Sphinx-III Speech Recognizer
Based on CMU’s HUB-4 Sphinx-III System
- From 1996 CMU DARPA spoken language evaluation
- Acoustic Models were built from 50 hours of BN’96 and WSJ training data.
Specifications
- 30x RT Performance
~1000 hours CPU time for ~35 hours of speech
- Achieved 35.5% WER on TREC-6 Evaluation Data
- Smaller Beam, No Adaptation