Synthetic conversational data boosts ASR training

Arxiv CS.SD (Sound/Audio ML)

Proposes an augmentation pipeline generating scenario-level dialogues for low-resource conversational ASR. Addresses scarcity of multi-speaker training data in niche domains. Method maps speaker metadata to synthetic conversations.

speech-to-textdata-augmentationspeech-recognition

Arxiv CS.SD (Sound/Audio ML)·M\'at\'e GedeonP\'eter Mihajlik

8 days ago