Feature Extraction from Non-Audible Murmur (NAM) for the Vocally Handicapped using Wavelet Transform

Non audible murmur is a body conducted silent speech through which the vocally handicapped can communicate. We propose a method of acquisition of Non Audible Murmur (NAM), (i.e., inaudible speech produced without vibrations of the vocal folds) from the vocally handicapped using the MEMS accelerometer, followed by its de-noising and Statistical Feature Extraction. The murmur is acquired by placing the sensor bonded to the surface of the skin over the soft-cartilage bone behind the ear. The resulting electrical signal is de-noised using Discrete Wavelet Transform (DWT). Statistical Features are extracted from the detailed coefficients of the de-noised murmur. General Terms Speech signal processing, Wavelet Transform