Linear Prediction Speech Coder Research Articles

A GEOSTATISTICAL MODEL FOR LINEAR PREDICTION ANALYSIS OF SPEECH

This paper presents a geostatistical model as a new approach to the linear prediction analysis of speech. The autocorrelation method of autoregressive modeling, which is widely applied in the linear predictive coding of speech, is used as a benchmark for comparison with the present algorithm. Before discussing the proposed model, we will briefly describe the concepts of linear prediction analysis of speech and how this is solved by the well-known method of autocorrelation. Following is the introduction of geostatistics including the ideas of regionalized variables, semi-variograms and kriging equations. We then propose a geostatistical model to the linear prediction modeling of speech signals. Examples on speech data are given to illustrate the effectiveness of the present algorithm in comparison with the autocorrelation method. Advantages offered by the proposed geostatistical algorithm over the autocorrelation method in the linear prediction analysis of speech are summarized as follows: (1) it is more effective due to the optimization of the kriging equations taking into account the biased condition; (2) it is more flexible by allowing different biased values for the fitting of the signal spectrum, and therefore may provide a means for adaptive LPC; (3) it can give a good estimate of the number of poles used in the LPC by means of the theoretical semi-variogram.

Fixed Point Implementation of the QCELP Speech Coder

The Qualcomm code excited linear prediction (QCELP) speech coder was adopted to increase the capacity of the CDMA Mobile System (CMS). In this paper, we implemented the QCELP speech coding algorithm by using TMS320C50 fixed point DSP chip. Also the fixed point simulation was done with C language. The computation complexity of QCELP on TMS320C50 was 10k words and data memory was 4k words. In the normal call test on the CMS, where mobile to mobile call test was done in the bypass mode without double vocoding, mean opinion score for the speech quality was he Qualcomm code excited linear prediction (QCELP) speech quality was 3.11.

Performance of ITU-T G.729 8 kb/s CS-ACELP speech codec with nonvoice narrowband signals

A report is given on the results of a series of objective measurements conducted by COMSAT in a laboratory environment aimed at characterizing the narrowband performance of the ITU-T G.729 8 kb/s conjugate-structure algebraic code-excited linear prediction (CS-ACELP) speech coder. The test procedures followed ITU-T Recommendation G.720, "Characterization of Low-Rate Voice Coder Performance with Non-Voice Signals". It was concluded that the G.729 algorithm has excellent performance with narrowband signals in general (e.g., single tones and DTMF signals). As for Signaling System No. 5 (SS5) interregister signals, the G.729 CS-ACELP frequently failed to correctly identify SS5 digit 6 in a number of occurrences, using worst-case analysis equipment. This indicates that the SS5 performance of G.729 codecs in trunks where SS5 is used should be carefully measured before the network planner decides on its deployment. Great care should also be taken for tandem connections, since no test has been performed for these configurations.

Efficient frequency domain postfiltering for multibandexcited linear predictive coding of speech

A frequency domain postfilter for multiband excited linear predictive (MBELP) coders is proposed. It is based on the principle of multiplying the signal band magnitudes by the postfilter magnitudes obtained by sampling the postfilter response at the pitch harmonics. This approach offers two major advantages: first, the signal energies before and after postfiltering can be equalised by simply rescaling the signal band magnitudes; secondly, the phase information will be retained owing to only the magnitude spectra being modified. A fast method is given for sampling the postfilter magnitude spectrum characterised by line spectral pair (LSP) parameters.

Algebraic vector quantization of LSF parameters with low storage and computational complexity

This correspondence presents an algebraic vector-quantization scheme for encoding line spectral frequency (LSF) parameters used in linear predictive coding (LPC) of speech. The scheme is based on low-dimensionality regular-point lattices. The algebraic codebook need not be stored, and the optimum codevector is found through simple rounding of the input vector. Thus, the scheme results in significant savings of memory and reduced computational complexity when compared to traditional vector-quantizer solutions. The quantizer achieves an average spectral distortion of about 1 dB at 28 b/frame for the telephone bandwidth.

