Domain Robust Feature Extraction For Rapid Low Resource ASR Development

Abstract

Developing a practical speech recognizer for a low resource language is challenging, not only because of the (potentially unknown) properties of the language, but also because test data may not be from the same domain as the available training data. In this paper, we focus on the latter challenge, i.e. domain mismatch, for systems trained using a sequence-based criterion. We demonstrate the effectiveness of using a pre-trained English recognizer, which is robust to such mismatched conditions, as a domain normalizing feature extractor on a low resource language. In our example, we use Turkish Conversational Speech and Broadcast News data. This enables rapid development of speech recognizers for new languages which can easily adapt to any domain. Testing in various cross-domain scenarios, we achieve relative improvements of around 25% in phoneme error rate, with improvements being around 50% for some domains.

Stats

• citations9
• S2 citations—
• github stars0
• HF likes0
• heat score7.50
• arxiv keydalmia2018domain

Related papers

• Audio-based Linguistic Feature Extraction For Enhancing Multi-lingual And Low-resource Text-to-speech (2024)0.00
• Extracting Domain Invariant Features By Unsupervised Learning For Robust Automatic Speech Recognition (2018)9.03
• Exploring End-to-end Techniques For Low-resource Speech Recognition (2018)5.84
• Sequence-based Multi-lingual Low Resource Speech Recognition (2018)12.40
• Toward Domain-invariant Speech Recognition Via Large Scale Training (2018)13.39
• Domain Adaptation Of Low-resource Target-domain Models Using Well-trained ASR Conformer Models (2022)4.52
• Spoken Term Detection Methods For Sparse Transcription In Very Low-resource Settings (2021)0.00
• Zero Shot Text To Speech Augmentation For Automatic Speech Recognition On Low-resource Accented Speech Corpora (2024)2.26