CN1262991C - Method and apparatus for tracking the phase of a quasi-periodic signal - Google Patents
Method and apparatus for tracking the phase of a quasi-periodic signal Download PDFInfo
- Publication number
- CN1262991C CN1262991C CNB008192006A CN00819200A CN1262991C CN 1262991 C CN1262991 C CN 1262991C CN B008192006 A CNB008192006 A CN B008192006A CN 00819200 A CN00819200 A CN 00819200A CN 1262991 C CN1262991 C CN 1262991C
- Authority
- CN
- China
- Prior art keywords
- phase
- signal
- frame
- phase place
- previous frame
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Lifetime
Links
- 238000000034 method Methods 0.000 title claims abstract description 49
- 230000000737 periodic effect Effects 0.000 claims abstract description 22
- 238000005259 measurement Methods 0.000 claims description 32
- 208000035126 Facies Diseases 0.000 claims description 14
- 230000007246 mechanism Effects 0.000 claims description 9
- 238000012544 monitoring process Methods 0.000 claims description 8
- 125000002015 acyclic group Chemical group 0.000 claims description 5
- 230000009466 transformation Effects 0.000 claims description 5
- 238000012806 monitoring device Methods 0.000 claims description 4
- 238000001228 spectrum Methods 0.000 description 64
- 238000005070 sampling Methods 0.000 description 24
- 238000004458 analytical method Methods 0.000 description 23
- 230000007704 transition Effects 0.000 description 15
- 238000012545 processing Methods 0.000 description 13
- 238000011002 quantification Methods 0.000 description 13
- 230000008569 process Effects 0.000 description 10
- 206010038743 Restlessness Diseases 0.000 description 8
- 238000004422 calculation algorithm Methods 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 7
- 238000013139 quantization Methods 0.000 description 7
- 230000005540 biological transmission Effects 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 238000004891 communication Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 230000006835 compression Effects 0.000 description 4
- 238000007906 compression Methods 0.000 description 4
- 230000005284 excitation Effects 0.000 description 4
- 239000002131 composite material Substances 0.000 description 3
- 230000008447 perception Effects 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 238000003786 synthesis reaction Methods 0.000 description 3
- 238000005311 autocorrelation function Methods 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 230000004069 differentiation Effects 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 230000008676 import Effects 0.000 description 2
- 239000002245 particle Substances 0.000 description 2
- 238000010183 spectrum analysis Methods 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000005236 sound signal Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
- G10L19/005—Correction of errors induced by the transmission channel, if related to the coding algorithm
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/48—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
- G10L19/02—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using spectral analysis, e.g. transform vocoders or subband vocoders
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
- G10L19/04—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using predictive techniques
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
Abstract
The present invention relates to a method for tracking phases of pseudo-periodicity signals. The present invention comprises the steps that for a frame with a periodic signal, the phase of the signal is estimated, and a closed-loop performance measuring value is used to monitor the performance of the estimated phase; for a frame with a periodic signal, and the performance of the estimated phase of the signal is lower than a preset threshold level, the phase of the signal is measured. In the estimated phase, if the previous frame is periodic, an initial phase value is set to be equal to the estimated final phase value of the previous frame; if the previous frame is nonperiodic, or a situation that the previous frame is periodic, and the performance of the estimated phase of the previous frame is lower than the preset threshold level exists, the initial phase value is set to be equal to the measured phase value of the previous frame; for a frame with a signal which is nonperiodic, the phase of the signal is measured. Whether the signal of a specified frame is periodic can be judged through open loop periodicity judgement.
Description
Background of invention
Invention field
The present invention is general relevant with the speech processes field, and is especially, relevant with the method and apparatus of the phase place that is used to follow the tracks of quasi-cycling signal.
Background
Voice transfer by digital technology has become universal, particularly in remote and digital cordless phones are used.And this makes people can send the minimal information amount and keep the quality aspect of discovering of realize voice to become interested simultaneously again by channel determining.If send voice by simple sampling and digitizing, then need to be approximately the data rate of per second 64 kilobits (kbps), just can reach the voice quality of traditional analog phone.Yet,, then, data rate is reduced widely by suitable coding, transmission and synthetic again at the receiver place by using speech analysis.
Come the device of the technology of compressed voice to call speech coder using by the parameter relevant of extracting with the model of human speech generation.Speech coder is divided into time block or analysis frame to the calling voice signal.Speech coder generally comprises encoder.Scrambler is analyzed the calling speech frame, and some has related parameter to extract, and then parameter quantification is become binary representation, that is, and and hyte or binary data grouping.By communication channel packet is sent to receiver and demoder.The decoder processes packet makes them go to quantize producing parameter, and uses the parameter synthetic speech frame again through going to quantize.
The function of speech coder is by removing all intrinsic in voice natural redundancies digitized voice signal to be compressed into the low bitrate signal.By representing to import speech frame, and use and quantize, represent that with one group of position parameter realizes digital compression with one group of parameter.If the input speech frame has figure place N
i, and the packet that speech coder produces has figure place N
o, then the compressibility that obtains by speech coder is C
r=N
i/ N
oAnd challenging be to keep under the situation of the high speech quality of decoded speech, realizing target compression simultaneously.The performance of speech coder depends on (1) speech model or above-mentioned analysis and synthesizes the processing combination and carry out fine or notly, and (2) are at every frame N
oThe targeted bit rates place of position, the parameter quantification processing execution gets fine or not.Therefore, the target of speech model is the key element of catching voice signal or target speech quality at every frame with less parameter group.
Speech coder can be used as the time domain coding device and realizes, described time domain coding device is attempted to use high time resolution to handle to catch the time domain speech waveform, voice segment (generally being the subframe of 5 milliseconds (ms)) is encoded at every turn.For each subframe,, can find pinpoint accuracy to represent from the code book space by many various searching algorithms of knowing in technical field.On the other hand, speech coder can be used as the Frequency Domain Coding device and realizes, the Frequency Domain Coding device attempts to catch with one group of parameter the short-term voice spectrum (analysis) of input speech frame, and uses corresponding synthetic the processing to create speech waveform again from frequency spectrum parameter.According to the known quantification technique of describing in " vector quantization and signal compression (1992) " of A.Gersho and R.M.Gray, the parameter quantification device is by representing parameter with the code vector of storing, thereby kept parameter.
The time domain speech coder that the crowd knows is linear prediction (CELP) scrambler of code exciting, in " digital processing of the voice signal " 396-453 (1978) of L.B.Rabiner and R.W.Schafer, describe the Linear Predictive Coder of described code exciting, all quote as a reference at this.In celp coder, it is relevant to analyze the short-term of removing in the voice signal by linear prediction (LP), or redundant, and short-term resonance peak filter factor is found in described linear prediction analysis.Short-term forecasting filtering is applied to the calling speech frame produces a LP residue signal, further make this residual signal modelling and quantification with long-term forecasting filtering parameter and follow-up random code book.Like this, the CELP coding makes the independent coding task that the task of coded time domain speech waveform is divided paired LP short-term filter factor coding and the LP residue is encoded.Can (that is, use identical figure place N by fixed rate for each frame
o) carry out time domain coding, or press variable bit rate (wherein, using different bit rate for dissimilar content frames) and carry out time domain coding.Variable rate coder attempts only to use the bit quantity that needs, and makes the parameter coding to scrambler reach the level that enough obtains aimed quality.At United States Patent (USP) the 5th, 414, a kind of example variable bit rate celp coder is described in No. 796, this patent has transferred assignee of the present invention, and all quotes as a reference at this.
Time domain coding device such as celp coder generally depends on the higher figure place N of every frame
oTo keep the degree of accuracy of time domain speech waveform.If the figure place N of every frame
oRelatively large, (for example, 8kpbs or more than), then this scrambler is general to transmit good speech quality.Yet at the low bitrate place (4kpbs or following), because limited available figure place, the time domain coding device just can not keep high-quality and firm performance.At the low bitrate place, the Waveform Matching ability of limited code book space constraint tradition time domain coding device, and in the commerce of higher rate is used, just can successfully use.
