This paper presents the complete design of an audio data hiding system destined to transmit a binary information via the audio communication channel for audio annotation applications. The proposed system is based on an innovative embedding strategy. It consists of 1) a new inaudibility control procedure that locally regulates the watermark transparency, 2) an informed embedding function that maximizes system robustness to additive channel perturbation by using a new criterion of robustness - that is by maintaining the error probability at a fixed value, and 3) an efficient and low computational cost synchronization mechanism. System performance in terms of inaudibility of the watermark, transmission reliability with respect to various perturbations, and computational cost is evaluated on real audio signals to determine the efficiency of the proposed embedding strategy

A particular application of audio data hiding systems and watermarking systems consists of using the audio signal as a transmission channel for binary information. The system should ensure a reliable and robust transmission for various channel perturbations but also propose a low computational cost for real-time applications. In this paper, we present a hybrid spread-spectrum data hiding system, which combines two reference systems taken from the State-Of-The-Art: the one based on a real-time receiver and the other one based on an informed embedding strategy with maximized robustness to additive perturbations. Experimental results permit to assess the efficiency of the system in terms of: (1) transmission reliability, which is significantly improved compared to reference systems, and (2) computational costs, which allows for the feasible real-time reception process of broadcast applications with off-line embedding.

This study compares four phone-based spoken document retrieval (SDR) approaches. In all cases, the indexing and retrieval system uses phonetic information only. The first retrieval method is based on the vector space model, using phone 3-grams as indexing terms. This approach is compared with 2 string-matching methods. A fourth method, combining the VSM approach with the slot detection step of stringmatching techniques is proposed. This method is tested on a collection of short German spoken documents, using three different sets of queries: text queries, clean spoken queries and noisy spoken queries.

A particular application of audio watermarking systems consists in using the audio signal as a transmission channel for binary information. In this context, reliability criteria with respect to transmission rate, such as bit error rate (BER), is a major issue that defines system performance. Current research has already proven the advantage of using informed embedding rather than blind embedding. In this paper, we present a new informed watermarking system based on an embedding strategy that maximizes robustness to perturbations such as compression and analog transmission. The proposed system performance is then compared to the equivalent blind embedding system performance. Experimental results permit us to assess the achieved efficiency. For transmission rates, up to 500 bits/s, BER are divided by 10 when channel perturbation is compression and divided by 4 in the case of analog transmission.

The paper presents a new detection method for a spread spectrum and perceptual watermarking system, viewed as a hidden data transmission system, where the generic detection operation is achieved by a Wiener deconvolution filter. We point out the insufficiencies of the generic reception scheme concerning truncation errors and an ill-conditioned deconvolution operation, and we propose a cascade realization of the reception filter, which significantly improves the detection performances (to a multiplicative factor of 20) at higher bit rates, even in the presence of MPEG compression.

This paper presents a phone-based approach of spoken document retrieval (SDR), developed in the framework of the emerging MPEG-7 standard. We describe an indexing and retrieval system that uses phonetic information only. The retrieval method is based on the vector space IR model, using phone N-grams as indexing terms. We propose a technique to expand the representation of documents by means of phone confusion probabilities in order to improve the retrieval performance. This method is tested on a collection of short German spoken documents, using 10 city names as queries.

In this paper, we present an MPEG-7-based audio classification and retrieval technique targeted for analysis of film material. The technique consists of low-level descriptors and high-level description schemes. For low-level descriptors, low-dimensional features such as audio spectrum projection based on audio spectrum basis descriptors is produced in order to find a balanced tradeoff between reducing dimensionality and retaining maximum information content. High-level description schemes are used to describe the modeling of reduced-dimension features, the procedure of audio classification, and retrieval. A classifier based on continuous hidden Markov models is applied. The sound model state path, which is selected according to the maximum-likelihood model, is stored in an MPEG-7 sound database and used as an index for query applications. Various experiments are presented where the speaker- and sound-recognition rates are compared for different feature extraction methods. Using independent component analysis, we achieved better results than normalized audio spectrum envelope and principal component analysis in a speaker recognition system. In audio classification experiments, audio sounds are classified into selected sound classes in real time with an accuracy of 96%.

Our purpose is to evaluate the efficiency of MPEG-7 audio descriptors for speaker recognition. The upcoming MPEG-7 standard provides audio feature descriptors, which are useful for many applications. One example application is a speaker recognition system, in which reduced-dimension log-spectral features based on MPEG-7 descriptors are used to train hidden Markov models for individual speakers. The feature extraction based on MPEG-7 descriptors consists of three main stages: Normalized Audio Spectrum Envelope (NASE), Principal Component Analysis (PCA) and Independent Component Analysis (ICA). An experimental study is presented where the speaker recognition rates are compared for different feature extraction methods. Using ICA, we achieved better results than NASE and PCA in a speaker recognition system.

