To address the challenging task of achieving effective voice communication over mobile ad hoc networks (MANETs), we introduce a new multiple description (MD) speech coder based on the AMR-WB (adaptive multirate wideband) standard. The MD coder splits the bitstream of the AMR-WB coder into two redundant sub-streams by directly selecting overlapping subsets of encoded data generated for each frame. The sub-streams are transmitted on different network paths. When both sub-streams arrive at the decoder, an output identical to that of AMR-WB is recovered. If only one substream arrives at the decoder, degraded, but still acceptable, speech quality is obtained. The performance of the multiple description coding system is tested for MANETS with a network simulator and an informal listening test. The results demonstrate that this approach makes effective use of the channel capacity and provides reliable end-to-end connections for voice communication in MANETs.

This paper introduces a sinusoidal modeling technique for low bit rate speech coding wherein the parameters for each sinusoidal component are sequentially extracted by a closed-loop analysis. The sinusoidal modeling of the speech linear prediction (LP) residual is performed within the general framework of matching pursuits with a dictionary of sinusoids. The frequency space of sinusoids is restricted to sets of frequency intervals or bins, which in conjunction with the closed-loop analysis allow us to map the frequencies of the sinusoids into a frequency vector that is efficiently quantized. In voiced frames, two sets of frequency vectors are generated: one of them represents harmonically related and the other one nonharmonically related components of the voiced segment. This approach eliminates the need for voicing dependent cutoff frequency that is difficult to estimate correctly and to quantize at low bit rates. In transition frames, to efficiently extract and quantize the set of frequencies needed for the sinusoidal representation of the LP residual, we introduce frequency bin vector quantization (FBVQ). FBVQ selects a vector of nonuniformly spaced frequencies from a frequency codebook in order to represent the frequency domain information in transition regions. Our use of FBVQ with closed-loop searching contribute to an improvement of speech quality in transition frames. The effectiveness of the coding scheme is enhanced by exploiting the critical band concept of auditory perception in defining the frequency bins. To demonstrate the viability and the advantages of the new models studied, we designed a 4 kbps matching pursuits sinusoidal speech coder. Subjective results indicate that the proposed coder at 4 kbps has quality exceeding the 6.3 kbps G.723.1 coder.

A novel paradigm based on pitch-adaptive windows is proposed for solving the problem of encoding the fixed codebook excitation in low bit-rate CELP coders. In this method, the nonzero excitation in the fixed codebook is substantially localized to a set of time intervals called windows. The positions of the windows are adaptive to the pitch peaks in the linear prediction residual signal. Thus, high coding efficiency is achieved by allocating most of the available FCB bits to the perceptually important segments of the excitation signal. The pitch-adaptive method is adopted in the design of a novel multimode variable-rate speech coder applicable to CDMA-based cellular telephony. Results demonstrate that the adaptive windows method yields excellent voice quality and intelligibility at average bit-rates in the range of 2.5-4.0 kbps.

This paper presents a matching pursuits sinusoidal speech coder which incorporates new techniques including a novel vector quantization (VQ) technique used for the weighted quantization of spectral magnitude vector, and an interframe quantization of spectral magnitudes using an interpolation matrix that minimize the weighted interpolation error. The paper describes a novel vector quantization technique, wherein the quantized vector is obtained by applying a linear transformation selected from a first codebook to a codevector selected from a second codebook. The transformation is selected from a family of linear transformations, represented by a matrix codebook. Vectors in the second codebook are called residual codevectors. In order to avoid high complexity during the search for the best linear transformation, each linear transformation is assigned a representative vector, such that the search can be done employing the representative vectors. The VQ design algorithm is based on joint optimization of the linear transformation and the residual codebooks. The introduced techniques are general enough to be used in any sinusoidal speech coding scheme. In this work we incorporated the techniques into the matching pursuits sinusoidal model to achieve high quality speech using sinusoidal speech coder at 4 kbps. Subjective tests indicate that the proposed coding model at 4 kbps has quality comparable to that of G.729 at 8 kbps.

