Jerome Boudy

Experiments with a Nonlinear Spectral Subtractor (NSS), Hidden Markov Models and the projection, for robust speech recognition in cars

Abstract Achieving reliable performance for a speech recogniser is an important challenge, especially in the context of mobile telephony applications where the user can access telephone functions through voice. The breakthrough of such a technology is appealing, since the driver can concentrate completely and safely on his task while composing and conversing in a “full” hands-free mode. This paper addresses the problem of speaker-dependent discrete utterance recognition in noise. Special reference is made to the mismatch effects due to the fact that training and testing are made in different environments. A novel technique for noise compensation is proposed: nonlinear spectral subtraction (NSS). Robust variance estimates and robust pdf evaluations (projection) are also introduced and combined with NSS into the HMM framework. We show that the lower limit of applicability of the projection (low SNR values) can be loosened after combination with NSS. Experimental results are reported. The performance of an HMM-based recogniser rises from 56% (no compensation) to 98% after speech enhancement. More than 3300 utterances have been used to evaluate the systems (three databases, two European languages). This result is achieved by the use of robust training/recognition schemes and by preprocessing the noisy speech by NSS.

Non-linear spectral subtraction (NSS) and hidden Markov models for robust speech recognition in car noise environments

The authors address the problem of speaker-dependent discrete utterance recognition in noise. Special reference is made to the mismatch effects due to the fact that training and testing are carried out in different environments. The authors extend their previous work (Lockwood and Boudy, 1991) where a robust hidden Markov model (HMM) training/recognition framework is proposed. Several new aspects are introduced: use of enhanced nonlinear spectral subtraction (NSS) schemes, introduction of root-MFCC parameters, use of dynamic features, and training of HMMs by a dynamic inference scheme (DIHMM). These enhancements are discussed from tests performed on bandlimited signals (200-3000 Hz). The authors show that these various optimizations allow a rise from 20% to over 99% in performance. A 93% recognition rate is already achievable on raw data using a weighted modified projection and a root-MFCC dynamic representation.<<ETX>>

Acoustical echo canceller with sub-band filtering

The echo canceller is designed to be placed between a hands-free acoustical interface and a communications network. It comprises a plurality of processing paths connected in parallel and each allocated to one of a plurality of adjacent sub-bands taken from the spectrum band of the output signal. Each path comprises an analysis filter receiving the echo-containing signal for transmission after correction, a second analysis filter receiving the incoming signal coming from the network, and feeding an adaptive filter that supplies an estimated echo in the respective sub-band to the subtractive input of the subtracter and a synthesis filter. The adaptive filters in at least some of the sub-bands implement a QR decomposition RLS algorithm on the incoming signal, using the fast version thereof, with or without recursive order.

Car noise processing for speech input

The authors address the problem of speech enhancement in a car environment. After describing the method used to characterize the noisy environment, the main effects of the noise on speech processing for speech recognition and transmission are presented. Most of the noise reduction methods proposed in past years have been developed and evaluated in the case of additive white noise. The authors have found that the results obtained by this approach, without modification, are poor. With preprocessing of the noisy signal, improvements are shown, especially for speech recognition.<<ETX>>

Acoustic echo cancellation using a fast QR-RLS algorithm and multirate schemes

High quality acoustic echo cancellation is now required by hands-free systems used in mobile radio and teleconference communications. The demand for fast convergence, good tracking capabilities, and reduced complexity cannot be met by classical adaptive filtering algorithms. In this paper, a new echo canceller using multirate systems and a fast QR-decomposition based RLS algorithm is investigated. Simulation results demonstrate the efficiency of this new combined structure for acoustic echo cancellation, and a fixed-point implementation of the proposed scheme confirms the expected numerical robustness of the fast QR-RLS algorithm.

Controlled convergence of QR least-squares adaptive algorithms-application to speech echo cancellation

With the increasing use of hands-free audio terminals in communication systems, the realization of a high quality hands-free function still remains a challenging research topic. The use of fast QR least-squares algorithms, combined with a multirate structure, have been proposed in Capman et al. (1995) for good performance and reduced complexity. In this paper, we propose to modify QR least-squares adaptive algorithms in order to control the adaptation process in the context of acoustic echo cancellation. This modification is based on a similar idea developed for fast transversal filters (Slock and Kailath 1991; and Benallal and Gilloire 1989). Simulation results demonstrate the efficiency of this modified adaptation process for QR-LS adaptive algorithms.