Improvements to Embedding-Matching Acoustic-to-Word ASR Using Multiple-Hypothesis Pronunciation-Based Embeddings

TL;DR

This paper enhances embedding-matching acoustic-to-word ASR by introducing multiple hypotheses and pronunciation-based embeddings, significantly improving accuracy especially for dynamic OOV words in digital assistant queries.

Contribution

The paper introduces two novel methods: generating multiple embeddings per instance and using pronunciation-based embeddings, to improve embedding-matching A2W accuracy.

Findings

Up to 18% reduction in word error rate on contact name queries.

Significant accuracy improvements with same training data and model size.

Effective handling of dynamic OOV words in real-world scenarios.

Abstract

In embedding-matching acoustic-to-word (A2W) ASR, every word in the vocabulary is represented by a fixed-dimension embedding vector that can be added or removed independently of the rest of the system. The approach is potentially an elegant solution for the dynamic out-of-vocabulary (OOV) words problem, where speaker- and context-dependent named entities like contact names must be incorporated into the ASR on-the-fly for every speech utterance at testing time. Challenges still remain, however, in improving the overall accuracy of embedding-matching A2W. In this paper, we contribute two methods that improve the accuracy of embedding-matching A2W. First, we propose internally producing multiple embeddings, instead of a single embedding, at each instance in time, which allows the A2W model to propose a richer set of hypotheses over multiple time segments in the audio. Second, we propose using word pronunciation embeddings rather than word orthography embeddings to reduce ambiguities introduced by words that have more than one sound. We show that the above ideas give significant accuracy improvement, with the same training data and nearly identical model size, in scenarios where dynamic OOV words play a crucial role. On a dataset of queries to a speech-based digital assistant that include many user-dependent contact names, we observe up to 18% decrease in word error rate using the proposed improvements.