This paper presents a novel vector quantization (VQ) technique in which the quantized vector is formed by adding the transformed outputs of a multistage codebook rather than just adding the outputs of the stages as in regular multistage vector quantization (MSVQ). The transformations are selected from a family of linear transformations represented by a codebook of matrices. This technique can be viewed as a generalized form of MSVQ. If the transformations are constrained to be the identity transformation, this technique becomes identical to the regular MSVQ. The design algorithm is based on joint optimization of the linear transformations and the stage codebooks. It is shown that the proposed technique yields high quality spectral magnitude quantization with performance exceeding that of multistage vector quantization (MSVQ) of similar complexity and bit rate.

This paper addresses the problem associated with variable-dimension vector quantization, and presents a new quantization technique that combines a variable-size nonsquare transform (NST) with a fixed-dimension vector quantizer. We show that all linear dimension conversion methods can be treated as special cases of a general approach for linear dimension conversion formulated as NST. By incorporating the speech perceptual properties, we introduce a technique called weighted nonsquare transform vector quantization (WNSTVQ) for the quantization of speech spectral vectors. We show that the total perceptual weighted distortion can be separated into the weighted modeling distortion, which is solely determined by the choice of transforms in WNSTVQ and the weighted quantizer distortion. We discuss the factors that influence the performance of the WNSTVQ system and provide a complexity analysis for two WNSTVQ implementations. Finally, experimental results are presented to show that the WNSTVQ system has the ability to trade performance for computational complexity and memory storage by selecting suitable transforms and/or the length of fixed-dimension vectors.

A new hybrid speech coding technique is presented in this paper, which combines a frequency-domain parametric coder (for stationary voiced and stationary unvoiced speech) with a time-domain waveform coder (for transition speech). Our hybrid coder uses a parametric representation for the excitation of a linear-prediction filter. The excitation of stationary voiced speech is a sum of harmonic cosines with interpolated magnitudes and a synthetic phase model, the excitation for stationary unvoiced speech is a spectrally shaped noise, and the excitation for transition speech is a set of signed pulses. Signal alignment when switching between the harmonic excitation of stationary voiced speech and the pulse model used for transition speech is required, and achieved by special alignment procedures. A 4 kb/s hybrid coder, which achieves high-quality reconstructed speech, is described. The 4 kb/s hybrid coder employs a neural network classifier, and a novel pitch detection and harmonic bandwidth estimation algorithm. The locations of excitation pulses for coding transitions are determined by analysis-by-synthesis. A simple and efficient dimension conversion and quantization of the harmonic. Spectral magnitudes of voiced speech was devised, combining the general nonsquare transform (NST) or dimension conversion and a weighted vector quantization (VQ) approach. Subjective listening tests demonstrate that the 4 kb/s hybrid coding scheme competes favorably with CELP coders at low bit-rates.

We introduce a general and powerful approach to sinusoidal modeling of speech wherein a closed-loop analysis-by-synthesis (AbS) technique sequentially extracts the parameters for each sinusoidal component. Low bit-rate speech coding is achieved by efficiently constraining the allowed frequencies of sinusoidal components into sets of frequency intervals or bins. In conjunction with the closed-loop analysis, the constrained frequency regions allow us to efficiently vector quantize the frequency information in each frame. In voiced frames, two sets of frequency vectors are generated: one for harmonically related components and the other for non-harmonically related components of the voiced segment. In transition frames, a vector of nonuniformly spaced frequencies is selected from a frequency codebook using frequency bin vector quantization (FBVQ) to represent the frequency domain information. The effectiveness of the coding scheme is enhanced by exploiting the critical band concept of auditory perception in defining the frequency bins. In transition segments, the sinusoidal phases are modeled and coded. Subjective tests with a partially quantized model indicate that, for a target rate of 4 kbps, the coder quality exceeds that of the G.729 standard at 8 kbps.

This paper proposes a bandwidth-scalable coding scheme based on the G.729 standard as a base layer coder. In the scheme, according to the channel conditions, the output speech of the decoder can be selected to be narrowband (4-kHz bandwidth) or wideband (8-kHz bandwidth). The proposed scheme consists of two layers: base and enhancement. The base coder uses the G.729 algorithm to encode narrowband speech. The enhancement coder is based on a full-band CELP model and it encodes wideband speech while making use of the available base layer information. Two bandwidth-scalable coders are designed: one is scalable with the 8 kbit/s G.729 base coder and another with the 6.4 kbit/s G.729 (Annex D) base coder. Subjective tests show that, for wideband speech, the proposed coders at 16 kbit/s achieve better performance than the 16 kbit/s MPEG-4 CELP with bandwidth scalability.

