A TTS-STT flywheel generates 22k synthetic entity-dense utterances to LoRA-fine-tune Whisper, lifting Telugu EHR from 0.027 to 0.473 and similar gains in Hindi/Tamil while releasing all data and code.
LASE: Language-adversarial speaker encoding for indic cross-script identity preservation
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
The TTS-STT Flywheel: Synthetic Entity-Dense Audio Closes the Indic ASR Gap Where Commercial and Open-Source Systems Fail
A TTS-STT flywheel generates 22k synthetic entity-dense utterances to LoRA-fine-tune Whisper, lifting Telugu EHR from 0.027 to 0.473 and similar gains in Hindi/Tamil while releasing all data and code.