KALMAN SPEECH ENHANCEMENT BASED ON IDENTIFICATION TECHNIQUES

Abstract : Speech enhancement can play a key role when aiming at friendly exchanging through the net. In that context, increasing the signal to noise ratio can be done through the use of Kalman filtering. However, Kalman filtering based-approaches usually require the explicit knowledge of the noise variances and the speech model parameters. In this paper, we propose to consider alternative approaches. Speech enhancement can be considered as an identification issue. For such a purpose, we take advantage of various results from the control framework. We present three new approaches based on subspace methods for identification. They have the advantage of minimizing the number of parameters to be estimated and avoid the use of a voice activity detector. Here, some results are proposed when speech is contaminated by additive white noise
Complete list of metadatas

https://hal.archives-ouvertes.fr/hal-00167772
Contributor : Eric Grivel <>
Submitted on : Wednesday, August 22, 2007 - 4:47:38 PM
Last modification on : Thursday, January 11, 2018 - 6:21:07 AM

Identifiers

  • HAL Id : hal-00167772, version 1

Citation

Eric Grivel, Mohamed Najim. KALMAN SPEECH ENHANCEMENT BASED ON IDENTIFICATION TECHNIQUES. Friendly Exchanging through the net", COST 254, 2000, Bordeaux, France. pp. ⟨hal-00167772⟩

Share

Metrics

Record views

123