Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
2
0
3
Current TTS evaluation metrics can be misleading for low-resource languages, as systems might "succeed" by speaking a neighboring language or producing unnatural audio that still yields a low WER.
Multilingual ASR models can completely fail on low-resource languages like Pashto, not just in accuracy but by hallucinating the wrong script.