Paper: Unsupervised Learning of Acoustic Sub-word Units

ACL ID P08-2042
Title Unsupervised Learning of Acoustic Sub-word Units
Venue Annual Meeting of the Association of Computational Linguistics
Session Main Conference
Year 2008

Accurate unsupervised learning of phonemes of a language directly from speech is demon- strated via an algorithm for joint unsupervised learning of the topology and parameters of a hidden Markov model (HMM); states and short state-sequences through this HMM cor- respond to the learnt sub-word units. The algorithm, originally proposed for unsuper- vised learning of allophonic variations within a given phoneme set, has been adapted to learn without any knowledge of the phonemes. An evaluation methodology is also proposed, whereby the state-sequence that aligns to a test utterance is transduced in an auto- matic manner to a phoneme-sequence and compared to its manual transcription. Over 85% phoneme recognition accuracy is demon- strated for speaker-dependent learning from fluent, large-vocabu...