This paper presents a 4 kb/s analysis-by-synthesis multimode harmonic coder (AbS-MHC). Novel features of this coder include a signal modification technique that allows time-domain analysis-by-synthesis parameter estimation in a sinusoidal coding framework, and a frequency-domain transition speech model with improved parameter estimation and quantization schemes. An efficient quantization scheme for harmonic magnitudes based on weighted non-square transform vector quantization (WNSTVQ) is also used. Subjective quality tests indicate that the 4 kb/s AbS-MHC coder outperforms the 5.3 kb/s G.723.1 standard CELP coder and produces speech quality very similar to the 6.3 kb/s G.723.1 coder.

This paper presents a 1.2 kbps speech coder based on the mixed excitation linear prediction (MELP) analysis algorithm. In the proposed coder, the MELP parameters of three consecutive frames are grouped into a superframe and jointly quantized to obtain a high coding efficiency. The interframe redundancy is exploited with distinct quantization schemes for different unvoiced/voiced (U/V) frame combinations in the superframe. Novel techniques for improving performance make use of the superframe structure. These include pitch vector quantization using pitch differentials, joint quantization of pitch and U/V decisions and LSF quantization with a forward-backward interpolation method. Subjective test results indicate that the 1.2 kbps speech coder achieves approximately the same quality as the proposed federal standard 2.4 kbps MELP coder.

Harmonic coders that synthesize speech without transmitting phase information abandon the benefits of closed-loop parameter estimation via waveform matching. In this paper, we show that effective closed loop parameter estimation can be achieved when a suitable time-scale modification is applied to the speech LP residual in harmonic coders. The concept is demonstrated here specifically for pitch estimation, but is more broadly applicable. For each of a set of pitch candidates generated by a time-domain pitch estimator, the residual is modified to match the pitch contour derived from that candidate. The best candidate is selected by evaluating for each candidate the match between the modified residual and the synthesized residual. The new pitch estimation algorithm significantly reduces gross pitch errors compared to a conventional time-domain pitch estimator and enhances the perceptual performance of a 4 kbps harmonic coder.

In this research, we have investigated about prosodic features of Farsi (Persian) language and quantified major stress rules and some intonation rules for speech synthesis purpose. The research is mostly concentrated on pitch variations and then on durational changes. We have implemented the proposed simplified prosodic rules using a Klatt formant synthesizer, specially modified for Farsi phonemes. In order to achieve to a better speech quality, we have exploited different allophonic forms for some consonants, leading to a total of 207 Farsi diphones synthesized by the speech synthesizer. Subjective listening tests show that the addition of the prosodic features drastically increases both the intelligibility and naturalness of the synthesized speech. The synthesizer is software-implemented on a Pentium PC and operates in real-time.

A major source of audible distortion in current low-bit-rate harmonic speech coding algorithms is the ineffective modeling of the transitional speech signals such as onsets, plosives, etc. A new method of modeling transitional speech based on a frequency domain approach is introduced in this paper. The approach uses a modified harmonic model able to produce non-periodic pulse sequences in conjunction with a closed-loop analysis-by-synthesis scheme for parameter estimation and quantization. The structure of a speech coding system based on this model is outlined. The proposed approach is shown to give better performance than transition encoding based on a standard CELP algorithm at rates of 4.8 kb/s.

In this paper we present a new approach for speech coding, which combines frequency-domain harmonic coding for periodic and "noise like" unvoiced segments of speech with a time-domain waveform coder for transition signals. This hybrid coder requires special handling of the boundary between voiced and transition segments. We outline the details of a 4 kbps hybrid coder and present subjective quality test results of this coder.