Method and apparatus for decoding LPC-encoded speech using a median filter modification of LPC filter factors to compensate for transmission errors

Disclosed herein are methods and apparatus for improving the quality of synthesized speech that is transmitted through a channel that is susceptible to transmission errors. In a presently preferred embodiment of the invention a speech signal is assumed to be first encoded using a Linear Predictive Coding (LPC) technique prior to transmission. The parameters that describe the short-term spectral behavior of the speech signal are received and then applied to and processed by a non-linear median processing block only on an occurrence of a predetermined number of transmission errors in the received LPC speech signal. The median-processed short term speech parameters are subsequently employed, together with a received excitation signal, in a synthesis filter to synthesize a speech signal of improved quality over what would be obtained if the short term speech parameters were not median processed to compensate for the transmission errors.

Volterra adaptive prediction of speech with application to waveform coding

AbstractRecent studies have shown that the airflow in the vocal tract is highly unstable and oscillates between its walls. Therefore linear prediction speech analysis, which is based on laminar airflow hypothesis, leads to approximate representations. This paper deals with nonlinear speech modeling and its exploitation to high quality medium‐rate coding. We first give evidence that the nonlinearities in speech can be described by a second‐order finite memory Volterra operator. An algorithm for performing adaptive nonlinear prediction is described. Application of the algorithm to speech coding is then reported and stability and computational issues are discussed. Performance evaluations and comparisons with linear predictive speech coding are reported and show that improvements in coding performances can be obtained.

Glottal source modeling for voice conversion

This paper describes recent advances in glottal source modeling for speech synthesis. In particular two procedures for modeling the glottal excitation waveform are described and applied to voice conversion. One model uses a polynomial to represent the glottal excitation waveform for one pitch period. The coefficients of the polynomial model form a vector that is used to design a glottal excitation code book with 32 entries for voiced excitation. The codebook is designed and trained using two sentences spoken by different speakers. Speech is synthesized using a quantized glottal excitation waveform for one speaker as the excitation for a glottal excitation linear predictive (GELP) synthesizer designed using tract parameters obtained from the speech of another speaker. Our implementation of the LP synthesizer is patterned after both a pitch-excited LP speech synthesizer and a code excited linear predictive (CELP) speech coder. In addition to the glottal excitation codebook, we use a stochastic codebook with 256 entries for unvoiced noise excitation. Analysis techniques are described for constructing both codebooks. The GELP synthesizer, which resynthesizes speech with high quality, provides the speech scientist with a simple speech synthesis procedure that uses established analysis techniques, that is able to reproduce all speech sounds, and yet also has an excitation model waveform that is related to the derivative of the glottal flow and the integral of the residue. Another approach uses the LF glottal volume-velocity waveform to model the characteristics of three voice types: modal, breathy, and vocal fry (creaky). We then convert a modal voice to sound like a breathy or vocal fry voice using the vocal tract characteristics for modal voice and the glottal volume-velocity waveform model for breathy and vocal fry voices as the excitation.

Pulsed residual excited linear prediction

Linear predictive coding of speech has been widely used at 16 kb/s in the form of adaptive predictive coding (APC) down to 4.8 kb/s in the form of code-excited linear prediction (CELP). Since its invention in 1984 there have been many variations of CELP which differ mainly in the way the final excitation signal (codebook) is produced and quantised. These variations either produce better speech quality or lower complexity. Three new excitation types, all of which are based on a pulsed residual, are proposed. The new pulsed residual excitations improve the speech quality significantly. In addition a novel mathematically equivalent codebook search method which reduces the search complexity significantly is described.

Efficient codebook search procedure forvector-sum excited linear predictive coding of speech

An efficient codebook search method for the EIA/TIA IS-54 vector-sum excited linear predictive (VSELP) speech coder is described. The method uses a two-stage search procedure. In the first stage, diagonal approximation of the correlation matrix of the filtered basis vectors is assumed and a simple sign detection procedure is used to identify a codeword which is close to the optimum codeword. In the second stage, a refinement search is carried out on those codewords which have a Hamming distance of one from the codeword obtained in the first stage. The new search procedure has a complexity only proportional to the bit rate which is much faster than the Gray code search employed in the IS-54 VSELP coder. Simulation results show that the SNR obtained using the proposed fast procedure is the same as that obtained in the standard VSELP coder.

