Librispeech Transducer Model with Internal Language Model Prior Correction

TL;DR

This paper introduces a Librispeech transducer model that effectively combines internal and external language models, achieving significant performance improvements through Bayesian-inspired subtraction of internal LM and better EOS handling.

Contribution

It proposes a novel method to incorporate and subtract the internal language model from the transducer, improving speech recognition accuracy on Librispeech.

Findings

Over 14% relative improvement with internal LM subtraction

Enhanced combination with external LM through separate non-blank probability

Improved performance by including external LM's EOS probability

Abstract

We present our transducer model on Librispeech. We study variants to include an external language model (LM) with shallow fusion and subtract an estimated internal LM. This is justified by a Bayesian interpretation where the transducer model prior is given by the estimated internal LM. The subtraction of the internal LM gives us over 14% relative improvement over normal shallow fusion. Our transducer has a separate probability distribution for the non-blank labels which allows for easier combination with the external LM, and easier estimation of the internal LM. We additionally take care of including the end-of-sentence (EOS) probability of the external LM in the last blank probability which further improves the performance. All our code and setups are published.