Current, for exploitation in to low bitrate (that is, 2.4 to 4kpbs and following scope in) the high-quality speech scrambler of work exists strong research interest and business demand.Range of application comprises wireless telephone, satellite communication, Internet Protocol telephone, various multimedia and voice flow application, voice mail and other voice storage system.Its driving force is to high performance demands with to the requirement of steadiness under the packet loss situation.Various recent voice coding standardization effort are another direct driving forces that advance the research and development of low rate speech coding algorithm.The low rate speech coder can be created more multichannel or user on each bandwidth that allows to use, and can be suitable for total budget of scrambler specification with the low rate speech coder of the extra play coupling of suitable channel coding, and under the channel error situation, transmit firm performance.
For by than the coding of low bitrate, developed the method for various frequency spectrums or frequency domain voice coding, wherein, with voice signal as frequency spectrum time-variations differentiation analyzes.For example, see R.J.McAulay and T.F.Quatieri " voice coding and synthetic in sinusoidal coding ", the 4th chapter (W.B.Kleijn and K.K.Paliwal edit, 1995).In the spectrum coding device, target is to imitate or predict the short-term voice spectrum of each phonetic entry frame with one group of frequency spectrum parameter, rather than simulated time-variation speech waveform accurately.Then, to the frequency spectrum parameter coding, and use the output frame of setting up voice through the parameter of decoding.Synthetic speech that is produced and original input speech waveform do not match, but have provided the similar quality of discovering.The example of many Frequency Domain Coding devices of knowing comprises multi-band excitation scrambler (MBE), Sine Transform Coding device (STC) and harmonic coding device (HC) in the art.This Frequency Domain Coding device provides the parameter model of the high-quality with succinct parameter group, can be used in the low bitrate place and accurately quantize with less available figure place.
But, the low bitrate coding has been forced harsh restriction to limited code distinguishability or limited code book space, the validity that this has just limited single encoded mechanism makes scrambler can not represent various types of voice segments with equal accuracy under the diversity of settings condition.For example, traditional low bitrate Frequency Domain Coding device does not send the phase information of speech frame.But, come reconstructed phase information by using prima facies place value and linear interpolation technology at random, artificial generation.For example, see people such as H.Yang " in the MBE model, being used for the synthetic quadratic phase interpolation method of speech sound " in 29 Electrontic Letters 856-57 (in May, 1993).Because the artificially produces phase information, so even quantification-removal quantizes to handle to keep sinusoidal wave amplitude kindly, but the output voice that the Frequency Domain Coding device is produced can not be aimed at (that is, main pulse is asynchronous) with original input voice.Therefore proved that it is difficult adopting any closed-loop characteristic mensuration (for example, such as signal to noise ratio (snr) or perception SNR) in the Frequency Domain Coding device.
Used the multi-mode coding techniques to carry out the low bitrate voice coding in conjunction with the open loop mode determination processing.Multimode and Variable-Rate Coding of Speech people such as Amitava Das, a kind of such multi-mode coding techniques has been described among the in SpeechCoding and Synthesis ch.7 (W.B.Kleijn and K.K.Paliwal edit, 1995).Traditional multi-mode scrambler is different patterns, or coding-decoding algorithm, is applied to dissimilar input speech frames.Customize every kind of pattern or coding-decoding processing and represent certain type voice segments with effective and efficient manner, for example, speech sound, unvoiced speech or ground unrest (non-voice).Outside open loop mode decision mechanism checks the input speech frame, and makes the relevant judgement which pattern is applied to this frame.Generally,, parameter is estimated temporarily, and be that the basis is carried out open loop mode and judged according to estimating with a kind of mode decision with feature frequency spectrum according to some by many parameters of extracting from incoming frame.Therefore when not knowing to export the definite situation (that is, aspect voice quality or other performance measurement, what degree the output voice will be close to the input voice) of voice in advance, make mode decision.
According to above-mentioned, be desirable to provide a kind of low bitrate Frequency Domain Coding device that can estimate phase information more accurately.A kind of multi-mode, hybrid domain scrambler further preferably are provided,, some speech frame are carried out time domain coding, and other speech frame is carried out Frequency Domain Coding according to the voice content of frame.Can further be desirable to provide a kind of hybrid domain scrambler, it can carry out time domain coding to some speech frame, and other speech frame is carried out Frequency Domain Coding according to closed loop coding mode decision mechanism.Preferably provide a kind of closed loop, multi-mode, hybrid domain speech coder more again, the time synchronized between output voice that the assurance scrambler produces and the raw tone that is input to scrambler.Described this speech coder in the relevant application of this proposition, described application is entitled as " closed-loop multimode formula mixed-domain linear prediction (MDLP) speech coder ", and it has transferred assignee of the present invention, and all quotes as a reference at this.
It would also be desirable to provide a kind of method, guarantee the output voice that scrambler produces and be input to time synchronized between the raw tone of scrambler.Therefore need a kind of method of accurately following the tracks of the phase place of quasi-cycling signal.
Summary of the invention
The present invention is directed to a kind of method of accurately following the tracks of the phase place of quasi-cycling signal.Therefore, in one aspect of the invention, a kind of method that signal phase is followed the tracks of comprises the following steps: for being that frame during is periodically estimated its signal phase at signal; Measure the performance of the estimated phase place of monitoring with closed-loop characteristic; For at signal being frame during periodically, measure the phase place of its signal; When the performance of estimated phase place drops on predetermined threshold level when following, provide estimated phase place as output phase; And on the performance of estimated phase place drops on predetermined threshold level the time, provide measured phase place as output phase; And, measure the step of the phase place of its signal for the frame during being aperiodicity at signal.
In another aspect of the present invention, a kind of is that method that frame during periodically carries out Phase Tracking quasi-periodicity comprises the following steps: to estimate its signal phase for being frame during periodically at signal to signal; Measure the performance of the estimated phase place of monitoring with closed-loop characteristic; For at signal being frame during periodically, measure the phase place of its signal; When the performance of estimated phase place drops on predetermined threshold level when following, provide estimated phase place as output phase; And on the performance of estimated phase place drops on predetermined threshold level the time, provide measured phase place as output phase.
Aspect another, a kind of device that signal phase is followed the tracks of comprises of the present invention: an estimation unit is used for estimating the phase place of its signal to being the frame during periodically at signal; One monitoring device is used for monitoring with the closed-loop characteristic measurement performance of estimated phase place; One is used for measuring the measurement mechanism of the phase place of its signal to being the frame during periodically at signal; One first actuating unit is used for performance when estimated phase place and drops on predetermined threshold level when following, provides estimated phase place as output phase; One second actuating unit in the time of on the performance of estimated phase place drops on predetermined threshold level, provides measured phase place as output phase; And one be used for the frame during being aperiodicity at signal is measured the measurement mechanism of the phase place of its signal.
Of the present invention aspect another, a kind of is that the device that frame during periodically carries out Phase Tracking quasi-periodicity comprises to signal: an estimation unit, be used to estimate at signal be periodically during the signal phase of frame; One monitoring device is used for monitoring with the closed-loop characteristic measurement performance of estimated phase place; And one be used for measuring the measurement mechanism of the phase place of its signal to being the frame during periodically at signal; One first actuating unit is used for performance when estimated phase place and drops on predetermined threshold level when following, provides estimated phase place as output phase; One second actuating unit in the time of on the performance of estimated phase place drops on predetermined threshold level, provides measured phase place as output phase.
Description of drawings
Fig. 1 is at the block scheme of each end by the communication channel of speech coder termination.
Fig. 2 is the block scheme that can predict the scrambler that use in (MDLP) speech coder at multi-mode, mixed-domain linear.
