Performance Improvements of Probabilistic Transcript-adapted ASR with Recurrent Neural Network and Language-specific Constraints

TL;DR

This paper presents two techniques using recurrent neural networks and language-specific constraints to refine probabilistic transcriptions, significantly improving the accuracy of cross-lingual ASR systems for non-native speakers.

Contribution

It introduces a noisy-channel model trained with RNNs and applies language-dependent constraints to enhance probabilistic transcriptions for better ASR adaptation.

Findings

Reduced phone error rate by 7% with the RNN model.

Achieved a 9% reduction in phone error rate using pronunciation constraints.

Both methods improve transcription quality and ASR performance.

Abstract

Mismatched transcriptions have been proposed as a mean to acquire probabilistic transcriptions from non-native speakers of a language.Prior work has demonstrated the value of these transcriptions by successfully adapting cross-lingual ASR systems for different tar-get languages. In this work, we describe two techniques to refine these probabilistic transcriptions: a noisy-channel model of non-native phone misperception is trained using a recurrent neural net-work, and decoded using minimally-resourced language-dependent pronunciation constraints. Both innovations improve quality of the transcript, and both innovations reduce phone error rate of a trainedASR, by 7% and 9% respectively