Débruitage multicapteur appliqué à la téléphonie mains-libres en automobile

Abstract : Hands-free car kits have gained a lot of popularity among drivers over the last years. These equipments fill the need for the users to communicate while driving their car, whether it is for professional or personal use. Having a phone conversation in a car is really challenging, as the inside of an automobile is a strongly adverse acoustic environment. Indeed, this compartment is strongly reverberant, because of the presence of important glass surface (such as the windbreaker and windows), and it is also very noisy. The noise comes from various sources, such as the engine, the contact of the tire on the road, the wind..., and those sources show different characteristics from one situation to another. Hence, the noise reduction for in-car voice pickup is a major element for the user’s comfort. The main objective of the work reported in this thesis is to build an efficient noise reduction solution for in-car telephony, using a plurality of microphones. We are in this work mostly interested in the freeway situation. Hence, a database of measurements has been recorded in a real car interior, to understand what are the spectral and spatial characteristics of the noise field in this situation. These measurements showed that the noise field has different characteristics in high and low frequencies. Indeed, the noise has more energy, and is more spatially coherent in the low frequency range. This leads us to propose hybrid subband systems, in order to use different algorithmic approaches in high and low frequencies. We made an adaptive implementation of the well-known Minimum Variance Distortionless Response beamforming, which is efficient when the input Signal-to-Noise Ratio is high, and the noise field shows a low spatial coherence. We also conducted an analysis on the impact of sensors’ positions, in order to build a microphone array which will allow this method to be efficient in the high frequency range. We also considered two different processings for the low frequency range : — one is based on Adaptive Noise Cancellation, which uses the high spatial coherence of recorded noises, — the other is based on Multichannel Wiener Filter. For both methods, an analysis of the impact of sensors’ positions has been made, in order to build an efficient microphone array. To assess the performance of these hybrid systems, a subjective evaluation has been conducted, through a listening test. This evaluation shows that the hybrid system using a Multichannel Wiener Filter in the low frequency range supresses a significant amount of noise, while keeping the voice distortion to a minimum, perceptually unnoticed.
Complete list of metadatas

https://pastel.archives-ouvertes.fr/tel-01156542
Contributor : Abes Star <>
Submitted on : Wednesday, May 27, 2015 - 3:22:21 PM
Last modification on : Friday, May 17, 2019 - 12:40:58 PM
Long-term archiving on : Tuesday, September 15, 2015 - 7:02:16 AM

File

TheseFoxV2.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-01156542, version 1

Collections

Citation

Charles Fox. Débruitage multicapteur appliqué à la téléphonie mains-libres en automobile. Traitement du signal et de l'image [eess.SP]. Télécom ParisTech, 2013. Français. ⟨NNT : 2013ENST0074⟩. ⟨tel-01156542⟩

Share

Metrics

Record views

431

Files downloads

296