Fig. 3 is the block scheme that can predict the demoder that use in (MDLP) speech coder at multi-mode, mixed-domain linear.
Fig. 4 is a process flow diagram, and the performed MDLP coding step of MDLP scrambler that can use in the scrambler of Fig. 2 is shown.
Fig. 5 is a process flow diagram, and the voice coding decision process is shown.
Fig. 6 is a closed-loop multimode formula MDLP speech coder.
Fig. 7 is the block scheme of the spectrum coding device that can use in the scrambler of the scrambler of Fig. 6 or Fig. 2.
Fig. 8 is the curve map of amplitude to frequency, is illustrated in the sinusoidal wave amplitude in the harmonic coding device.
Fig. 9 is a process flow diagram, and the mode decision that is illustrated in the multi-mode MDLP speech coder is handled.
Figure 10 A is the voice signal amplitude to the legend of time, and Figure 10 B is the legend of the remaining amplitude of linear prediction (LP) to the time.
Figure 11 A is that speed/pattern in closed loop coding is judged is to the curve map of frame index; Figure 11 B is that perception signal to noise ratio (S/N ratio) (PSNR) in closed loop coding is judged is to the curve map of frame index; And Figure 11 C is speed/pattern when not existing closed loop coding to judge and the PSNR curve map to frame index.
Figure 12 is a kind of block scheme of device that is used to follow the tracks of the phase place of quasi-cycling signal.
Embodiment
In Fig. 1, first scrambler 10 receives digitize voice sampling s (n), and sampling s (n) is encoded, and is used for sending to first demoder 14 on transmission medium 12 or communication channel 12.14 pairs of encoded speech samples of demoder are decoded, and synthetic output voice signal s
SYNTH(n).For transmission in the opposite direction, second scrambler, 16 pairs of digitizing speech samples s (n) encodes, and this sampling is to send on communication channel 18.Second demoder receives encoded speech sample, and it is decoded, and produces synthetic output voice signal s
SYNTH(n).
Speech sample s (n) expression is according to any voice signal that carries out digitizing and quantification in many in the art the whole bag of tricks of knowing (for example, comprising pulse code modulation (pcm), compression expansion μ-Lv or A-rule).As knowing crowd in the art, speech sample s (n) is organized into input data frame, wherein, each frame comprises the digital speech sampling s (n) of predetermined number.In example embodiment, use the sampling rate of 8kHz, each 20ms frame comprises 160 samplings.Among the embodiment that is described below, can be advantageously serving as that the basis changes data transmission rate frame by frame, from 8kpbs (full rate) to 4kpbs (half rate) to 2kpbs (1/4th speed) to 1kpbs (1/8th speed).On the other hand, can use other data rate.As used herein, term " at full speed " or " at a high speed " generally are meant the data rate more than or equal to 8kpbs, and term " half rate " or " low rate " generally are meant the data rate that is less than or equal to 4kpbs.It is favourable changing data transmission rate, because for the frame that comprises less relatively voice messaging, can use than low bitrate selectively.Those skilled in the art that will appreciate that, can use other sampling rate, frame sign and data transmission rate.
First scrambler 10 and second demoder 20 constitute first speech coder or speech coder together.Similarly, second scrambler 16 and first demoder 14 constitute second speech coder together.Those skilled in the art that will appreciate that, can realize speech coder with digital signal processor (DSP), special IC (ASIC), discrete gate logic, firmware or any traditional programmable software modules and microprocessor.Software module can reside in RAM storer, flash memory, register or in the write medium of many any other forms of knowing in the art.On the other hand, any traditional processor, controller or state machine can replace microprocessor.At United States Patent (USP) the 5th, 727, (this patent has transferred assignee of the present invention in No. 123, and all quote as a reference) at this, and the U.S. Patent application the 08/197th that is entitled as " vocoder ASIC " that proposes on February 16th, 1994, (this patent has transferred assignee of the present invention, and all quotes as a reference at this) describes the example ASIC that is in particular voice coding and designs in No. 417.
As in Fig. 2, describing, according to an embodiment, multi-mode mixed-domain linear prediction (MDLP) scrambler 100 that can use in speech coder comprises mode decision module 102, spacing estimation module 104, linear prediction (LP) analysis module 106, LP analysis filter 108, LP quantization modules 110 and MDLP residue scrambler 112.Input speech frame s (n) is offered mode decision module 102, spacing estimation module 104, LP analysis module 106 and LP analysis filter 108.Mode decision module 102 is according to the periodicity of each input speech frame s (n) with such as other parameter generating mode index I that extracts such as energy, frequency spectrum inclination angle, zero passage speed
MWith pattern M.Propose on March 11st, 1997, the U. S. application sequence number 08/815 that is entitled as " method and apparatus that is used to carry out the variable rate speech coding of changing down ", the whole bag of tricks of speech frame being classified according to is periodically described in 354, this application has transferred assignee of the present invention, and all quotes as a reference at this.These methods are included among interim standard TIA/EIA IS-127 of telecommunications industry association and the TIA/EIA IS-733.
In Fig. 3, the demoder 200 that uses in speech coder comprises LP parameter decoder module 202, residue decoder module 204, mode decoding module 206 and LP composite filter 208.206 couples of mode index I of mode decoding module
MReceive the decode, produce pattern M from it.LP parameter decoder module 202 receiving mode M and LP index I
LP202 pairs of values that received of LP parameter decoder module are decoded, to produce the LP parameter that quantizes.Residue decoder module 204 receives residue index I
R, spacing index I
pWith mode index I
M204 pairs of values that received of residue decoder module are decoded, to produce the residual signal that quantizes
The residual signal that quantizes
offers LP composite filter 208 with the LP parameter, its therefrom synthetic output voice signal through decoding
Except MDLP residue scrambler 112, operation and enforcement in the various modules of the demoder 200 of many in the art scramblers 100 of knowing Fig. 2 and Fig. 3, and at above-mentioned United States Patent (USP) the 5th, among " digital processing of the voice signal " 396-453 (1978) of 414, No. 796 and L.B.Rabiner and R.W.Schafer description is arranged.
According to an embodiment, MDLP scrambler (not shown) is carried out in the step shown in the process flow diagram of Fig. 4.The MDLP scrambler can be the MDLP residue scrambler 112 of Fig. 2.In step 300, MDLP scrambler checking mode M is full rate (FR), or 1/4th speed (QR) or 1/8th speed (ER).If pattern M is FR, QR or ER, then the MDLP scrambler forwards step 302 to.In step 302, the MDLP scrambler puts on residue index I to corresponding speed (FR, QR or ER---according to the value of M)
RBeing high precision, high-rate coded for the FR pattern, and advantageously the time domain coding of CELP coding puts on the LP residue frame, or puts on speech frame on the other hand.Transmit frame (after the further signal Processing that comprises number-Mo conversion and modulation) then.In one embodiment, frame is the LP residue frame of expression predicated error.In another embodiment, frame is the speech frame of expression speech sample.
On the other hand, if in step 300, pattern M is not FR, QR or ER, (that is) if pattern M is half rate (HR), and then the MDLP scrambler forwards step 304 to.In step 304, spectrum coding (more advantageously harmonic coding) is put on the LP residue with half rate, or put on voice signal.The MDLP scrambler forwards step 306 to then.In step 306, obtain distortion measurement D by encoded voice being decoded and itself and original incoming frame being compared.The MDLP scrambler forwards step 308 to then.In step 308, distortion measurement D and predetermined threshold T compare.If distortion measurement D greater than threshold value T, then modulates the corresponding quantization parameter of frame half rate, spectrum coding and send.On the other hand, if distortion measurement D is not more than threshold value T, then the MDLP scrambler forwards step 310 to.In step 310, rate is carried out recompile to the frame through decoding at full speed in time domain.Can use any traditional two-forty high precision encryption algorithm, such as, the CELP coding preferably used.Modulate and send the parameter of the FR pattern quantization that is associated with this frame then.