Excitation pulse positioning method in a linear predictive speech coder

A method for positioning excitation pulses for a linear predictive coder (LPC) operating according to the multi-pulse principle, i.e. a number of such pulses are positioned at specific time points and with specific amplitudes. The time points and the amplitudes are determined from the predictive parameters (a k ) and the predictive residue signal (d k ), by correlation between a speech representative signal (y) and a composed synthesized signal (y). All possible time positions for the excitation pulses within a given frame interval are provided. The possible time positions are divided into a number (n f ) of phase positions and each phase position is divided into a number of phases (f). All phases are vacant for the first excitation pulse. When this pulse has been positioned, the phase determined for this pulse is denied to the following excitation pulses until all pulses in a frame have been positioned.

Fully vector-quantized neural network-based code-excited nonlinear predictive speech coding

Recent studies have shown that nonlinear predictors can achieve about 2-3 dB improvement in speech prediction over conventional linear predictors. In this paper, we exploit the advantage of the nonlinear prediction capability of neural networks and apply it to the design of improved predictive speech coders. Our studies concentrate on the following three aspects: (a) the development of short-term (formant) and long-term (pitch) nonlinear predictive vector quantizers (b) the analysis of the output variance of the nonlinear predictive filter with respect to the input disturbance (c) the design of nonlinear predictive speech coders. The above studies have resulted in a fully vector-quantized, code-excited, nonlinear predictive speech coder. Performance evaluations and comparisons with linear predictive speech coding are presented. These tests have shown the applicability of nonlinear prediction in speech coding and the improvement in coding performance. >

Apparatus for linear predictive coding and decoding of speech using residual wave form time‐acess compression

A speech coding and decoding apparatus for use in linear predictive coding of a speech signal includes coding apparatus having a pitch analyzing element for separating an analysis frame into one or more blocks and calculating the strength of correlativity between pitch periods of a residual waveform in each block. A residual partially compresing element compresses the time axis of the residual waveform in a block having a high correlativity strength. A residual quantizing element quantizes the compressed residual waveform while preferentially allotting quantization allotment bits to the compressed portion of the residual waveform. Decoding apparatus includes a residual inverse quantizing element for inversely quantizing the residual waveform by the same bit allotment. A residual partially expanding element expands the compressed portion of the residual waveform to its original length and supplies it to a linear predictive synthesis filter.

A non bit-exact approach for implementation verification of the CCITT LD-CELP speech coder

This paper presents the design methodology behind the floating point verification procedure for the Low-Delay Code-Excited Linear Prediction speech coder, recently selected by the CCITT as Recommendation G.728. This procedure is based on a non bit-exact specification, which is different from previous CCITT speech coder verification procedures. This approach gives additional freedom for the implementor of the algorithm, and will allow for more efficient implementations on various kinds of hardware. However, this flexibility also means that different implementations will respond slightly different to verification test signals. To cope with this, explicit objective measurements of such deviations are used in the verification process. These measurements are simple weighted and unweighted signal-to-noise ratios (WSNR and SNR). In addition to the objective measurements, certain restrictions had to be placed on the test sequence design. In spite of the input restrictions, a set of test sequences giving reasonably good coverage of the LD-CELP algorithm and state space has been found. Evaluation experiments are reported, showing that these sequences have a satisfactory error detecting capability. A final discussion concludes that the chosen verification approach is indeed feasible as an implementor s tool.

Digital speech coder having improved long term lag parameter determination

First Page

Low-delay hybrid vector excitation linear predictive speech coding

A hybrid approach in determining the excitation vector in a low-delay code excited linear predictive coder is proposed. By a judicious division of the composite excitation vector into long-term and short-term components, and the use of switched quantisation, substantial improvement in coding quality is obtained.

