Speech enhancement for forensic applications using dynamic time warping and wavelet packet analysis

This paper presents two novel speech enhancement techniques which are applicable to forensic audio recordings. The first technique is one for removing a background signal from a single channel recording in order to obtain the target signal. The method assumes that it is possible to obtain a copy of the background signal at a later instance. The copy of the background signal is aligned onto the single channel recording and then it is used in the cancellation process. The second technique may be used for removing a varying tone from a signal and is based on wavelet packet analysis. Wavelet analysis isolates the varying tones in particular wavelet packets which are located using regression analysis and then suppressed using an attenuation function.