Self-supervised language learning from raw audio: Lessons from the Zero Resource Speech Challenge
This work addresses the challenge of developing speech technology for low-resource languages or scenarios where expert annotations are unavailable, though it is incremental as it reviews existing efforts.
The paper tackles the problem of building speech processing systems from raw audio without textual labels by reviewing the Zero Resource Speech Challenge series, which defines tasks and benchmarks for model comparison, leading to cumulative progress in self-supervised learning.
Recent progress in self-supervised or unsupervised machine learning has opened the possibility of building a full speech processing system from raw audio without using any textual representations or expert labels such as phonemes, dictionaries or parse trees. The contribution of the Zero Resource Speech Challenge series since 2015 has been to break down this long-term objective into four well-defined tasks -- Acoustic Unit Discovery, Spoken Term Discovery, Discrete Resynthesis, and Spoken Language Modeling -- and introduce associated metrics and benchmarks enabling model comparison and cumulative progress. We present an overview of the six editions of this challenge series since 2015, discuss the lessons learned, and outline the areas which need more work or give puzzling results.