De-Noising of a Room Impulse Response by Applying Wavelets
Ðorðe M. Damnjanovič
Dejan G. Čirič
Bratislav B. Predič
Abstract
One of the techniques most commonly used for de-noising of audio signals, images or speech is wavelet denoising. However, with regard to de-noising of room impulse responses (RIRs), the usage of wavelets can be hardly found in literature. This paper explores the potentials of applying
wavelets for the reduction of negative effects of noise in these responses, since noise represents a serious problem in RIRs. The analysis is based on implementing various wavelets on both synthesized and measured RIRs containing different noises. The influence of changing the parameters of
wavelets is observed in order to find an optimal set of parameters that will provide the best de-noising results. The dynamic range improvement of the backward integrated RIR as well as the noise floor reduction are used as control measures to assess the wavelet de-noising quality. The results
show that the noise floor can be reduced and dynamic range can be improved significantly (e.g., by more than 10 dB) by applying the optimal set of wavelet parameters. The wavelet de-noising is compared with some other noise compensation methods using the estimated reverberation time as a control
measure. In addition to de-noising, wavelets seem to be promising for detection of the truncation time of a RIR.