A new efficient algorithm to compute the LSP parameters for speech coding

In this paper, the split Levinson algorithm is used to develop an efficient algorithm to compute the Line Spectrum Pairs (LSP) in Linear Predictive Coding (LPC) of speech. We propose two new real functions defined from the reciprocal and antireciprocal parts of the predictor polynomials obtained from the split Levinson algorithm. These functions are shown to obey three-term recurrence relations. Thus the LSP parameters are directly available from the eigenvalues of tridiagonal matrices, the entries of which are computed from only one version of the split Levinson algorithm. When compared with other existing methods, this algorithm is better in terms of complexity. Der Split-Levinson-Algorithmus wird benutzt zur Entwicklung eines effizienten Algorithmus' für die Berechnung der Line-Spektrum-Pairs (LSP) bei der linear-prädiktiven Kodierung (LPC) von Sprache. Wir schlagen zwei neue reelle Funktionen vor, die wir mit Hilfe des reziproken und des anti-reziproken Teils der Prädiktor-Polynome definieren, welche man aus dem Split-Levinson-Algorithmus erhält. Es wird gezeigt, daβ diese Funktionen Rekursionsbeziehungen in drei Termen Gehorchen. Auf diese Weise Kann man die LSP-Parameter unmittelbar aus den Eigenwerten tridiagonaler Matrizen erhalten, deren Elemente sich aus nur einer Version des Split-Levinson-Algorithmus berechnen lassen. Im vergleich zu anderen bekannten Methoden weist dieser Algorithmus eine geringere Komplexität auf. Dans ce papier, l'algorithme de Levinson éclaté est utilisé pour développer un algorithme efficace pour le calcul des paires de raies spectrales (LSP) en prédiction linéaire pour le codage de la parole. Nous proposons deux nouvelles fonctions réelles, ayant des propriétés intéréssantes, définies à partir des polynômes préditeurs symétriques et anti-symétriques. Les relations de récurrences à trois termes pour les fonctions réelles sont déduites. Les paramètres LSP sont ensuite calculés directement à partir des valeurs propres de matrices tridiagonales symétriques dont les éléments sont calculés par une seule version de l'algorithme de Levinson éclaté. Comparé à d'autres méthodes existantes, cet algorithme est le moins complexe en terme de nombre total d'opérations.

A low-delay CELP coder for the CCITT 16 kb/s speech coding standard

A low-delay code-excited linear prediction (LD-CELP) speech coder which is expected to be standardized in 1992 as a CCITT G Series Recommendation for universal applications of speech coding at 16 kb/s is presented. The coder achieves a one-way coding delay of less than 2 ms by making both the LPC predictor and the excitation gain backward-adaptive and by using a small excitation vector size of five samples. The official CCITT laboratory tests revealed that the speech quality of this 16 kb/s LD-CELP coder is either equivalent to or better than that of the CCITT G.721 standard 32-kb/s ADPCM coder for almost all conditions tested. A description of the LD-CELP algorithm, its implementation on the DSP32C for CCITT testing, and performance results from these tests are presented. >

Method for improving speech quality in code excited linear predictive speech coding

By reconciling differences between the estimator and the filter of a code excited linear predictive (CELP) voice coder, higher quality is achieved in the output speech. The pulse amplitudes and pitch tap gain are solved for simultaneously to minimize the estimator bias in the CELP excitation. Increased signal to noise ratio is accomplished by modifying the pitch predictor such that the pitch synthesis filter accurately reflects the estimation procedure used to find the pitch tap gain, and by improving the excitation analysis technique such that the pitch predictor tap gain and codeword gain are solved for simultaneously, rather than sequentially. These modifications do not result in an increased transmission rate or significant increase in complexity of the CELP coding algorithm.

On the design of connectionist vector quantizers

The Kohonen self-organizing feature map has been widely used for the design of connectionist vector quantizers (VQ). One of the features of the Kohonen algorithm is that the weight update gain sequence η ( m) is a decreasing function of the number of iterations, and if incorrectly chosen can lead to very long training times. Here we derive the time-optimal gain sequence and demonstrate its efficacy for a number of cases. The performance is demonstrated using a Gauss-Markov source and compared with a VQ designed using the LBG algorithm. It is demonstrated that the new method is time optimal and that its performance tends to that of the VQ with the LBG algorithm. The special case of a connectionist VQ for linear predictive or auto-regressive data is analysed. The optimal design is derived for the Itakura distance measures. It is again compared with the LBG design. Two connectionist models for finite-state vector quantizers (FSVQ) are proposed. These models learn not only the distribution but also the conditional distribution of a source. The connectionist FSVQs exploit the redundancy between the vectors (frames) of a highly correlated source, and further improve the rate-distortion performance. Their architectures and learning algorithms are presented. To evaluate the performance of the proposed quantizers, comparisons between quantized and unquantized speech spectra are presented. The connectionist VQ and FSVQ for linear predictive data are finally applied to a multipulse linear predictive speech coder using data from the TIMIT database. Comparisons are made of waveforms and rate-distortion functions.