What it is
Fork of Bark that adds batched, streaming long-form text-to-audio generation with interactive playback and multi-GPU inference support. Generate long-form speech and other audio from text with automatic chunking, batched decoding, streaming segments, live playback, and optional multi-GPU execution.
Key results
- 3x 4090: long jobs drop from 2-5 min to 30-60 s
- Streams decoded audio in ~3 s segments via generate_audio_long(stream=True)
- Batches all 4 stages natively instead of sequential per prompt