This paper presents a robust voicing estimation algorithm for low bit rate harmonic speech coding. The algorithm is based on waveform time-warping followed by spectral matching based on voiced and unvoiced local spectral models. The objective of time warping is to reduce the effect of pitch variations on the voicing decision. Several adaptive techniques are used to improve the flexibility and robustness of the conventional spectral matching algorithm. An objective evaluation of the new voicing algorithm is obtained by comparing to manually estimated voicing values. Subjective tests of a sinusoidal coder using the new voicing algorithm show significantly better performance than the standard spectral matching under both clean and noisy environment.

This paper presents a variable-rate CELP codec which achieves good communications speech quality at an average rate of about 3 kb/s. The codec operates as a source-controlled variable rate coder with rates of 4.9 kb/s for voiced and transition sounds, 3.0 kb/s for unvoiced sounds and 670 b/s for silent frames. New techniques used in the codec include prediction of the fixed codebook target vector and joint optimization of the adaptive and fixed codebook search. The prediction of the fixed codebook target vector is based on fixed codebook selections in previous subframes and a running estimate for the fundamental frequency. Informal subjective testing (MOS) indicates that the proposed codec, at an average rate of less than 3.2 kb/s, achieves better quality than fixed rate standard codecs with rates in the range 4-4.8 kb/s.

Several techniques for low-rate speech coding have emerged, requiring quantization of the spectral magnitudes. The set of spectral magnitudes may be considered as a variable-dimension vector with dimension dependent on the pitch period. The authors present a technique called nonsquare transform vector quantization (NSTVQ). This technique addresses the problem of variable-dimension vector quantization by combining a fixed-dimension vector quantizer with a variable-sized nonsquare transform. The experimental results presented show that NSTVQ outperforms existing harmonic magnitude quantization techniques.

We present spectral excitation coding (SEC), a speech codec based on a sinusoidal model applied to the excitation signal. A phase dispersion algorithm allows the same model to be used for voiced as well as unvoiced and transitional sounds. The phase dispersion algorithm significantly improves the perceived quality resulting in more natural reconstructed speech. A new technique for variable dimension vector quantization called nonsquare transform vector quantization (NSTVQ) is used for quantization of the harmonic magnitudes. The SEC system at 2.45 kb/s achieved an MOS score 0.8 points higher than the 2.4 kb/s ZPC-10 standard. A preliminary 1.85 kb/s SEC system which uses zero-bit magnitude quantization is also presented. Informal listening tests indicate that the quality of the 1.85 kb/s system exceeds that of the LPC-10 standard.

A common aspect of speech transmission through packetised networks is the need to consider the discarded (missing) packets as a result of error detection or network overload. The missing packets and the possible mistracking that results in the speech decoder lead to significant quality degradation. We introduce a packet recovery technique for CELP based speech coders. The proposed technique extrapolates independently the excitation signal and the short-term synthesis filter. A recovery strategy based on speech classification (voiced, unvoiced, transition, silence) is discussed. The extrapolation of the short-term filter uses a least-squares fading memory polynomial filter applied to the reflection coefficients. Objective and subjective quality evaluations of the recovery system applied to the LD-CELP G.728 standard and a variable rate CELP system for random and burst frame erasures are presented. The results indicate that the system is robust up to a frame erasure rate of 10%. Very little degradation in quality was observed at erasure rates up to 3%.

Although the mobile communication channels are time-varying, most systems allocate the combined rate between the speech coder and error correction coder according to a nominal channel condition. This generally leads to a pessimistic design and consequently an inefficient utilization of the available resources, such as bandwidth and power. This paper describes an adaptive coding system that adjusts the rate allocation according to actual channel conditions. Two types of variable rate speech coders are considered : the embedded coders and the multimode coders and both are based on code excited linear prediction (CELP). On the other hand, the variable rate channel coders are based on the rate compatible punctured convolutional codes (RCPC). A channel estimator is used at the receiver to track both the short term and the long term fading condition in the channel. The estimated channel state information is then used to vary the rate allocation between the speech and the channel coder, on a frame by frame basis. This is achieved by sending an appropriate rate adjustment command through a feedback channel. Experimental results show that the objective and the subjective speech quality of the adaptive coders are superior than their non-adaptive counterparts. Improvements of up to 1.35 dB in SEGSNR of the speech signal and up to 0.9 in informal MOS for a combined rate of 12.8 kbit/s have been found. In addition, we found that the multimode coders perform better than their embedded counterparts.>

