and P ₈ are delayed as in FIG.
The output audio signal output by the processing according to FIG.
It becomes as shown in. Since packet P ₄ is delayed, the packet P ₃ of the decoded decoded voice signal V ₄ of the audio signal V ₃ packet P ₄ at a time t _3e ended is too late, the interpolation sound signal V ₃ generated from the audio signal to it ′ Is output continuously to the decoded voice signal V ₃ , and thereafter, in this example, at time t
₅ decoded voice signal V ₄ of the packet P ₄ is obtained, arrives at which time compressed audio signal V ₄ ^* is continuously output the interpolated sound signal V ₃ ', the correct time of late packets P ₄ is expected When the end time t _4e of the decoded audio signal V ₄ is _reached , the transmission of the compressed audio signal V ₄ ^* is stopped, and the decoded audio signal V ₅ of the next packet P ₄ is output from the time t _4e . That is, when a packet P ₄ in this example lags estimated arrival time for a predetermined time or more, would exit the decoded speech signal V ₃ of the packet P _3, the decoded speech signal V ₄ of the packet P ₄ to the end time t _3e In time, an interpolated audio signal is output, and the decoded audio signal V ₄ of the delayed packet P ₄ is output.
Is obtained before the end time t _4e of the decoded audio signal V ₄ when the packet P ₄ is not delayed, the compressed audio signal V ₄ ^* of the decoded audio signal V ₄ is output until the time t 4 _e after this is obtained. I do.

[0025] In this case, the compressed audio signal V ₄ ^* insertion interval t ₅ ~t _4e is combined compressed audio signal V ₄ ^* at the beginning of one pitch period T _P in t _5, or compressed voice of one pitch period T _P The end of the signal V ₄ ^* may be adjusted to t _4e . At time t _7e decoded audio signal V ₇ of the packet P ₇ is completed in the same manner, the decoded audio signal packet P ₈ is too late, it is interpolated by the interpolation sound signal V ₇ ', delayed packets P
_In this example, the compressed audio signal V ₈ ^{* of the} decoded audio signal V ₈ is output until the end time t _{8 e} of the decoded audio signal of the packet P ₈ when the packet P ₈ is received without delay after the time t _9. . In this way, there is no cut section due to the interpolated speech, and the compressed speech signals V ₄ ^* , V ₈ ^* are output, so that the phoneme contents of the decoded speech signals V ₄ , V ₈ are not lost. Also, since the total time length of the interpolated voice signal V ₃ ′ and the compressed voice signal V ₄ ^* matches the decoded voice signal length of one packet, there is no delay in the final output voice, so that voice interactive communication is possible. is there.

[0026]Embodiment of Claim 2 FIG. 5 shows a voice packet to which the embodiment of the second aspect of the present invention is applied.
FIG. 2 shows a block diagram of a receiving side of packet communication. In this case,
The output side of the signal section 8 includes an interpolation section 41 and a silent section detection section 42.
And the silent section time axis compression unit 44 and the contact N of the switch.
The output side of the silence section time axis compression section 44 is connected to a switch.
Switch 44 is connected to the contact A. In the control unit 40, FIG.
As shown in the flow diagram of
(S₁), After that, the voice packet to be decoded
It is determined whether the cost is late (S _Two). Delay
If it is not a packet, a decryption process is performed (S_Three),That
It is checked whether there is a silent section in the decoded audio signal (S_Four), Silence
If not, send decoded audio signal to output buffer 9
(S_Five), If there is a silent section, check if compression processing is necessary
(S₆), If the compression processing is not necessary,
To the output buffer 9 through the respective contacts N of the changers 43 and 45
Send (S_Five) And output to the output terminal 10 (S₇).

[0027] In the case was delayed packets in step S _2, the delay packet arrives, performs speech interpolation process until decoding (S _8, S _9). In this case, the interpolation processing is performed by repeating the processing based on the extraction of the pitch period of the waveform described in the section of the related art, or repeatedly using the previous transmission parameter in the case of the CELP system. If a delayed packet is obtained during the interpolation, a speech decoding process is performed (S
₁₀ ) After the interpolated audio signal, the signal is output from the terminal 10 via the output buffer 9. In this state, a cut section cannot be formed in the output voice, but the output is delayed by the time required for interpolation. Therefore, the silent section detecting section 42 detects the silent section of the decoded audio signal, and if the silent section is detected (S ₄ ) and the compression processing is required (S ₆ ), the silent section time axis compressing section 44 in compressed amount of time required silence decoded audio signal to the interpolator (S _11). Thereby, output delay can be eliminated.

