Improving Reverberant Speech Training Using Diffuse Acoustic Simulation

Abstract

We present an efficient and realistic geometric acoustic simulation approach for generating and augmenting training data in speech-related machine learning tasks. Our physically-based acoustic simulation method is capable of modeling occlusion, specular and diffuse reflections of sound in complicated acoustic environments, whereas the classical image method can only model specular reflections in simple room settings. We show that by using our synthetic training data, the same neural networks gain significant performance improvement on real test sets in far-field speech recognition by 1.58% and keyword spotting by 21%, without fine-tuning using real impulse responses.

Accepted to ICASSP 2020, impulse response generation code at https://github.com/RoyJames/pygsound

References in corpus (4)

• Regression and Classification for Direction-of-Arrival Estimation with Convolutional Recurrent Neural Networks
• Scene-Aware Audio Rendering via Deep Acoustic Analysis
• Low-frequency Compensated Synthetic Impulse Responses for Improved Far-field Speech Recognition
• Interactive Sound Rendering on Mobile Devices using Ray-Parameterized Reverberation Filters

Cited by in corpus (9)

• Regression and Classification for Direction-of-Arrival Estimation with Convolutional Recurrent Neural Networks
• Scene-Aware Audio Rendering via Deep Acoustic Analysis
• MESH2IR: Neural Acoustic Impulse Response Generator for Complex 3D Scenes
• GWA: A Large High-Quality Acoustic Dataset for Audio Processing
• Low-frequency Compensated Synthetic Impulse Responses for Improved Far-field Speech Recognition
• Sound Synthesis, Propagation, and Rendering: A Survey
• A study on more realistic room simulation for far-field keyword spotting
• Scene-aware Far-field Automatic Speech Recognition
• Improving Reverberant Speech Separation with Multi-stage Training and Curriculum Learning