Attention-Based End-to-End Speech Recognition on Voice Search

TL;DR

This paper presents an attention-based end-to-end speech recognition model for Mandarin voice search, achieving low error rates through novel techniques like character embeddings and attention smoothing.

Contribution

The study introduces effective training tricks and attention mechanisms to improve Mandarin speech recognition with end-to-end models, overcoming language-specific challenges.

Findings

Achieved CER of 3.58% and SER of 7.43% without language model.

Improved CER to 2.81% and SER to 5.77% with a trigram language model.

Demonstrated effectiveness of attention smoothing and character embeddings.

Abstract

Recently, there has been a growing interest in end-to-end speech recognition that directly transcribes speech to text without any predefined alignments. In this paper, we explore the use of attention-based encoder-decoder model for Mandarin speech recognition on a voice search task. Previous attempts have shown that applying attention-based encoder-decoder to Mandarin speech recognition was quite difficult due to the logographic orthography of Mandarin, the large vocabulary and the conditional dependency of the attention model. In this paper, we use character embedding to deal with the large vocabulary. Several tricks are used for effective model training, including L2 regularization, Gaussian weight noise and frame skipping. We compare two attention mechanisms and use attention smoothing to cover long context in the attention model. Taken together, these tricks allow us to finally achieve a character error rate (CER) of 3.58% and a sentence error rate (SER) of 7.43% on the MiTV voice search dataset. While together with a trigram language model, CER and SER reach 2.81% and 5.77%, respectively.