WAVELET PACKET ANALYSIS OF BODY-CONDUCTED MURMUR FOR THE VOCALLY HANDICAPPED Sheena Christabel Pravin 1 , Samyuktha Sundar 2 , Krithika Aravindan 2 1 Assistant Professor, 2 Student, Rajalakshmi Engineering College Email id: sheenachrsitabelpravin@gmail.com, samyukthasundar22@gmail.com, krithika.aravindan7@gmail.com Abstract- The Non-Audible Murmur (NAM) is a body- conducted speech that can aid the communication of the vocally handicapped. The body conducted murmur is acquired by placing the MEMS vibration sensor on the skin covering the soft-cartilage bone behind the ear lobe. The recorded NAM is de-noised using Wavelet Packet Transform (WPT) in three stages. The Symlet-8 is the wavelet chosen for de-noising the acquired murmur signal with decimated inputs to the Digital Filter Bank (DFB). Features are extracted from the de-noised murmur by Statistical Analysis. Keyword: NAM, WPT, DFB, Statistical Analysis I. INTRODUCTION Today, when growth in technology is reaching new heights, technology for the needy is the need of the hour. Speech processing for the vocally handicapped is one such area with tremendous scope for research and development. The vocally handicapped are unable to produce well-defined speech; however, they are able to produce murmurs, which are sometimes inaudible to the human ear. An alternative to communicate is the interpretation of their murmur into intelligible text which can be fed to a speech synthesizer. This would be a boon to the vocally handicapped. Silent speech interfaces have recently been studied as a suitable technology to enable speech communication to take place without the necessity of emitting an audible acoustic signal [1]. A serious drawback of body-conducted speech is that severe degradation of speech quality is caused by essential mechanisms of body conduction. Consequently, naturalness and intelligibility of body- conducted speech are much poorer than those of natural voices. To use it in human-to-human speech communication, its speech quality improvements are essential l[2]. To enhance body-conducted speech, it should be de-noised for feature extraction, which is carried out in this paper. Fourier analysis, using the Fourier transform, is a powerful tool for analyzing the components of a stationary signal (a stationary signal is a signal that repeats). For example, the Fourier transform is a powerful tool for processing signals that are composed of some combination of sine and cosine signals[3]. The Fourier transform is less useful in analyzing non-stationary data, where there is no repetition within the region sampled. Having in mind the limitations of the Fourier Transform (poor time localization) and of the Short-Time Fourier Transform (fixed time and frequency localization), Grossman and Morlet gave in 1984 the formulation of the Continuous Wavelet Transform. Unlike the first two, that were decomposing the signal into a basis of complex exponentials, the Wavelet Transform decomposes the signal over a set of dilated and translated wavelets. Wavelet transforms (of which there are, at least formally, an infinite number) allow the components of a non-stationary signal to be analyzed. Wavelets also allow filters to be constructed for stationary and non-stationary signals. This paper looks at acquiring this Non-Audible Murmur(NAM) as body conducted vibrations and de- noising the murmur using Wavelet Packet Transform. The approximate and detail coefficients from the Wavelet analysis aids in feature extraction from the murmur signal. II. NON-AUDIBLE MURMUR ACQUISITION Murmur samples are taken from vocally handicapped students using a tri-axial accelerometer – ADXL335. It is a MEMS vibration sensor. Figure1. Experimental setup