A study on more realistic room simulation for far-field keyword spotting

Document detail

ID

oai:arXiv.org:2006.02774

Topic

Computer Science - Sound Electrical Engineering and Systems...

Author

Bezzam, Eric Scheibler, Robin Cadoux, Cyril Gisselbrecht, Thibault

Year

2020

listing date

3/31/2022

Keywords

impact study

Metrics

Abstract

We investigate the impact of more realistic room simulation for training far-field keyword spotting systems without fine-tuning on in-domain data.

To this end, we study the impact of incorporating the following factors in the room impulse response (RIR) generation: air absorption, surface- and frequency-dependent coefficients of real materials, and stochastic ray tracing.

Through an ablation study, a wake word task is used to measure the impact of these factors in comparison with a ground-truth set of measured RIRs.

On a hold-out set of re-recordings under clean and noisy far-field conditions, we demonstrate up to $35.8\%$ relative improvement over the commonly-used (single absorption coefficient) image source method.

Source code is made available in the Pyroomacoustics package, allowing others to incorporate these techniques in their work.

;Comment: 7 pages, 4 figures, accepted at APSIPA 2020, room impulse response generation code can be found at https://github.com/ebezzam/room-simulation

Bezzam, Eric,Scheibler, Robin,Cadoux, Cyril,Gisselbrecht, Thibault, 2020, A study on more realistic room simulation for far-field keyword spotting