Regarding the silent section detection, the transmission packet
If an identifier of silence or not is given in advance
Uses that identifier. If there is no identifier, receive
On the side, for example, the power P of the current frame_CAnd the average
WaP_VPower ratio (P_C/ P _V) Is below a certain threshold
If there is, it is determined that it is a silent section. Time axis pressure in silent section
As the compression method, the time required for compression is
Just disconnect from signal and connect silence section before and after disconnection
Is fine. When background noise is included in the silent section
Is the pitch period T in the time axis compression shown in FIG._P
Instead of taking a predetermined period, a weighting window
May be superimposed. One packet silence section
If it is shorter than the time of the interpolated voice, divide it into multiple sections
By applying silence section compression, the compression ratio in each section
And sound degradation is small.

FIG. 4 (c) shows the output voice timing of the present embodiment with respect to the received packet of FIG. 4 (a). Here, by interpolating the interpolated sound signal V ₃ 'from the time t _3e due to the delay of the packet P _4, the decoded voice signal V ₄ of the packet P _4, which is delayed in time t ₅ is obtained which immediately interpolated audio signal V Following the ₃ ', all of the decoded audio signal V ₄ and outputs a later decryption compress the silent interval between V ₅ and V ₆ in the speech signal, V _5, V ₆ from the short signal V ₅ ♯, V ₆ ♯ absorbs the length of the interpolated voice signal V ₃ ′. Similarly, the interpolation signal V ₇ ′ is interpolated by the delay of the packet P ₈ ,
When the decoding signal V ₈ of the packet P ₈ is obtained, it allowed to continue in its entirety V ₈ to the interpolated signal V ₇ ', there is no silent section in the decoded audio signal V ₉ of immediately, further followed decoded speech signal V _{In this case} , a silent section in _{10 is} compressed by the length of the interpolation signal V ₇ ′ to obtain a compressed audio signal V ₁₀ #. Thus no cut section by interpolation sound signal, also is not because each decoded voice signal V _4, V ₈ of delayed packets P ₄ and P ₈ is output as the phoneme content is lost. Further, since the time length required for the interpolated audio signals V ₃ ′ and V ₇ ′ is the same as the silent section compression time of the audio signals V ₅ ♯, V ₆ ♯ and V ₁₀ 、, the final output audio delay However, real-time voice conversation communication is possible.

Third Embodiment FIG. 7 is a block diagram showing the receiving side of voice packet communication to which the third embodiment of the present invention is applied. In FIG.
The output side of the decoding unit 8 is a silent / sound section determination unit 52, a silent section time axis compression unit 54, and a sound section time axis compression unit 55.
And the contact N of the switch 53, and connected to the contacts A ₁ and A ₂ of the switch 53 on the output side of the silent section time axis compression section 54 and the output side of the sound section time axis compression section 55, respectively. ing. As shown in the flow chart shown in FIG. 8, when the control unit 50 receives a packet (S ₁ ), it determines whether or not a voice packet to be decoded is delayed (S ₂ ). The decoded speech signal is generated by processing (S ₃ ), it is determined whether the decoded speech signal is a silent section (S ₄ ), and it is determined whether a compression process is required for both a silent section and a sound section. examined (S _5, S _6), one may not need the compression processing decoded speech signal is sent to the output buffer 9 (S _7), the audio signal is output from the terminal 10 through the output buffer 9 ( S _8).

[0031] if it was delayed packets in step S _2, the delay a packet arrives until the decoded audio signal, the audio interpolation processing is carried out (S _9, S _10), the interpolation process in this case, The repetition processing based on the pitch period extraction of the waveform described in the section of the related art, or the previous transmission parameter is repeatedly used in the case of the CELP system. During the interpolation, arrives delayed packet, when the audio decoding process is executed (S _11), through the output buffer 9 following the interpolated sound signal, the decoded audio signal is output from the pin 10. With this processing alone, a cut section cannot be formed in the output voice, but the output is delayed by the time required for interpolation. So silence /
The voiced section determination unit 52 determines whether or not the decoded voice is silent or non-voiced. For the voice signal determined to be silent, (S ₄ )
If you need a compression process compresses (S _6), the time required for interpolation in the silent section time-base compression unit 54 (S _12). Also for the voiced and determined audio signal in step S _4, the case that requires compression process compresses (S _5), the time required for interpolation in the sound interval time-base compression unit 55 (S ₁₃ ). As a result, output delay can be eliminated.

