Data processing: speech signal processing – linguistics – language – Speech signal processing – For storage or transmission
Reexamination Certificate
2000-12-20
2002-11-19
{tilde over (S)}mits, T{overscore (a)}livaldis Ivars (Department: 2641)
Data processing: speech signal processing, linguistics, language
Speech signal processing
For storage or transmission
C379S283000, C379S282000
Reexamination Certificate
active
06484139
ABSTRACT:
TECHNICAL FIELD
The present invention relates to an voice encoder used in voice digital wired communication and radio communication, in particular to a method for improving an voice encoder for transmitting non-voice signal using an voice frequency band, such as dual tone multi frequency, DTMF, signals and push button, PB, signals.
BACKGROUND ART
The reduction of communication cost is the most important issue in the private (i.e., local) network. In order to achieve highly efficient transmission of voice signals occupying most part of communication traffics, cases where highly efficient voice encoder based on voice encoding and decoding is applied is increasing, which is exemplified by 8 kbit/s conjugate-structure algebraic-code-excited linear prediction, CS-ACELP, voice encoding method (ITU-T Recommendation G, 729 compliant).
The voice encoding algorithm where the transmission speed is 8 kbit/s has a structure where input signals are specific to voice signals in order to obtain high quality voice with less information amount. This will be described with reference to the 8 kbit/s CS-ACELP system.
FIG. 9
shows a schematic block diagram of an encoder, and
FIG. 10
shows a detail block diagram of the encoder.
This encoding method has an encoding algorithm where the human vocalizing mechanism is modeled. In other words, it is based on CELP method, which uses a composite filter
6
(linear filter corresponding to a voice spectral envelope) where human vocal tract information is modeled to drive time series signals (outputs of an adder
15
) stored in a code book corresponding to the human vocal cords information.
The detailed description of the algorithm can be found in ITU-T Recommendation G. 729, “Coding of Speech at 8 kbit/s using Conjugate-Structure Algebraic-Code Excited Linear Prediction (CS-ACELP)”.
In the coding algorithm specific to voices, higher efficient transmission tends to deteriorate transmission characteristics of signals (such as DTMF signals, PB signals, No. 5 signaling, modem signals) other than voice: signals using the voice frequency band in a transmission path using the highly efficient voice encoder.
Of one example showing the condition, details of LSP quantizer portion will be described with reference to FIG.
11
.
FIG. 11
shows an LSP quantifier portion (
309
) within an encoder based on the CS-ACELP method shown in FIG.
9
.
FIG. 11
includes an MA prediction component calculator
308
for calculating Moving Average (MA) of an LSP, a multiplier
330
, adders
331
,
332
, and
333
, a quantized error weighting coefficient calculator portion
338
for calculating a weighting coefficient based on an input LSP coefficient, a least square error calculator
334
for calculating a square error between a quantized LSP vector calculated in the adder
332
and an LSP vector calculated based on an input voice signals and multiplying it by the weighting coefficient calculated in
334
to select a least square error among quantifier LSP vector candidates, the first stage LSP codebook
335
, the second stage LSP codebook
336
, and an MA prediction coefficient codebook
337
where a plurality kinds of sets of MA coefficients.
Since the LSP quantization method using this structure is described in detail in “CS-ACELP no LSP to gain no ryoushikahou”, Kataoka et al., NTT R&D, Vol. 45 No. 4, 1996, pp. 331-336. Thus, the description is omitted here. It is known that the LSP quantization method is used so that voice signal spectral envelop information can be quantized efficiently.
According to the CS-ACELP voice coding method, the quantization of LSP coefficients is achieved by following three processes. That is, the LSP quantizer portion
309
has three processing function blocks as shown below:
(1) an MA (Moving Average) prediction component calculator portion
308
for subtracting a predictable component between frames in order to achieve efficient quantization;
(2) the first stage LSP quantization code book
335
for using an adaptive code book learned from voices to achieve rough quantization; and
(3) the second stage LSP quantization code note
336
for finely adjusting random number series for an target LSP, which is quantized roughly in the first stage.
The MA (Moving Average) in (1) is used so that signals with few radical changes in frequency characteristics, that is, having strong correlation between frames can be quantized efficiently. Further, the adaptive code book of (2) is used so that a schematic form of a spectral envelope specific to audio signals can be expressed efficiently with a few information amounts. Furthermore, when the random code book of (3) is used in addition to the learned code book of (2) so that slight changes in spectral envelop can be followed flexibly. In consideration of the above-described reasons, it can be said that the LSP quantifier portion
309
is a well suitable method for coding characteristics of voice spectral envelope information efficiently. On the other hand, in order to code non-voice signals, especially DTMF signals, characteristics as described below must be considered:
Voice signals and DTMF signals differ significantly in spectral envelope;
radical changes in spectral characteristics are found between a signal continue time and a pause time. Gains also changes radically. However, a change amount in spectral characteristics and gains only for the duration of DTMF signals is extremely small;
Since quantization distortion of LSP is reflected on frequency distortion of DTMF as it is, the LSP quantization distortion must be small as much as possible; and
For the duration of the DTMF signals, the frequency characteristic is extremely stable.
In consideration of the above-described viewpoints, it cannot be said that the LSP quantizer portion
301
is an effective method for coding the spectral envelope of DTMF signals.
As described in the example above, the non-voice signals such as DTMF signals have different characteristics from those of voice signals in several viewpoints. Thus, when the non-voice signals are coded, it is not suitable to use a same method as one used for voice signals under the condition where the transmission bit rate is low and redundancy for coding is small.
By the way, in the private network, for the call set-up in the telephone communication, the in-channel signalling is performed by using DTMF signals instead of the common channel signalling. In this case, if an allocated transmission path uses the voice coding, it deteriorates transmission characteristics of the DTMF signals. As a result, the call set-up frequently cannot be achieved normally.
As the first solution for overcoming the problem, a device configuration in
FIG. 12
as disclosed in Japanese Unexamined Patent Application Publication No. 9-81199 may be adopted. This configuration includes a unit for, identifying a voice signal and a non-voice signal such as a DTMF signal on the transmission side and memories for storing patterns in which the DTMF signal is pre-decoded on the transmission side and a receiver side. When the identification unit identified a DTMF signal input, an index of a memory holding the coded patterns corresponding to a number of :DTMF to the receiver side, where the index was identified to generate a DTMF signal corresponding to the digit.
As the second solution for overcoming the problem, a device configuration in
FIG. 13
may be adopted, for example. An encoder
101
includes one which is optimized for coding voice signals and one which is optimized for compressively coding non-voice signals (such as DTMF signal) with less distortion. The configuration includes a unit for identifying whether a signal to be transmitted is voice or non-voice and selecting one of the function blocks based on the determination result from the identification unit for coding processing. Further, the configuration includes a unit for folding the determination result into an encoder output so that transmission can be achieved without changing its transmission bit-rate and with least deterioration in voice quality. Furthermore, a sear
Mitsubishi Denki & Kabushiki Kaisha
{tilde over (S)}mits T{overscore (a)}livaldis Ivars
LandOfFree
Voice frequency-band encoder having separate quantizing... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Voice frequency-band encoder having separate quantizing..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Voice frequency-band encoder having separate quantizing... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2982099