As shown in the process flow diagram of Fig. 5, according to an embodiment, the speech sample that closed-loop multimode formula MDLP speech coder is used for sending in processing is followed one group of step.In step 400, speech coder is received in the digital sample of the voice signal in the successive frame.Speech coder forwards step 402 to according to the given frame that is received.In step 402, speech coder detects the energy of frame.This energy is the measuring of speech activity of frame.Carry out speech detection by comparing to square summation of the amplitude of digital speech sample and energy value and threshold value as a result.In one embodiment, threshold value is carried out self-adaptation according to the variation level of ground unrest.In No. the 5th, 414,796, above-mentioned United States Patent (USP), a kind of demonstration variable thresholding voice activity detector is described.Some unvoiced speech sound can be extremely low-energy sampling, and this just may as background noise encode it mistakenly.In order to prevent this from occurring, can use the frequency spectrum inclination angle of low-yield sampling, from ground unrest, distinguishing unvoiced speech, in No. the 5th, 414,796, above-mentioned United States Patent (USP), describe.
After the energy that detects frame, speech coder forwards step 404 to.In step 404, speech coder judges whether the frame energy that is detected is enough to frame classification as comprising voice messaging.If the frame energy that is detected is reduced under the predetermined threshold level, then speech coder forwards step 406 to.In step 406, speech coder with frame as background noise (that is, noiseless or quiet) coding.In one embodiment, with 1/8 speed, or 1kpbs carries out time domain coding to ground unrest.If in step 404, the frame energy that is detected meets or surpasses predetermined threshold level, be frame classification speech frame then, and speech coder forwards step 408 to.
In step 408, speech coder determines whether this frame is periodic.Periodically the various known method of judging comprises, for example, uses zero crossing and uses normalized autocorrelation function (NACF).Especially, propose on March 11st, 1997, the U.S. Patent application the 08/815th that is entitled as " method and apparatus that is used to carry out the variable bit rate speech coding of changing down ", describe in No. 354 and use zero crossing and NACF to come sense cycle, this application has transferred assignee of the present invention, and all quotes as a reference at this.In addition, the said method that is used for distinguishing speech sound and unvoiced speech is included in interim standard TIA/EIA IS-127 of telecommunications industry association's industry and TIA/EIA IS-733.If it is periodic not judging frame in step 408, then speech coder forwards step 410 to.In step 410, speech coder is encoded frame as unvoiced speech.In one embodiment, with 1/4 speed, or 2kpbs, the unvoiced speech frame is carried out time domain coding.If it is periodic determining this frame in step 408, then speech coder forwards step 412 to.
In step 412, speech coder uses many in the art periodicity detection method of knowing (described at above-mentioned U. S. application sequence number 08/815,354) to determine whether frame has enough periodicity.If judge this frame enough periodicity is not arranged, then speech coder forwards step 414 to.In step 414, with frame as the transition voice (that is the transition from the unvoiced speech to the speech sound) carry out time domain coding.In one embodiment, rate, or 8kpbs is at full speed carried out time domain coding to the transition speech frame.
If in step 412, speech coder determines that frame has enough periodicity, and then speech coder forwards step 416 to.In step 416, speech coder is encoded frame as speech sound.In one embodiment, with half rate, or 4kpbs, the speech sound frame is carried out spectrum coding.More advantageously, the speech sound frame is carried out spectrum coding with the harmonic coding device, following described with reference to figure 7.On the other hand, can use other spectrum coding device, for example, Sine Transform Coding device or multi-band excitation scrambler are as knowing the crowd of institute in the art.Then, speech coder forwards step 418 to.In step 418, speech coder is decoded to encoded speech sound frame.Then, speech coder forwards step 420 to.In step 420, through the speech sound frame of decoding with compare corresponding to the input speech sample of this frame, obtaining the measured value of synthetic speech distortion, and judge whether half rate speech sound spectrum coding model works in acceptable limits.Then, speech coder forwards step 422 to.
In step 422, speech coder judges through the speech sound frame of decoding with corresponding to the error between the input speech sample of this frame whether be reduced under the predetermined threshold.According to an embodiment, make this judgement in following mode with reference to figure 6 descriptions.If coding distortion is reduced under the predetermined threshold, then speech coder forwards step 426 to.In step 426, speech coder uses the parameter of step 416 that this frame is sent as speech sound.If in step 422, coding distortion meets or surpasses predetermined threshold, and then speech coder forwards step 414 to, to the frame of the digital speech sampling that in step 400, receives as the transition voice at full speed rate carry out time domain coding.
Be noted that step 400-410 comprises open loop coding determinating mode.On the other hand, step 412-426 comprises closed loop coding determinating mode.
Among the embodiment shown in Figure 6, a kind of closed-loop multimode formula MDLP speech coder comprises the analog-digital converter (A/D) 500 that is coupled to frame buffer 502, and frame buffer 502 is coupled to processor controls 504 successively.Energy calculator 506, speech sound detecting device 508, ground unrest scrambler 510, two-forty time domain coding device 512 and low rate spectrum coding device 514 are coupled to processor controls 504.Frequency spectrum demoder 516 is coupled to spectrum coding device 514, and Error Calculator 518 is coupled to frequency spectrum demoder 516 and processor controls 504.Threshold value comparer 520 is coupled to Error Calculator 518 and processor controls 504.Impact damper 522 is coupled to spectrum coding device 514, frequency spectrum demoder 516 and threshold value comparer 520.
In the embodiment of Fig. 6, the speech coder parts preferably realize that as the firmware in speech coder or other software-driven module speech coder itself advantageously resides among DSP or the ASIC.Those skilled in the art that will appreciate that, can implement the speech coder parts preferably with many other known methods equivalently.Processor controls 504 is microprocessor advantageously, but can use controller, state machine or discrete logic to realize in addition.
In the multi-mode scrambler of Fig. 6, voice signal is offered A/D 500.The frame that A/D 500 becomes analog signal conversion digital speech to sample, S (n).The digital speech sampling is offered frame buffer 502.Processor controls 504 obtains the digital speech sampling from frame buffer 502, and they are offered energy calculator 506.The ENERGY E that energy calculator 506 is sampled according to following formula computing voice:
Wherein, these frames are that 20ms is long, and sampling rate is 8kHz.Calculated energy E is sent it back processor controls 504.
The speech energy and the speech activity threshold value of 504 pairs of calculating of processor controls compare.If calculated energy is lower than the speech activity threshold value, then processor controls 504 is sent to ground unrest scrambler 510 to the digital speech sampling from frame buffer 502.Ground unrest scrambler 510 uses and keeps ground unrests and estimate required minimum number of bits this frame of encoding.
If calculated energy is more than or equal to the speech activity threshold value, then processor controls 504 is sent to speech sound detecting device 508 to the digital speech sampling from frame buffer 502.Speech sound detecting device 508 judges whether the periodicity of speech frame allows to use the low bitrate spectrum coding to carry out efficient coding.The method of judging the periodicity level in the speech frame knows for the crowd of institute in the art, and comprises, for example, and use normalizing autocorrelation function (NACF) and zero crossing.These methods and other method were described in above-mentioned U. S. application sequence number 08/815,354.
Speech sound detecting device 508 offers processor controls 504 to signal, and this signal has indicated this speech frame whether to comprise enough periodic voice, to encode effectively by spectrum coding device 514.If speech sound detecting device 508 judges that speech frame lacks enough periodicity, then processor controls 504 is sent to high-rate coded device 512 to the digital speech sampling, and it carries out time domain coding with predetermined maximum data rate to voice.In one embodiment, predetermined maximum rate is 8kpbs, and high-rate coded device 512 is celp coders.