The silent / voiced section determination section 52 transmits the transmission packet.
When a silent or sound identifier is assigned to the
Uses that identifier. If there is no identifier, receive
On the side, for example, the power P of the current section_CAnd average power P of sound section
_VAnd the ratio (P_C/ P_V) Is below a certain threshold
It is assumed to be a sound section, and otherwise, a sound section. Nothing
As the time axis compression method for the sound section, the time required for compression is
Silence section before and after cutting by cutting directly from the decoded audio signal
Only need to be connected. Silence periods include background noise, etc.
If it is rare, the time axis compression shown in FIG.
And the pitch period T _PInstead of a specific cycle
Alternatively, a weighting window may be applied and superimposed. Aru-ku
The compression method is described in the embodiment of the first aspect of the present invention.
The time axis compression method by TDHS shown in FIG. 3 is used.

The interpolation time is long, and the succeeding section (one packet)
Compression time in the decoded audio signal period)
If it takes only a short time compared to
By applying silence / voice interval compression,
Of the time to compress on the time axis, ie the compression ratio is low
This makes it possible to perform processing with less sound degradation. FIG.
FIG. 4D shows the present embodiment for the received packet of FIG.
3 shows the output audio timing. Here, corresponding to FIG.
Are assigned the same reference numerals, and the interpolated audio signal V_Three′
Packet P_FourOf the decoded audio signal V_FourTo connect this
In the example, the decoded audio signal V_FourIs compressed in the time axis,
Compressed signal V_Four ^*Is connected. However, it differs from FIG.
And the decoded audio signal V_FourOf the compressed signal
Use all. With this compression time alone, the interpolation signal
V_Three′ Is not enough for the time length,
V_Five, V₆Each silent section is compressed and the silent compressed signal V
_Five♯, V₆順次 are sequentially connected, and the sound compression signal V_Four ^*
And silence compression signal V_Five♯, V ₆The sum of each compression time with ♯
Interpolation signal V_Three′. Similarly
Signal V₇'The decoded voice signal V₈, V₉, V_TenIn
V₈, V₉For each is a signal that has been compressed with a sound time axis
V₈ ^*, V₉ ^*As V_TenAbout silence section compression
Signal V_Ten圧 縮 and compression of these three decoded audio signals
The total time is the interpolation signal V₇'Is equal to the length
You. Also in this case, the sound compression signal V₈ ^*, V₉ ^*Each
Decoded audio signal V₈, V₉Disconnect each compressed signal
Everything is used without doing.

In this manner, cutting is performed by the interpolated audio signal.
No section and delayed packet V_Four ^*And V₈ ^*Is output
The phonetic content is not lost. Also interpolation
Audio signal V_Three'And V₇'Is the signal V_Five♯,
V₆♯, V_Ten♯ and V_Four ^*, V ₈ ^*, V₉ ^*Silence with / Yes
The same as the total sound section compression time, so the final output
There is no delay in voice and real-time voice dialogue communication is possible.
You.

[0035] Example Figure 1 of another aspect, FIG. 5 shows a simple changeover switch as each of the switching / connecting portion 34,45,56 in Figure 7, the interpolation sound signal and a decoded audio signal delayed packets The connection with can also be as follows. That is, in the invention of claim 4, when the interpolated speech signal has a pitch periodicity, for example, as shown in FIGS. 9 (a) and 9 (b), the pitch period T _P starts from the start time t ₀ of the interpolated speech signal V _K ′. Integer multiple of (T
_The interpolated audio signal V _K ′ up to time t _{1 of i} = nxT _P (in the example, n = 2) is used. Then, the waveform at the start time t ₀ of the interpolated audio signal V _K ′ and the waveform at the interpolation end time t ₁ correspond to the same position of one pitch, and thereafter, the decoded audio signal V _{K + 1} of the delayed packet Are not greatly discontinuous at the connection boundary t ₁ as shown in FIG.

When the decoded speech signal V _{K + 1} of the delayed packet is required, if the decoded speech in the past is necessary, the interpolation signal V _K ′ is not used and the interpolation is started on the assumption that there is no delay. Assuming that it follows the time point t ₀ , decoding is performed using the immediately preceding audio information, that is, the decoded audio signal V _K (the invention of claim 5). By doing so, even if a delay packet is generated on the receiving side, the subsequent voice decoding is not affected by the interpolated voice signal, and the same voice as on the transmitting side can be output.

