MultiTalkFT Fine tuning corpus for full duplex multi speaker dialogue. Schemas data {zh,en}{, multichannel}.jsonl (one record per line): field type description path string relative path to the audio file voice string relative path to speaker prompt duration float clip duration in seconds system string persona / system prompt transcripts/ .parquet : column type description audio path string matches data .jsonl path id string duration float num channels int32 original conversation speaker count speaker to channel string JSON encoded {speaker: channel index} voice string JSON encoded {speaker: relative voice path} alignments string JSON encoded flat list [[word, [start, end], speaker label], …] training string JSON encoded {system prompt, voice prompt (relative), …} Quick load
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy