Audio8
The publisher of audio8-tts, a compact multilingual text-to-speech model with zero-shot voice-cloning. Discovered while ingesting audio8-tts.
Little is public about the organization itself. It has no company site this wiki could find — the
GitHub org and the Hugging Face org
are the only presences, with Audio8_TTS (895 stars, Python, Apache-2.0) created 2026-07-29 and
the first checkpoint published 2026-08-19. Country, funding and team are unstated in every
artifact read. The model card is written in English with Chinese in the code examples, and the
licence names an authoritative English text with a Chinese translation for reference — noted as an
observation, not an inference about where the team sits.
Two things it has declared, both unusual enough to be the reason this node exists:
- A revenue-capped licence. The Audio8 Community License v1.0 makes commercial use free below US$2M in annual revenue and requires a written licence above it — a third shape beside the permissive and research-only poles open-weight-tts was built around.
- Behavioural use terms on a cloning model. Consent before impersonation, disclosure of synthetic audio. free-voice-clone-list's 53-model census found no such language anywhere; the one exception it did contain was on a model that cannot clone.
It ships two checkpoint sizes (0.1B and 0.6B, the latter only referenced in evaluation tables so far) and packages the small one for CPU under ONNX Runtime.
Connections
- publisher of audio8-tts
- credits fish-audio-s2-pro's DualAR architecture as the design it works from
Related
audio8-tts · text-to-speech · open-weight-tts · voice-cloning · fish-audio-s2-pro