Further, as shown in FIGS. 9A and 9B, the interpolated speech signal V _K ′ and the delayed packet decoded speech V _{K + 1} are interpolated as a window function, that is, the interpolated speech signal which is a connected signal. V _K gradually decreases from the connection time t ₁ , and conversely, the decoded voice signal V _{K + 1,} which is the connection signal, is multiplied by each window function that gradually becomes ₁ from the connection time t ₁ , added and connected. By doing so, even if the audio signal changes during interpolation,
Since weighting and addition are continuously performed, it is possible to reduce the discontinuity of the connection boundary t ₁ , and it is possible to suppress quality deterioration due to connection (the invention of claim 6).

[0038]

As described above, according to the present invention, if a packet is delayed within a certain time limit, only the delay time is used.
The audio signal is interpolated by the decoded audio signal of the packet that arrived earlier, and then the decoded audio signal of the delayed packet is connected. At that time, the decoded audio signal itself or a silent section thereafter, or a silent and voiced section. Since the time axis compression is performed, it is possible to realize a packet speech decoding method that eliminates the loss of delayed frame phonemes, outputs a smooth interpolated speech signal, and does not increase the time delay.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an example of a receiving apparatus to which the invention of claim 1 is applied.

FIG. 2 is a flowchart showing an example of the reception decoding processing procedure.

FIG. 3 is a waveform chart for explaining time axis compression processing.

FIG. 4 is a block diagram showing an example of received packets (a).
Output examples (b), (c), and
FIG.

FIG. 5 is a block diagram showing an example of a receiving apparatus to which the invention of claim 2 is applied.

FIG. 6 is a flowchart showing an example of the reception decoding processing procedure.

FIG. 7 is a block diagram showing an example of a receiving apparatus to which the invention of claim 3 is applied.

FIG. 8 is a flowchart showing an example of the reception decoding processing procedure.

FIG. 9 is a waveform chart for explaining the invention according to claims 4 to 6, which is a method for connecting an interpolation audio signal and a decoded audio signal of a delay packet.

FIG. 10 is a block diagram showing a general configuration of an audio signal packet transmission / reception transmission system.

FIG. 11 is a diagram showing a relationship among a transmission packet, a reception packet, and a decoded audio signal.

FIG. 12 is a diagram showing a relationship between a delay packet and various conventional decoded audio signals corresponding thereto.

FIG. 13 is a flowchart showing a conventional received packet decoding processing procedure.

FIG. 14 is a flowchart showing a conventional decoding processing procedure including a delay packet.

FIG. 15 is a block diagram showing a conventional decoding device that performs voice interpolation on a delayed packet.

FIG. 16 is a flowchart showing a conventional processing procedure.

FIG. 17 is a waveform diagram for explaining connection between a conventional interpolated audio signal and a delayed packet decoded audio signal.

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-54-139417 (JP, A) JP-A-2-183648 (JP, A) JP-A-5-88697 (JP, A) JP-A-4- 219797 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G10L 19/00

Claims (6)

(57) [Claims] When a packet is delayed by a predetermined time or more from an estimated arrival time, a speech signal is generated by interpolation processing from encoded information of a previously arrived packet, and is output continuously with the output speech signal up to that time. When the delayed packet is received before the scheduled arrival time of the next packet, the decoded speech of the delayed packet is compressed on the time axis, and the decoded speech output when it arrives at the original scheduled arrival time is output. And outputting the interpolated audio signal continuously until the last time. 2. When a packet is delayed by a predetermined time or more from an estimated arrival time, an audio signal is generated by interpolation processing from encoded information of a previously arrived packet, and output as a continuous output audio signal. When the delayed packet is received within a predetermined time, the decoded audio signal of the delayed packet is output following the interpolated audio signal, and the decoded audio signal of the packet after the decoded audio signal is output. A packet voice decoding method comprising compressing a silent section by a time length of the interpolation voice signal. 3. A speech section in a decoded speech signal of a packet after outputting the decoded speech signal of the delayed packet following the interpolation speech signal is time-compressed, and the interpolation is performed by summing the compression with the compression of the silence section. 3. The packet audio decoding method according to claim 2, wherein the packet audio decoding time is equal to a time length of the audio signal. 4. The apparatus according to claim 1, wherein when the interpolated audio signal has a pitch periodicity, a section of the interpolated audio signal is set to an integral multiple of the pitch period. Packet voice decoding method. 5. The method according to claim 1, wherein when decoding of a packet requires past speech information, speech information immediately before said interpolation speech signal is used for decoding of said delayed packet. The packet audio decoding method according to any one of the above. 6. The connection between the interpolated audio signal and the decoded audio signal of the delayed packet by multiplying an interpolation window function by the interpolated audio signal and the decoded audio signal and adding them together. 4. The packet audio decoding method according to claim 1, wherein: