No abstract

This paper describes the details of a bilingual speech recognition system, AmritaRec, developed for English and Tamil. The performance results of the system is compared with that of a monolingual English speech recognition system adapted to Tamil using cross language transfer and cross language adaptation techniques.

In this paper an automated method to recognize the musical instruments playing the musical signals is presented. Various features of the musical instruments and musical signals are investigated. The features can broadly be grouped into three categories: temporal, spectral, and cepstral features. A composite neural network structure is proposed as the classifier. The performance of the composite neural network using a set of carefully chosen features is compared with that of the traditional neural network. Experimental results show that the accuracy achieved using composite structure (94%) is significantly higher than that using the traditional structure (88%) when more than four musical instruments are to be distinguished.

A non-blind two-channel time-frequency digital bits audio watermarking scheme with error-correcting code is described in this paper. The proposed method operates by encoding the watermark bits with cyclic code before embedding them into the audio signal. Time-frequency compressionexpansion technique is used to embed the watermark bits. The coefficients to be deleted or added for the time-frequency compression-expansion technique are determined using psychoacoustic model. Both channels of the stereo audio signal are used for watermark embedding. This combination of cycliccode and two-channel approach using the robust time-frequency technique of coding watermark bits has resulted in perfect recovery of watermark under attacks.

No abstract