Lattice low-delay vector excitation coding (LLD-VXC) is a speech coding system based on analysis-bysynthesis excitation coding and backward adaptation of the synthesis filter. The introduction of a lattice filter as a (high order) short-term predictor has significant advantages, such as fast tracking of speech signal nonstationarities, simple stability verification, and uniform distribution of the computational load. The objective of this paper is to present a Lattice LD-VXC (LLD-VXC) codec and experimental results obtained at rates of 8, 9.6, and 16 kb/s. A sign algorithm for lattice filter adaptation is introduced in order to reduce computational complexity. An LLD-VXC codec with a 20th-order lattice predictor, a 10thorder lattice weighting filter, and a backward pitch predictor achieved toll quality at 16 kb/s and good communications quality at 8-9.6 kb/s with a delay of less than 2 ms and reasonable complexity.

A tree-searched multistage vector quantization (VQ) scheme for linear prediction coding (LPC) parameters which achieves spectral distortion lower than 1 dB with low complexity and good robustness using rates as low as 22 b/frame is presented. The M-L search is used, and it is shown that it achieves performance close to that of the optimal search for a relatively small M. A joint codebook design strategy for multistage VQ which improves convergence speed and the VQ performance measures is presented. The best performance/complexity tradeoffs are obtained with relatively small size codebooks cascaded in a 3-6 stage configuration. It is shown experimentally that as the number of stages is increased above the optimal performance/complexity tradeoff, the quantizer robustness and outlier performance can be improved at the expense of a slight increase in rate. Results for log area ratio (LAR) and line spectral pairs (LSPs) parameters are presented. A training technique that reduces outliers at the expense of a slight average performance degradation is introduced. The method significantly outperforms the split codebook approach.>

In vector transform quantization (VTQ), vectors consisting of M consecutive samples of a waveform are transformed into a set of M coefficients that are quantized by m>

The authors present a tree searched multi-stage vector quantization (TS-MSVQ) scheme which achieves spectral distortion lower than 1 dB with low complexity and good robustness using 24 b/frame. The M-L search is used and it is shown that it achieves performance close to that of the optimal search for a relatively small M. The best performance/complexity trade-offs are obtained with relatively small size codebooks cascaded in a three-four stage configuration. Results for log-area ratio (LAR) and line spectral pain (LSP) parameters are presented. A training technique which reduces outliers at the expense of a slight average performance degradation is introduced. The robustness across different languages and input spectral shapings is studied. Finally, it is shown that TS-MSVQ significantly outperforms the split-codebook approach.>

Speech codecs based on code excited linear prediction (CELP) traditionally use an adaptive short-term filter, an adaptive codebook (long-term filter), and a fixed (stochastic) excitation codebook. The authors examined the possibility of replacing the fixed stochastic codebook by an adaptive codebook with adaptation based on the characteristics of the unquantized residual. In a typical 4-kb/s CELP codec, the authors use the spectral magnitude and phase of the unquantized residual to experimentally estimate an upper bound on the performance improvement which could be obtained by excitation codebook adaptation. The results suggest that adaptation methods based only on the spectral magnitude (including fractal-based codebooks) are unlikely to result in significant improvement. Adaptation based on the spectral phase information, on the other, shows a significant potential for improving CELP speech quality. The authors also present results of a preliminary test designed to investigate the effect of quantization noise on phased-based adaptation of excitation codebooks.>

The need for a 16-kb/s speech coding algorithm that has very low coding delay while achieving essentially the same high quality as the 32-kb/s adaptive differential pulse code modulation (ADPCM) standard G.721 is addressed. The authors describe low-delay vector excitation coding (LD-VXC), a new coding algorithm which provides high quality with less than 2 ms of coding delay and is robust to transmission errors. The algorithm combines techniques such as vector quantization, analysis-by-synthesis, and perceptual weighting together with backward adaptive linear predictive encoding, and uses a novel long-term predictor employing backward adaptive pitch tracking. Perceptually based nose shaping and postfiltering contribute to the masking of audible quantization noise.>

