Developers who want a fast, free open speech model that runs on-device; the hallucination claims are Hume's own tests.
Kenning's take
What it does
TADA (Text-Acoustic Dual Alignment) is an open-source speech generation model from Hume AI. It pairs each piece of text with exactly one chunk of audio, which Hume says makes it more than five times faster than similar LLM-based speech models and stops it from skipping or adding words. The code and 1B and 3B models are free to download, and the model is small enough to run on a device.
- One-to-one alignment of text and audio
- Real-time factor of 0.09, per Hume
- Zero hallucinations on 1,000+ LibriTTSR test samples, per Hume
- 1B English and 3B multilingual models
- MLX versions for Apple Silicon
- Light enough for on-device use
Who it suits. Developers and researchers building voice apps who want a fast, reliable open speech model.
Key facts
- Category
- AI models
- Made by
- Hume AI
- Based in
- United States
- Released
- 10 March 2026
- How to get it
- free model weights on Hugging Face and code on GitHub
- Works on
- Hugging Face (open weights), GitHub, On-device (MLX)
- In the UK
- Yes
- Free plan
- Free open weights (tada-1b, tada-3b-ml) under the Llama 3.2 licence
FAQ
When was TADA released?
Hume AI open-sourced TADA on 10 March 2026.
Is TADA free?
Yes, the code and models are free to download. The models on Hugging Face use the Llama 3.2 licence.
What languages does TADA support?
The 3B multilingual model lists English, Japanese, German, French, Spanish, Chinese, Arabic, Italian, Polish and Portuguese on Hugging Face.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools
ACE-Step 1.5
Open-source music generation model that writes full songs with lyrics and runs on ordinary consumer GPUs.
AlphaGenome Atlas
Google DeepMind database of AI-predicted effects for every possible single-letter change in human DNA.
Amazon Nova 2
Amazon's second family of Nova AI models, covering reasoning, speech conversation and combined text-and-image generation.
Cisco Antares
Cisco's small open-weight models that find where known security vulnerabilities sit in a codebase, running locally.