André Goalic

On the Recognition of Cochlear Implant-Like Spectrally Reduced Speech With MFCC and HMM-Based ASR

This correspondence investigates the recognition of cochlear implant-like spectrally reduced speech (SRS) using mel frequency cepstral coefficient (MFCC) and hidden Markov model (HMM)-based automatic speech recognition (ASR). The SRS was synthesized from subband temporal envelopes extracted from original clean test speech, whereas the acoustic models were trained on a different set of original clean speech signals of the same speech database. It was shown that changing the bandwidth of the subband temporal envelopes had no significant effect on the ASR word accuracy. In addition, increasing the number of frequency subbands of the SRS from 4 to 16 improved significantly the system performance. Furthermore, the ASR word accuracy attained with the original clean speech can be achieved by using the 16-, 24-, or 32-subband SRS. The experiments were carried out by using the TI-digits speech database and the HTK speech recognition toolkit.

A novel framework for noise robust ASR using cochlear implant-like spectrally reduced speech

We propose a novel framework for noise robust automatic speech recognition (ASR) based on cochlear implant-like spectrally reduced speech (SRS). Two experimental protocols (EPs) are proposed in order to clarify the advantage of using SRS for noise robust ASR. These two EPs assess the SRS in both the training and testing environments. Speech enhancement was used in one of two EPs to improve the quality of testing speech. In training, SRS is synthesized from original clean speech whereas in testing, SRS is synthesized directly from noisy speech or from enhanced speech signals. The synthesized SRS is recognized with the ASR systems trained on SRS signals, with the same synthesis parameters. Experiments show that the ASR results, in terms of word accuracy, calculated with ASR systems using SRS, are significantly improved compared to the baseline non-SRS ASR systems. We propose also a measure of the training and testing mismatch based on the Kullback-Leibler divergence. The numerical results show that using the SRS in ASR systems helps in reducing significantly the training and testing mismatch due to environmental noise. The training of the HMM-based ASR systems and the recognition tests were performed by using the HTK toolkit and the Aurora 2 speech database.

On Normalized MSE Analysis of Speech Fundamental Frequency in the Cochlear Implant-Like Spectrally Reduced Speech

In this paper, we present a quantitative study on the speech fundamental frequency (F0) of the cochlear implant-like spectrally reduced speech (SRS). The SRS was synthesized from the subband amplitude and frequency modulations (AM and FM) of original clean speech utterances selected from the TI-digits database. The SRS synthesis algorithm was derived from the frequency amplitude modulation encoding (FAME) strategy, proposed by Nie et al., 2005. The normalized mses (NMSEs), calculated between the F0 of the original clean speech and that of the SRSs, were analyzed. The NMSEs analysis of F0 revealed the greater F0 distortion in the AM-based SRS, which is the acoustic simulation of present-day cochlear implants, compared to the FAME-based SRS. This evidence supports the fact that current cochlear implant users have difficulty in the speaker recognition task as reported by Zeng et al., 2005. Further, the analysis results showed that it is better to keep the rapidly varying FM components to reduce the F0 distortion in the FAME-based SRS at low spectral resolution.

1995-2005: ten years of active research on underwater acoustic communications in Brest

On the last ten years, GESMA and its main partners, Sercel Brest and ENST Bretagne realized different projects related to underwater acoustic communication. The underwater acoustic channel is so harsh that different strategies have to be considered according to the desired application. From point to point to network and from short to long range, this paper aims to recall these different developments and most significant results.

Potential applications of low complexity non-binary high code rate block turbo codes

High code rate block turbo codes (BTC) using Bose-Chaudhuri-Hocquenghem (BCH) codes have already demonstrated near Shannon performances for quadrature phase-shift keying (QPSK) over additive white Gaussian noise (AWGN) channel. We show here that very low complexity Reed-Solomon (RS) BTC can also achieve reliable transmission at less than 1 dB from Shannon limit with the same transmission conditions. Performances of this RS-BTC over binary symmetric channel (BSC) and binary erasure channel (BEC) are also investigated in this paper. Potential applications for this study are high speed optical transmission and data storage.

Long range real-time underwater acoustic communication at low bit rate with channel coding protection

The TRIDENT project was launched a few years ago by GESMA. The First objectives were to develop a multiple rate underwater acoustic link for images, text and data transmission. This link was designed to provide a wireless communication to AUVs . Recently the platform was extended to low bit rate speech transmission using a MELP [1] coder working at 2400 bit/s. This allows us to use a lower carrier frequency able to reach a longer range. In order to increase the link reliability, channel coding was added to the system. Different kinds of error correcting schemes including Convolutional Code (CC) and Reed Solomon codes(RS) were checked during river and sea trials. The purpose of this paper is to present the whole platform including both the emitter and the receiver and an overview of the results during the trials. Among the numerous possible applications, one can note mine warfare, offshore activities with the development of digital acoustic links between divers, boats, submarines etc...or a digital underwater acoustic phone. Such an acoustic link is likely to increase underwater activity safety and also widely develop such activities.