Resegmentation of SWITCHBOARD.

摘要: The SWITCHBOARD (SWB) corpus is one of the most important benchmarks for recognition tasks involving large vocabulary conversational speech (LVCSR). high error rates on SWB are largely attributable to an acoustic model mismatch, frequency poorly articulated monosyllabic words, and variations in pronunciations. It imperative improve quality segmentations transcriptions training data achieve better modeling. By adapting existing models only a small subset such improved transcriptions, we have achieved 2% absolute improvement performance.