The authors report results on lattice low delay vector excitation coding (LLD-VXC) for rates in the range of 8-16 kb/s. The LLD-VXC, which is based on a backward adaptive analysis-by-synthesis configuration, an adaptive lattice short-term predictor, an adaptive lattice perceptual weighting filter, and an adaptive pitch predictor, offers toll speech quality at 16 kb/s with moderate complexity and a total communication delay of under 2 ms. For bit rates below 16 kb/s, the LLD-VXC keeps the same basic structures and algorithms but uses different code-vector dimensions and codebook sizes to achieve different bit rates. For reducing computational complexity, the sign-error-sign-data least mean square adaptive lattice algorithm was derived and implemented in LLD-VXC. The sign algorithm significantly reduced computational complexity at the expense of a slight speech quality degradation.>

A custom VLSI architecture for implementing low-delay speech coding algorithms is presented. The architecture consists of a variable number of adaptive arithmetic units (AAUs) in parallel; each AAU has an adaptive datapath to perform the required filtering operations on one element (sample) of data. The number of AAUs is optimized for power consumption and chip area. A configuration with 4 AAUs provides a good tradeoff between low power and small area. It implements the lattice low-delay vector excitation coding algorithm with an estimated power consumption of less than 300 mW and area of 90 mm/sup 2/, which compares favorably with implementations on general-purpose DSP chips.>

Adaptive vector transform quantization (AVTQ) as a coding system is discussed. The optimal bit assignment is derived based on vector quantization asymptotic theory for different PDFs (probability density functions) of the transform coefficients. Strategies for shaping the quantization noise spectrum and for adapting the bit assignment to the changes in the speech statistics are discussed. A good estimate of the efficiency of any coding system is given by the system coding gain over scalar PCM (pulse code modulation). Based on the optimal bit allocation, the coding gain of the vector transform quantization (VTQ) system operating on a stationary input signal is derived. The VTQ coding gain demonstrates a significant advantage of vector quantization over scalar quantization within the framework of transform coding. System simulation results are presented for a first-order Gauss-Markov process and for typical speech waveforms. The results of fixed and adaptive systems are compared for speech input. Also, the AVTQ results are compared to known scalar speech coding systems.>

This paper presents a new approach to efficient vector quantization. The proposed Transform Domain Vector Quantization (TDVQ) system consists of a two-stage coding scheme: A Karhunen-Loewe Transform stage through which minimal representation of the source is achieved followed by a feature extraction based Vector Quantizer (VQ). Based on the compact structure of the source in its eigenspace, two different non-related techniques are employed to circumvent ccmplexity in VQ's: an optimal fast search algorithm which substantially alleviates the computational complexity of encoding correlated sources and a sub-optimal partial representation technique which reduces both memory and computational requirements. The complexity and the performance of the TDVQ system are presented both for an highly correlated first order Gauss-Markov source and for speech waveforms, confirming its advantage over the traditionally used time domain VQ.

Vector quantization, in its simplest form, may be regarded as a generalization of PCM (independent quantization of each sample of a waveform) to what might be called "vector PCM," where a block of consecutive samples, a vector, is simultaneously quantized as one unit. In theory, a performance arbitrarily close to the ultimate rate-distortion limit is achievable with waveform vector quantization if the dimension of the vector,k, is large enough. The main obstacle in effectively using vector quantization is complexity. A vector quantizer of dimensionkoperating at a rate ofrbits/sample requires a number of computations on the order ofk2^{kr}and a memory of the same order. However, a low-dimensional vector quantizer (dimensions 4-8) achieves a remarkable improvement over scalar quantization (PCM). Consequently, using the vector quantizer as a building block and imbedding it with other waveform data compression techniques may lead to the development of a new and powerful class of waveform coding systems. This paper proposes and analyzes a waveform coding system, adaptive vector predictive coding (AVPC), in which a low-dimensionality vector quantizer is used in an adaptive predictive coding scheme. In the encoding process, a locally generated prediction of the current input vector is subtracted from the current vector, and the resulting error vector is coded by a vector quantizer. Each frame consisting of many vectors is classified into one ofmstatistical types. This classification determines which one ofmfixed predictors and ofmvector quantizers will be used for encoding the current frame.