Over the past two months, we’ve open-sourced multiple generations of on-device Audio models to the community through Audio8. The current portfolio includes: • ASR: 0.1B, 0.3B, 0.6B, and 3B • TTS: 0.1B, 0.3B, and 0.6B These releases are designed for practical local inference, with deployment targets spanning phones, PCs,... and other resource-constrained devices.show more

These releases are more than a collection of model checkpoints. Audio8 is an end-to-end approach to on-device audio intelligence, covering model design and technical report, deployment-aware export, optimized runtimes, and hardware-specific execution paths. ONNX, INT8/INT4, and iOS ANE variants allow the models to adapt to different memory, compute, and power budgets while remaining practical for local deployment.

Our ASR releases, sorted by model scale, with deployment variants grouped together: - 0.1B / 0.3B line Audio8-ASR-0.1B iOS ANE ONNX Runtime - 0.6B line ARK-ASR-0.6B INT8 ONNX 3B line ARK-ASR-3B Our TTS releases, sorted by model scale: - 0.1B line ONNX INT8 Preview 0.1B 0.3B line GPA 0.6B line Preview 0.6B ONNX INT4 The Hugging Face page also includes GPA-v1.5, an additional 1B-scale TTS release:

We’ll continue open-sourcing more Audio8 versions, model sizes, deployment formats, and capabilities. Thank you to everyone who has downloaded the models, tested them locally, and shared feedback with us. On September 10, we’ll make a major open-source release: Edge0 — a breakthrough in on-device AI.

no hindi? 👀

Local inference sounds safer than cloud until your own box gets popped. Mine did on January 28th, an attacker spun up a fake admin role, botched demoting the real one, wiped the database, then dropped in Kinsing anyway. Running it yourself just moves where the risk lives.

Very useful!

on device audio models that actually run on phones love this

Big sound tiny footprint AI voice that runs right on your device

On device audio from 0 1B to 3B is huge local inference finally gets real options for every device size

Impressive progress—bringing capable audio models to local devices is a huge win for accessible AI.

Powerful audio running right where you are on device

on device audio is the real unlock here

Open source Audio8 bringing real local voice AI everywhere

Clear release. On device audio models help when data should stay local.

I do love it. Open 0.1B audio models are the practical edge path. Phones and offline assistants need local inference, not giant server clusters.

Strong lineup for bringing capable audio AI directly onto devices

Huge contribution to the open-source community a full range of on-device ASR and TTS models optimized for phones and PCs is exactly what local AI needed!

Impressive progress bringing capable audio AI directly to everyday devices

Incredible contribution open source audio models empowering local AI inference for everyone.🎙️🔥✨

Shipping real on device audio models at this scale is huge

Samuel Zeng: Audio8 open-sourced on-device ASR and TTS models from 0.1B to 3B for practical local inference

curious what accuracy looks like at 0.1B on a domain-specific task vs the 3B

Buena pregunta. He probado herramientas similares y las locales me han funcionado mejor que las APIs caras. ¿Cuál estás evaluando tú ahora mismo?