This paper presents a speech enhancement method for noise robust front-end and speech reconstruction at the back-end of Distributed Speech Recognition (DSR). The speech noise removal algorithm is based on a two stage noise filtering LSAHT by log spectral amplitude speech estimator (LSA) and harmonic tunneling (HT) prior to feature extraction. The noise reduced features are transmitted with some parameters, viz., pitch period, the number of harmonic peaks from the mobile terminal to the server along noise-robust mel-frequency cepstral coefficients. Speech reconstruction at the back end is achieved by sinusoidal speech representation. Finally, the performance of the system is measured by the segmental signal-noise ratio, MOS tests, and the recognition accuracy of an Automatic Speech Recognition (ASR) in comparison to other noise reduction methods.

An audio signal parametric modeling scheme is proposed that permits higher performance for representing strong sound transients. The exponentially damped sinusoids (EDS) model is considered in association with a high resolution parameter estimation approach. Such a technique is well adapted to almost every audio signal but is unfortunately not efficient when dealing with signals presenting strong temporal variations, such as percussive music signals, and causes pre-echo artifacts and weak onset dynamic reproduction which are prejudicial to listening. A system, based on the EDS model, has been developed with a transient detector and dynamic time segmentation and modeling that allows to overcome such artifacts.

This paper presents the application of the promising public key watermarking method to the audio domain. Its detection process does not need the original content nor the secret key used in the embedding process. It is the translation, in the watermarking domain, of a public key pair cryptosystem. We start to build the detector with some basic assumptions. This leads to a hypothesis test based on probability likelihood. But real audio signals do not satisfy the assumption of a Gaussian probability density function. Moreover, the use of an advanced human perception model to hide the watermark makes the detection issue a tough problem. Our works result in a new detection process offering a good test's power for a low probability of false alarm.

This paper deals with the problem of incorrect data rejection in a large vocabulary directory task. Two different strategies are investigated to improve the rejection of noises and OOV data. An incremental adaptation algorithm is first proposed to adapt word models and a garbage model to field data. The second method consists in post-processing the recogniser hypotheses by computing for each of them a confidence measure based on frame level likelihood ratios. Both methods yield a noticeable reduction in the false alarm rate on noises and OOV data. Their combination leads to a further false alarm rate reduction.

This paper describes a full-rate (FR) candidate speech codec for a new Adaptive Multi-Rate (AMR) codec standard. The proposed codec is a multi-rate codec, operating in various modes that have different bit-rate partitionings between source and channel coding. The source coding is based on Algebraic Code Excited Linear Predictive Coding (ACELP). Convolutional coding is used for the channel coding. Subjective tests show that at poor channel error conditions with low C/I-ratios of 7 dB, the lowest mode of the proposed codec produces speech quality comparable to G.728 in clean channel condition. The codec also provides substantial improvement on error robustness over the Enhanced Full Rate (EFR) GSM codec, by selecting an optimal coding rate to suit the channel condition. The codec performance passes twenty-six out of the twenty-nine test conditions, conducted according to [2]. Therefore, the proposed codec has passed the ETSI qualification test.

A 64 kbps coder of wideband (15 kHz) monophonic audio signals is described. Its structure is based on the transform coded excitation scheme, adopted to 7 kHz band signals. Significant modifications are proposed, that yield reduction of delay while keeping an almost transparent quality of speech and music, equivalent to that provided by the MPEG-1, layer II audio standard at the same bit rate. Algorithmic delay has been reduced to 17 ms-approximately 1/3 the delay of the MPEG coder.

We describe a systematic procedure to replace vector quantization (VQ) with trellis-coded vector quantization (TCVQ) in existing CELP coders. Following this procedure, we design an 8/16 kbit/s TCVQ CELP coder. We analyze the performance of this method in terms of quality and complexity. Our results show that a CELP coder using TCVQ produces significantly better quality than the same coder using VQ, with reasonable complexity. By modifying the TCVQ CELP coder parameters one can favorably trade coding quality against complexity and/or delay.

We analyze the performance of a CELP coder where the vector quantization (VQ) of the excitation is replaced with trellis-coded vector quantization (TCVQ). Our results show that TCVQ performs significantly better than VQ, with reasonable complexity. This makes TCVQ a fair choice for trading quality against complexity and/or delay. We describe a systematic procedure to replace VQ with TCVQ for existing CELP coders. We propose an optimization algorithm to appropriately populate the trellis. We show how pseudo-Gray coding can be applied to the TCVQ codebook to improve intrinsic coder robustness to channel errors. Finally, we evaluate the complexity and performance of the method.

