Serialized Output Training by Learned Dominance

TL;DR

This paper introduces a model-based serialization strategy for multi-talker speech recognition that autonomously orders speech components, outperforming previous permutation-invariant methods on standard datasets.

Contribution

The study proposes an auxiliary module within the Attention Encoder-Decoder that identifies and orders speech components based on dominance, addressing label-permutation issues.

Findings

Significantly outperforms PIT and FIFO baselines on LibriSpeech and LibriMix.

The serialization module identifies dominant speakers by loudness and gender.

Effective in 2-mix and 3-mix scenarios.

Abstract

Serialized Output Training (SOT) has showcased state-of-the-art performance in multi-talker speech recognition by sequentially decoding the speech of individual speakers. To address the challenging label-permutation issue, prior methods have relied on either the Permutation Invariant Training (PIT) or the time-based First-In-First-Out (FIFO) rule. This study presents a model-based serialization strategy that incorporates an auxiliary module into the Attention Encoder-Decoder architecture, autonomously identifying the crucial factors to order the output sequence of the speech components in multi-talker speech. Experiments conducted on the LibriSpeech and LibriMix databases reveal that our approach significantly outperforms the PIT and FIFO baselines in both 2-mix and 3-mix scenarios. Further analysis shows that the serialization module identifies dominant speech components in a mixture by factors including loudness and gender, and orders speech components based on the dominance score.