If originally speech sound detecting device 508 judges voice signal and has enough periodicity to encode effectively by spectrum coding device 514 that then processor controls 504 is sent to spectrum coding device 514 to the digital speech sampling from frame buffer 502.Describe a kind of example frequency spectrum scrambler in detail below with reference to Fig. 7.
The spacing frequency F that spectrum coding device 514 is extracted estimated
0, the spacing frequency the amplitude A of harmonic wave
IAnd voice messaging V
CSpectrum coding device 514 offers impact damper 522 and frequency spectrum demoder 516 to these parameters.Frequency spectrum demoder 516 can advantageously be modeled to the demoder of the scrambler in the traditional C ELP scrambler.Frequency spectrum demoder 516 produces the synthetic speech sampling according to frequency spectrum codec format (describing below with reference to Fig. 7),
And synthetic speech sampling offered Error Calculator 518.Processor controls 504 sends to Error Calculator 518 to speech sample S (n).
The MSE that calculates is offered threshold value comparer 520, and whether within the acceptable range it judge distortion level, that is, whether distortion level is reduced under the predetermined threshold.
If the MSE that calculates within the acceptable range, then threshold value comparer 520 offers impact damper 502 to signal, and the data of spectrum coding are exported from speech coder.On the other hand, if MSE is not within the acceptable range, then threshold value comparer 520 offers processor controls 504 to signal, and processor controls 504 is sent to two-forty time domain coding device 512 to digital sample from frame buffer 502 successively.Time domain coding device 512 is encoded to frame with predetermined maximum rate, and abandons the content of impact damper 522.
In the embodiment of Fig. 6, the type of employed spectrum coding is a harmonic coding, such as following with reference to figure 7 description, but on the other hand, can be the spectrum coding of any kind, for example, sinusoidal wave transition coding or multi-band excitation coding.For example, use the multi-band excitation coding of describing in 166, and using, the sinusoidal wave transition coding of describing in 068 for example at United States Patent (USP) the 4th, 865 at United States Patent (USP) the 5th, 195.
Be equal to or less than the transition frames and the sound frame of cyclic parameter for the phase distortion threshold value, by two-forty time domain coding device 512, advantageously rate or 8kpbs use the CELP coding to the multi-mode scrambler of Fig. 6 at full speed.On the other hand, for this frame, can use the two-forty time domain coding of any other form known.Therefore, just transition frames (and the inadequate sound frame of periodicity) is encoded,, the waveform of input end and output is mated preferably so that by keeping phase information preferably with high precision.In one embodiment, after handling the continuous sound frame of predetermined number that threshold value surpasses the period measurement value, the multi-mode scrambler for a frame no matter the judgement of threshold value comparer 520 how, all switches to full rate CELP coding from the half rate spectrum coding.
Be noted that energy calculator 506 and speech sound detecting device 508 constitute the open loop coding together with processor controls 504 and judge.In contrast to this, spectrum coding device 514, frequency spectrum demoder 516, Error Calculator 518, threshold value comparer 520 and impact damper 522 constitute the judgement of closed loop coding together with processor controls 504.
In an embodiment who describes with reference to figure 7, use spectrum coding and preferably use harmonic coding, with low bitrate enough periodic sound frames is encoded.General Definition spectrum coding device is an algorithm, but described algorithm is attempted to keep the time-evolution of voice spectrum feature with the significant method of perception by each speech frame being simulated and encoding in frequency domain.The pith of these algorithms is: (1) spectrum analysis or parameter estimation; (2) parameter quantification; And (3) analyze the output speech waveform that has through decoding parametric.Therefore, its target is the key character that keeps the short-term voice spectrum with one group of frequency spectrum parameter, to parameter coding, uses then through the synthetic output of decoding frequency spectrum parameter voice.Generally, synthetic output voice are as the weighted sum of sine wave.Sinusoidal wave amplitude, frequency and phase place be analyze during estimated frequency spectrum parameter.
" analysis-by-synthesis " is the technology that a kind of crowd knows in the CELP coding techniques, and do not utilize this technology in spectrum coding.The main cause that analysis-by-synthesis shall not be applied to the spectrum coding device is because the losing of initial phase information, even be well suited in function from perceptible viewpoint speech model, but synthetic speech all can (MSE) may be very high.Therefore, another advantage that correctly produces initial phase is to obtain a kind of ability, directly speech sample and reconstruct voice is compared to allow to judge whether precision encoding speech frame of speech model.
In spectrum coding, following synthetic output speech frame:
S[n]=S
v[n]+S
uv[n],n=1,2,…,N,
Wherein, N is the hits of every frame, and S
vAnd S
UvIt is respectively sound and noiseless component.The sinusoidal wave synthetic following sound component of having created of handling of suing for peace:
Wherein, L is sinusoidal wave sum, f
kBe the frequency of in short-term spectrum, being concerned about, A (k n) is sinusoidal wave amplitude, and θ (k n) is sinusoidal wave phase place.Estimate amplitude, frequency and phase parameter by the spectrum analysis processing from the short-term spectrum of incoming frame.Noiseless component can together be created with sound part in independent sine wave summation is synthetic, maybe can handle separate computations by special-purpose no phonosynthesis, is added back to S then
vIn.
In the embodiment of Fig. 7, use the particular type spectrum coding device be referred to as the harmonic coding device, with low bitrate enough periodic sound frames is carried out spectrum coding.The harmonic coding device be characterized by a frame sinusoidal wave and, the segment of analysis frame.The frequency that each sine wave in sinusoidal wave summation has is the spacing F of this frame
0Integral multiple.In a further embodiment, wherein, the spectrum coding device of employed particular type is not the harmonic coding device, obtains the sine wave freuqency of each frame from one group of real number between 0 and 2 π.In the embodiment of Fig. 7, advantageously be chosen in each sinusoidal wave amplitude and phase place in the summation, so that summation will mate best with the signal on the one-period, shown in the legend of Fig. 8.Generally, the harmonic coding device uses external sort, is designated sound or noiseless to each input speech frame.For sound frame, sinusoidal wave frequency limitation in estimating spacing (F
0) harmonic wave, that is, and f
k=kF
0For unvoiced speech, use the peak value of short-term spectrum to determine sinusoidal wave.Interpolation amplitude and phase place be with following imitation their differentiation on frame, as:
A(k,n)=C
1(k)*n+C
2(k)
θ(k,n)=B
1(k)*n
2+B
2(k)*n+B
3(k)
Wherein, the characteristic frequency position f outside the short-term Fourier transform (STFT) of the input speech frame of getting window
k(=kf
0) locate, from the instantaneous value estimation coefficient [Ci (k), Bi (k)] of amplitude, frequency and phase place.Each sinusoidal wave parameter to be sent is amplitude and frequency.Do not send phase place, but according in several known technologies any to its simulation as an alternative, for example, described known technology comprises the quadratic phase model, or any traditional phase place polynomial expression formula.
As shown in FIG. 7, the harmonic coding device comprises the spacing extractor 600 that is coupled to window logic 602 and discrete Fourier transformation (DFT) and frequency analysis logic 604.Also be coupled to DFT and frequency analysis logic 604 receiving the spacing extractor 600 of speech sample S (n) as input.DFT and frequency analysis logic 604 are coupled to residue scrambler 606.Spacing extractor 600, DFT and frequency analysis logic 604 and the residue scrambler 606 each all be coupled to parameter quantification device 608.Parameter quantification device 608 is coupled to channel encoder 610, successively, channel encoder 610 is coupled to transmitter 612.By standard radio frequency (RF) interface (for example, such as CDMA (CDMA) air interface) transmitter 612 is coupled to receiver 614.Receiver 614 is coupled to channel decoder 616, successively, channel decoder 616 is coupled to quantizer 618.Going quantizer 618 to be coupled to sinusoidal wave summation voice operation demonstrator 620.Also sine wave summation voice operation demonstrator 620 is coupled to phase estimating device 622, it receives previous frame information as input.The sinusoidal wave summation of configuration voice operation demonstrator 620 is to produce synthetic speech output S
SYNTH(n).
