Speaker conditioning of acoustic models using affine transformation for multi-speaker speech recognition

TL;DR

This paper introduces a speaker conditioning method using affine transformations in acoustic models, significantly improving multi-speaker speech recognition accuracy in overlapping speech scenarios.

Contribution

It proposes a general affine transformation-based speaker conditioning approach applicable to any acoustic model architecture, demonstrated with a ResNet model.

Findings

Achieved +9% relative WER reduction on clean speech

Achieved +20% relative WER reduction on overlapping speech

Effective fusion of speaker info with acoustic features

Abstract

This study addresses the problem of single-channel Automatic Speech Recognition of a target speaker within an overlap speech scenario. In the proposed method, the hidden representations in the acoustic model are modulated by speaker auxiliary information to recognize only the desired speaker. Affine transformation layers are inserted into the acoustic model network to integrate speaker information with the acoustic features. The speaker conditioning process allows the acoustic model to perform computation in the context of target-speaker auxiliary information. The proposed speaker conditioning method is a general approach and can be applied to any acoustic model architecture. Here, we employ speaker conditioning on a ResNet acoustic model. Experiments on the WSJ corpus show that the proposed speaker conditioning method is an effective solution to fuse speaker auxiliary information with acoustic features for multi-speaker speech recognition, achieving +9% and +20% relative WER reduction for clean and overlap speech scenarios, respectively, compared to the original ResNet acoustic model baseline.