CIF-T: A Novel CIF-based Transducer Architecture for Automatic Speech Recognition

TL;DR

This paper introduces CIF-T, a new speech recognition model combining CIF mechanism with RNN-T to reduce computation and enhance predictor network role, achieving state-of-the-art results on multiple datasets.

Contribution

The paper presents CIF-T, a novel architecture that replaces RNN-T loss with CIF-based alignment, reducing computation and improving predictor network utilization.

Findings

CIF-T achieves state-of-the-art accuracy on AISHELL-1 and WenetSpeech datasets.

CIF-T reduces computational overhead compared to traditional RNN-T models.

The proposed enhancements improve speech recognition performance.

Abstract

RNN-T models are widely used in ASR, which rely on the RNN-T loss to achieve length alignment between input audio and target sequence. However, the implementation complexity and the alignment-based optimization target of RNN-T loss lead to computational redundancy and a reduced role for predictor network, respectively. In this paper, we propose a novel model named CIF-Transducer (CIF-T) which incorporates the Continuous Integrate-and-Fire (CIF) mechanism with the RNN-T model to achieve efficient alignment. In this way, the RNN-T loss is abandoned, thus bringing a computational reduction and allowing the predictor network a more significant role. We also introduce Funnel-CIF, Context Blocks, Unified Gating and Bilinear Pooling joint network, and auxiliary training strategy to further improve performance. Experiments on the 178-hour AISHELL-1 and 10000-hour WenetSpeech datasets show that CIF-T achieves state-of-the-art results with lower computational overhead compared to RNN-T models.