Can realize spacing extractor 600, window logic 602, DFT and frequency analysis logic 604, residue scrambler 606, parameter quantification device 608, channel encoder 610, channel decoder 616 with the various distinct methods (for example, comprising firmware or software module) that those skilled in the art that crowd knows, remove quantizer 618, sinusoidal wave summation voice operation demonstrator 620 and phase estimating device 622.Can realize transmitter 612 and receiver 614 with any equivalence margin RF parts that those skilled in the art that crowd knows.
In the harmonic coding device of Fig. 7, spacing extractor 600 receives input sample S (n), the spacing of extracting frequency information F
0Make sampling multiply by suitable window function by window logic 602 then, the segment of speech frame is analyzed with permission.DFT and frequency analysis logic 604 used the DFT of the pitch information calculating sampling that spacing extractor 600 provide, producing the complex spectrum point, from the described complex spectrum point harmonic amplitude A that extracted
I, shown in the legend of Fig. 8, wherein, L represents the sum of harmonic wave.DFT is offered residue scrambler 606, the residue scrambler 606 acoustic intelligence V that extracted
c
Be noted that as shown in FIG. 8 V
cThe point of parametric representation on frequency axis, more than the point, frequency spectrum is the unvoiced sound signal feature, and no longer is harmonic wave at this.In contrast to this, at a V
cBelow, frequency spectrum is a harmonic wave, and is the speech sound feature.
A
I, F
0And V
cComponent offers parameter quantification device 608, and it quantizes information.Offering channel encoder 610 through quantitative information with block form, channel encoder 610 quantizes grouping with low bitrate (for example, such as half rate, or 4kpbs).Grouping is offered transmitter 612, and 612 pairs of groupings of transmitter are modulated, and the generation signal is sent to receiver 614 aloft.Receiver 614 receives and restituted signal, and encoded packet delivery is arrived channel decoder 616.616 pairs of groupings of channel decoder are decoded, and offering quantizer 618 through decoded packet.Removing quantizer 618 makes information remove quantification.Information is offered sinusoidal wave summation voice operation demonstrator 620.
The sinusoidal wave summation of configuration voice operation demonstrator 620 makes it according to above-mentioned S[n] formula a plurality of sine waves of simulation short-term voice spectrum are synthesized.Sinusoidal wave frequency f
kBe basic frequency F
0Multiple or harmonic wave, described basic frequency F
0It is the spacing periodic frequency of quasi periodic (that is transition) speech sound section.
Sinusoidal wave summation voice operation demonstrator 620 also receives the phase information from phase estimating device 622.Phase estimating device 622 receives the previous frame information, that is, near the A of previous frame
I, F
0And V
cParameter.Phase estimating device 622 also receives N sampling of the reconstruct of previous frame, and wherein, N is frame length (that is, N is the hits of every frame).Phase estimating device 622 is judged the initial phase of frame according to the information of previous frame.The initial phase judgement is offered sinusoidal wave summation voice operation demonstrator 620.Information and initial phase according to present frame calculates (phase estimating device 622 calculates according to the described initial phase of past frame information and executing), and sinusoidal wave summation voice operation demonstrator 620 produces the synthetic speech frame.As mentioned above.
As mentioned above, the harmonic coding device comes synthetic or reconstruct speech frame by using previous frame information and predicted phase ground from the frame to the frame linearity to change.In being referred to as the above-mentioned synthetic model of quadratic phase model usually, coefficient B
3(k) initial phase of the synthetic current sound frame of expression.In judging phase place, traditional harmonic coding device is arranged to zero to initial phase, perhaps produces the prima facies place value randomly or with some pseudorandom production method.For predicted phase more accurately, according to judge near previous frame be still transition speech frame of speech sound frame (that is, enough periodic frame), phase estimating device 622 one of uses in two kinds of possibility methods of judgement initial phases.If previous frame is the speech sound frame, then use the prima facies place value of the last estimation phase value of this frame as present frame.On the other hand, if previous frame be categorized as transition frames, then obtain the prima facies place value of present frame from the frequency spectrum of previous frame, this is that DFT by the demoder output of carrying out previous frame obtains.Therefore, phase estimating device 622 has utilized available precise phase information (is that rate is handled at full speed because of the previous frame as transition frames).
In one embodiment, a kind of closed-loop multimode formula MDLP speech coder is followed the speech processes step of describing in the process flow diagram of Fig. 9.Speech coder is encoded to the LP residue of each input speech frame by selecting only coding mode.Some pattern is encoded to LP residue or speech residual in time domain, and other pattern is represented LP residue or speech residual in frequency domain.The group of pattern has: the full rate time domain (T pattern) that is used for transition frames; The half rate frequency domain (V model) that is used for speech frame; / 4th speed time domains (U pattern) that are used for silent frame; And 1/8th speed time domains (N pattern) that are used for noise frame.
Those skilled in the art that will appreciate that, can follow step shown in Figure 9 voice signal or corresponding LP residue are encoded.The waveform character of noise, noiseless, transition and speech sound can be regarded as the function of time in the legend of Figure 10 A.Noise, noiseless, transition and the remaining waveform character of sound LP can be regarded as the function of time in the legend of Figure 10 B.
In step 700, any input speech residual S (n) that puts in four kinds of patterns (T, V, U or N) is done the open loop mode judgement to relevant.If apply the T pattern, then in step 702, under the T pattern, i.e. rate processed voice residue at full speed in time domain.If apply the U pattern, then in step 704, under the U pattern, promptly in time domain with 1/4th rate processing speech residual.If apply the N pattern, then in step 706, under the N pattern, promptly in time domain with 1/8th rate processing speech residual.If apply V model, then in step 708, under V model, promptly in frequency domain, remain with the half rate processed voice.
In step 710, the voice of coding in step 708 are decoded, and compare with input speech residual S (n), and calculated performance measured value D.In step 712, performance measurement D and predetermined threshold T are compared.If performance measurement D is more than or equal to threshold value T, then in step 714, the speech residual through spectrum coding of step 708 allows to send.On the other hand, if performance measurement D less than threshold value T, then in step 716, handles input speech residual S (n) under the T pattern.In a further embodiment, do not calculate performance measurement, and do not define threshold value.But under V model, handled after the speech residual frame of predetermined number, under the T pattern, handle next frame.
Advantageously, determination step shown in Figure 9 allows only just to use high bit rate T pattern when needs, by utilized the periodicity of speech sound segmentation than the low bitrate V model, when the execution of V model is improper, prevented that by switching to full rate any quality from descending simultaneously.Therefore, can produce high speech quality with the mean speed that is starkly lower than full rate near the speech quality of full rate.In addition, can come the controlled target speech quality by selected performance measurement and selected threshold value.
By keeping model phase locus to approach to import the phase locus of voice, " upgrading " also can improve the performance that follow-up V model is used to the T pattern.When the performance inconsistency in V model in good time, step 710 and 712 closed-loop characteristic inspection switch to the T pattern, thereby improve the performance that follow-up V model is handled by " refreshing " prima facies place value, this allows pattern phase locus to become once more near original input voice phase locus.By as at example as shown in the legend of Figure 11 A-C, carry out in V model from the 5th frame that begins is improper, as obviously seeing by employed PSNR distortion measurement.As a result, closed loop judge and the absence of upgrading in, the phase locus of simulation obviously departs from original input voice phase locus, causes the serious reduction of PSNR, as shown in Figure 11 C.In addition, the performance of the subsequent frame of handling under V model reduces.Yet,, under closed loop is judged, the 5th frame is switched to the T mode treatment as shown in Figure 11 A.By upgrading the performance of the 5th frame is improved greatly, as seeing significantly in the raising of the PSNR shown in Figure 11 B.In addition, also improved the performance of the subsequent frame of under V model, handling.
