TOOLORA / Local reading edition
Dia
Build English two-speaker dialogue with speaker tags and audio prefixes for voice conditioning; current Dia2 provides checkpoints, scripts and WAV output.
Overview
Dia is Nari Labs’ dialogue-speech family; this entry opens Dia2. Scripts distinguish speakers with [S1]/[S2] tags and can use per-speaker audio prefixes to condition English dialogue saved as WAV. Current documentation limits generation to two minutes. The model has a streaming design, but the separate TTS Server with real streaming remains planned rather than an available hosted API. Main code uses Apache 2.0; Mimi and other components retain their licenses.
- Company country / region
- United States · Company headquarters
Official sources and references
- Official websitehttps://github.com/nari-labs/dia2
- Documentationhttps://huggingface.co/nari-labs/Dia2-2B
- Documentationhttps://github.com/nari-labs/dia
Descriptions summarize public information; consult official websites for features, pricing and regional availability.