The authors investigate several algorithms that construct the input for the synthesis filter in the CELP coder, they present them under the same formalism, and compare their performances. They model the excitation vector by a linear combination of K signals, which are issued from K codebooks and multiplied by K associated gains. They demonstrate that this generalized form incorporates several particular coders such as code excited linear predictive coders, multipulse coders, self excited vocoders, etc. The least squares minimization problem is presented afterwards. In the case of orthogonal codebooks, they show that the optimal solution of this least squares problem is equivalent to orthogonal transform coding. They use the Karhunen-Loeve transform to design the corresponding orthogonal codebooks. In the case of nonorthogonal codebooks, they are restricted to suboptimal iterative algorithms for index selection and gain computation. They present some new algorithms based on orthogonalization procedures and QR factorizations that attempt to reduce this suboptimality. In a particular case, when the excitation is modeled using one gain coefficient (for example, ternary excitation or concatenation of short codebook vectors), an iterative angle minimization algorithm is proposed for index selection. The different extraction algorithms are compared with regard to the resulting coder complexity and synthetic speech quality. They find a particularly attractive method that consists of modeling the excitation with one unique gain.>

The authors investigate three algorithms that orthogonalize codebooks in a multi-stage code excited linear prediction (CELP) speech coder. They carry out the same processing, a locally optimal modeling of the perceptual signal, but the computational costs differ. The authors show that the recursive modified Gram-Schmidt algorithm extra computational cost is less than the other two. An orthogonal codebook is defined a priori and the authors observe an equivalence to orthogonal transform coding. Three methods based on the Karhunen-Loeve transform for designing this codebook are compared. A partitioned shape-gain VQ is applied in the transform domain.>

A generalized coder that includes all types of excitation is presented. In this analysis-by-synthesis scheme, maximization formulae are the same regardless of the kind of excitation. The total excitation is expressed as a linear combination of excitation vectors. Given the number of excitation vectors or, equivalently, the bit rate of the coder, finding the vectors and their corresponding gains is a specific least-squares problem. The standard way this problem is usually solved is given, and three alternative mathematical approaches are proposed. These approaches are a Gram-Schmidt orthogonalization, a Choleski decomposition, a Householder transform. All these procedure have the same geometrical interpretation and lead to the same floating point simulation results.>

This paper describes a custom integrated circuit designed to perform LPC-based feature measurements. The chip is decomposed in 4 operators, working concurrently, implementing the functions of preem-phasizing, correlation, Parcor extraction and filtering. By using bit-serial architecture, each block can be adapted to the computation requirement. The projected chip complexity is approximately 25000 transistors (9 mm2 without I/O pads in 2µ NMOS technology) plus 2640 bits of RAM. The maximum sampling frequency is about 25 kHz which is sufficient for speech coding.

The problem of determining sparse MA models has received much attention in recent years and is of fundamental importance in various application areas such as speech (multi-pulse excitation) or seismic data (wavelet time of arrival). This paper addresses the problem of selection and identification of non-zero coefficients in the MA models (pulse position and amplitude). The selection is done globally in the Fourier transform domain using a (complex) Pisarenko procedure, instead of sequentially. Moreover, the pulses being frequently placed in contiguous locations as a short solvo, a new MA identification method is proposed for this special case. This method only uses the AR model coefficients and the prediction residual as entries.

This paper discusses two custom integrated circuits designed to perform the functions of signal correlation and lattice filtering (MA or AR). Each circuit is decomposed into P operators, each being a direct implementation of the equations. To allow concurrent use of an arbitrary number of operators and to simplify inter-module connections (both within and between chips), a bit-serial architecture was adopted. These chips can be cascaded; computation speed is independent of model order in both types of calculations. These chips have been designed to operate at a sample frequency between 0 and 300 kHz for the correlator, 0 and 150 kHz for the lattice filter.

A general framework for deriving and interpreting analysis and synthesis spectra of the autoregressive (AR) and moving average (MA) type is presented. Investigation of AR linear transformations of finite dimensional data records yields a set of intermediate MA techniques associated with approximation of the inverse correlation matrix R-1. The corresponding spectrum we call a parameterized maximum likelihood method (pMLM) spectrum. Investigation of MA linear transformations yields a set of intermediate MA techniques associated with approximation of the correlation matrix R. The corresponding spectrum we call a parameterized Bartlett spectrum (pBA). Simulations on synthetic AR, MA and ARMA data sets illustrate the techniques and lead to interesting remarks concerning the use of parameterizations of R and R-1to differentiate between data sets of AR and MA type.