By split-hair initial phase estimated value is provided, guarantee that the synthetic speech residual signal of V model that is produced accurately aligns in time with original input speech residual S (n), determination step shown in Figure 9 has improved the quality that V model is represented.In the following manner, the initial phase that before obtains the speech residual section of first V model processing from being right after through decoded frame.For each harmonic wave,, then initial phase is arranged to equal the last estimation phase place of previous frame under V model if previous frame is handled.For each harmonic wave,, then initial phase is arranged to equal the actual harmonic phase of previous frame under the T pattern if previous frame is handled.By using complete previous frame to take over to decode remaining DFT, can obtain the actual harmonic phase of previous frame.On the other hand, by handling the various spacing cycle of previous frame, with spacing-method of synchronization, the DFT of the decoded frame of taking over can obtain the actual harmonic phase of previous frame.
In the embodiment that reference Figure 12 describes, the successive frame of quasi-cycling signal S is input in the analysis logic 800.For example, quasi-cycling signal S can be a voice signal for example.Some frame of this signal is periodic, and other frame is not periodic or acyclic.The amplitude of analysis logic 800 measuring-signals, and the amplitude A of output through measuring.Analysis logic 800 is gone back the phase place of measuring-signal, and output is through Measurement Phase P.Amplitude A is offered combinator 802.Also phase value P
OUTOffer combinator 802.Phase value P
OUTCan be through Measurement Phase value P, perhaps phase value P
OUTCan be the phase value P that estimates
EST, as described below.Combinator 802 composite signals, and output is through synthetic signal S
SYNTH
Also quasi-cycling signal S is offered sorted logic 804, it is categorized into aperiodicity or periodicity to signal.For the aperiodicity frame of signal, the phase place P that offers combinator 802
OUTBe arranged to equal Measurement Phase P.The periodic frame of signal is offered closed loop phase estimation logic 806.Also quasi-cycling signal S is offered closed loop phase estimation logic 806.Closed loop phase estimation logic 806 is estimated phase place, and phase place P is estimated in output
ESTEstimate that phase place is according to prima facies place value P
INIT, it is input to closed loop phase estimation logic 806.If sorted logic 804 is categorized into periodic frame to the previous frame that provides, then the prima facies place value is the last estimation phase value of this signal previous frame.If sorted logic 804 is categorized into the aperiodicity frame to previous frame, then the prima facies place value is the Measurement Phase value P of previous frame.
Estimating phase place P
ESTOffer Error Calculation logic 808.Quasi-cycling signal S is also offered Error Calculation logic 808.Also Measurement Phase P is offered Error Calculation logic 808.In addition, Error Calculation logic 808 receives and synthesizes signal S by combinator 802 synthetic warps
SYNTH'.Through synthetic signal S
SYNTH' be as the phase place P that is input to combinator 802
OUTEqual to estimate phase place P
ESTThe time by the synthetic signal S of the synthetic warps of combinator 802
SYNTH Error Calculation logic 808 is come the calculated distortion measured value by comparing and measuring phase value with the estimation phase value, or error measuring value E.In a further embodiment, Error Calculation logic 808 by quasi-cycling signal relatively incoming frame and quasi-cycling signal come the calculated distortion measured value through synthetic frame, or error measuring value E.
Distortion measurement E is offered Compare Logic 810.810 couples of distortion measurement E of Compare Logic and predetermined threshold T compare.If distortion measurement E greater than predetermined threshold T, then is arranged to Measurement Phase P to equal to offer the phase value P of combinator 802
OUTOn the other hand, if distortion measurement E is not more than predetermined threshold T, then estimating phase place P
ESTBe arranged to equal to offer the phase value P of combinator 802
OUT
Therefore, a kind of novel method and equipment that is used to follow the tracks of the phase place of quasi-cycling signal has been described.Those skilled in the art that will appreciate that, can realize or carry out various example logic pieces and the algorithm steps of describing in conjunction with the embodiment that is disclosed here with digital signal processor (DSP), special IC (ASIC), discrete gate or transistor logic, the discrete hardware components such as register and FIFO, the processor of carrying out one group of firmware instructions or any conventional programmable software module and processor.Advantageously, processor can be a microprocessor, but on the other hand, processor can be any conventional processors, controller, microcontroller or state machine.Software module can reside in RAM storer, flash memory, register or in the art in the write medium of many any other forms of knowing.Those skilled in the art that will appreciate that, advantageously represent data, instruction, order, information, signal, position, code element and the chip of indication in the top whole instructions by voltage, electric current, electromagnetic wave, magnetic field or particle, optical field or particle or their any combination.
Therefore, illustrated and described preferred embodiment of the present invention.Yet those skilled in the art that can understand, can make many changes to the embodiment that discloses here and without departing from the spirit and scope of the present invention.Therefore, except that according to the following claims, the present invention is unrestricted.
Claims (18)
1, a kind of method that signal phase is followed the tracks of is characterized in that described method comprises the following steps:
For being that frame during is periodically estimated its signal phase at signal;
Measure the performance of the estimated phase place of monitoring with closed-loop characteristic;
For at signal being frame during periodically, measure the phase place of its signal;
When the performance of estimated phase place drops on predetermined threshold level when following, provide estimated phase place as output phase; And
In the time of on the performance of estimated phase place drops on predetermined threshold level, provide measured phase place as output phase; And
For the frame during being aperiodicity at signal, measure the step of the phase place of its signal.
2, the method for claim 1 is characterized in that, further comprises for given frame judging periodically that with open loop this signal is periodicity or acyclic step.
3, the method for claim 1 is characterized in that, described estimating step comprises the step that constitutes the multi-term expression of phase place according to harmonic-model.
4, the method for claim 1 is characterized in that, described estimating step comprises if previous frame is periodic, the step that the prima facies place value equals the estimated last phase value of previous frame then is set
5, the method for claim 1 is characterized in that, described estimating step comprises if previous frame is acyclic, the step that the prima facies place value equals the measured phase value of previous frame then is set.
6, method as claimed in claim 5 is characterized in that, obtains the measured phase value of previous frame from the discrete Fourier transformation of previous frame.
7, the method for claim 1, it is characterized in that, described estimating step comprises if previous frame is periodic, and the performance of the estimated phase place of previous frame drops under the predetermined threshold level, and the step that the prima facies place value equals the measured phase value of previous frame then is set.
8, method as claimed in claim 7 is characterized in that, obtains the measured phase value of previous frame from the discrete Fourier transformation of previous frame.
9, a kind of is the method that frame during periodically carries out Phase Tracking quasi-periodicity to signal, it is characterized in that, comprises the following steps:
For at signal being frame during periodically, estimate its signal phase;
Measure the performance of the estimated phase place of monitoring with closed-loop characteristic;
For at signal being frame during periodically, measure the phase place of its signal;
When the performance of estimated phase place drops on predetermined threshold level when following, provide estimated phase place as output phase; And
In the time of on the performance of estimated phase place drops on predetermined threshold level, provide measured phase place as output phase.
10, a kind of device that signal phase is followed the tracks of is characterized in that described device comprises:
One estimation unit is used for estimating the phase place of its signal to being the frame during periodically at signal;
One monitoring device is used for monitoring with the closed-loop characteristic measurement performance of estimated phase place;
One is used for measuring the measurement mechanism of the phase place of its signal to being the frame during periodically at signal;
One first actuating unit is used for performance when estimated phase place and drops on predetermined threshold level when following, provides estimated phase place as output phase;
One second actuating unit in the time of on the performance of estimated phase place drops on predetermined threshold level, provides measured phase place as output phase; And
One is used for the frame during being aperiodicity at signal is measured the measurement mechanism of the phase place of its signal.
11, the device as signal phase being followed the tracks of as described in the claim 10 is characterized in that, further comprises a decision maker, is used for judging periodically that with open loop this signal is periodically or acyclic for given frame.
12, the device as signal phase being followed the tracks of as described in the claim 10 is characterized in that, described estimation unit comprises the device that constitutes the multi-term expression of phase place according to harmonic-model.
13, the device as signal phase being followed the tracks of as described in the claim 10 is characterized in that, described estimation unit comprises if previous frame is periodic, the device that the prima facies place value equals the estimated last phase value of previous frame then is set.
14, the device as signal phase being followed the tracks of as described in the claim 10 is characterized in that described estimation unit comprises if previous frame is acyclic, the device that the prima facies place value equals the measured phase value of previous frame then is set.
15, the device as signal phase being followed the tracks of as described in the claim 14 is characterized in that, obtains the measured phase value of previous frame from the discrete Fourier transformation of previous frame.
16, the device as signal phase being followed the tracks of as described in the claim 10, it is characterized in that, described estimation unit comprises if previous frame is periodic, and the estimated phase performance of previous frame drops under the predetermined threshold level, and the device that the prima facies place value equals the measured phase value of previous frame then is set.
17, the device as signal phase being followed the tracks of as described in the claim 16 is characterized in that, obtains the measured phase value of previous frame from the discrete Fourier transformation of previous frame.
18, a kind of is the device that frame during periodically carries out Phase Tracking quasi-periodicity to signal, it is characterized in that, comprising:
One estimation unit, be used to estimate at signal be periodically during the signal phase of frame;
One monitoring device is used for monitoring with the closed-loop characteristic measurement performance of estimated phase place; And
One is used for measuring the measurement mechanism of the phase place of its signal to being the frame during periodically at signal;
One first actuating unit is used for performance when estimated phase place and drops on predetermined threshold level when following, provides estimated phase place as output phase;
One second actuating unit in the time of on the performance of estimated phase place drops on predetermined threshold level, provides measured phase place as output phase.
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/US2000/005141 WO2002003381A1 (en) | 2000-02-29 | 2000-02-29 | Method and apparatus for tracking the phase of a quasi-periodic signal |
Publications (2)
Publication Number | Publication Date |
---|---|
CN1437746A CN1437746A (en) | 2003-08-20 |
CN1262991C true CN1262991C (en) | 2006-07-05 |
Family
ID=21741099
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CNB008192006A Expired - Lifetime CN1262991C (en) | 2000-02-29 | 2000-02-29 | Method and apparatus for tracking the phase of a quasi-periodic signal |
Country Status (8)
Country | Link |
---|---|
EP (1) | EP1259955B1 (en) |
JP (1) | JP4567289B2 (en) |
KR (1) | KR100711040B1 (en) |
CN (1) | CN1262991C (en) |
AU (1) | AU2000233852A1 (en) |
DE (1) | DE60025471T2 (en) |
HK (1) | HK1055834A1 (en) |
WO (1) | WO2002003381A1 (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103811011A (en) * | 2012-11-02 | 2014-05-21 | 富士通株式会社 | Audio sine wave detection method and device |
Families Citing this family (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104347082B (en) * | 2013-07-24 | 2017-10-24 | 富士通株式会社 | String ripple frame detection method and equipment and audio coding method and equipment |
EP2963645A1 (en) | 2014-07-01 | 2016-01-06 | Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V. | Calculator and method for determining phase correction data for an audio signal |
CN108776319B (en) * | 2018-04-25 | 2022-11-08 | 中国电力科学研究院有限公司 | Optical fiber current transformer data accuracy self-diagnosis method and system |
CN109917360A (en) * | 2019-03-01 | 2019-06-21 | 吉林大学 | A kind of irregular PRI estimation method of aliasing pulse |
Family Cites Families (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP0215915A4 (en) * | 1985-03-18 | 1987-11-25 | Massachusetts Inst Technology | Processing of acoustic waveforms. |
CA1332982C (en) * | 1987-04-02 | 1994-11-08 | Robert J. Mcauley | Coding of acoustic waveforms |
US5023910A (en) * | 1988-04-08 | 1991-06-11 | At&T Bell Laboratories | Vector quantization in a harmonic speech coding arrangement |
JPH02288739A (en) * | 1989-04-28 | 1990-11-28 | Fujitsu Ltd | Voice coding and decoding transmission system |
US5765127A (en) * | 1992-03-18 | 1998-06-09 | Sony Corp | High efficiency encoding method |
US5787387A (en) * | 1994-07-11 | 1998-07-28 | Voxware, Inc. | Harmonic adaptive speech coding method and system |
JP3680374B2 (en) * | 1995-09-28 | 2005-08-10 | ソニー株式会社 | Speech synthesis method |
JPH10214100A (en) * | 1997-01-31 | 1998-08-11 | Sony Corp | Voice synthesizing method |
WO1999010719A1 (en) * | 1997-08-29 | 1999-03-04 | The Regents Of The University Of California | Method and apparatus for hybrid coding of speech at 4kbps |
JPH11224099A (en) * | 1998-02-06 | 1999-08-17 | Sony Corp | Device and method for phase quantization |
-
2000
- 2000-02-29 JP JP2002507369A patent/JP4567289B2/en not_active Expired - Lifetime
- 2000-02-29 KR KR1020027011075A patent/KR100711040B1/en active IP Right Grant
- 2000-02-29 CN CNB008192006A patent/CN1262991C/en not_active Expired - Lifetime
- 2000-02-29 WO PCT/US2000/005141 patent/WO2002003381A1/en active IP Right Grant
- 2000-02-29 DE DE60025471T patent/DE60025471T2/en not_active Expired - Lifetime
- 2000-02-29 AU AU2000233852A patent/AU2000233852A1/en not_active Abandoned
- 2000-02-29 EP EP00912054A patent/EP1259955B1/en not_active Expired - Lifetime
-
2003
- 2003-11-07 HK HK03108079A patent/HK1055834A1/en not_active IP Right Cessation
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103811011A (en) * | 2012-11-02 | 2014-05-21 | 富士通株式会社 | Audio sine wave detection method and device |
Also Published As
Publication number | Publication date |
---|---|
EP1259955A1 (en) | 2002-11-27 |
JP2004502203A (en) | 2004-01-22 |
EP1259955B1 (en) | 2006-01-11 |
DE60025471D1 (en) | 2006-04-06 |
AU2000233852A1 (en) | 2002-01-14 |
CN1437746A (en) | 2003-08-20 |
HK1055834A1 (en) | 2004-01-21 |
KR100711040B1 (en) | 2007-04-24 |
KR20020081352A (en) | 2002-10-26 |
WO2002003381A1 (en) | 2002-01-10 |
DE60025471T2 (en) | 2006-08-24 |
JP4567289B2 (en) | 2010-10-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN1266674C (en) | Closed-loop multimode mixed-domain linear prediction (MDLP) speech coder | |
CN100350453C (en) | Method and apparatus for robust speech classification | |
CN1154086C (en) | CELP transcoding | |
US7426466B2 (en) | Method and apparatus for quantizing pitch, amplitude, phase and linear spectrum of voiced speech | |
US6640209B1 (en) | Closed-loop multimode mixed-domain linear prediction (MDLP) speech coder | |
CN1302459C (en) | A low-bit-rate coding method and apparatus for unvoiced speed | |
CN1815558B (en) | Low bit-rate coding of unvoiced segments of speech | |
CN1922659A (en) | Coding model selection | |
CN1188832C (en) | Multipulse interpolative coding of transition speech frames | |
CN1212607C (en) | Predictive speech coder using coding scheme selection patterns to reduce sensitivity to frame errors | |
US6397175B1 (en) | Method and apparatus for subsampling phase spectrum information | |
US6449592B1 (en) | Method and apparatus for tracking the phase of a quasi-periodic signal | |
CN1262991C (en) | Method and apparatus for tracking the phase of a quasi-periodic signal |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CX01 | Expiry of patent term | ||
CX01 | Expiry of patent term |
Granted publication date: 20060705 |