LiquidAI releases browser demo for LFM3-Omni voice AI
LiquidAI has made a public demo available on Hugging Face that runs the LFM3-Omni-2B model directly in web browsers using WebGPU. This approach allows users to experience local voice AI without installing specialized server software, provided their device has a compatible GPU and browser support. The demo files were added on October 30, 2026, and a developer shared details on Reddit the…
Key points
- LiquidAI released a WebGPU browser demo for LFM3-Omni-2B on Hugging Face on October 30, 2026.
- The model handles text and audio in a single architecture, avoiding separate transcription and speech synthesis steps.
- The demo requires a compatible GPU and browser, with an initial data download and a 30-second voice input limit.
The model is multimodal, handling text, images, and audio in a single architecture rather than using separate components for transcription and speech synthesis. The interface includes text and voice input options, with a 30-second limit for single voice recordings. While the model card lists Japanese as a supported language, the author has not verified the accuracy or speed of Japanese speech recognition in this specific browser environment.
Users should note that "local execution" still requires an initial data download to load the model, which consumes bandwidth and device memory. Performance varies by hardware; speed figures cited in community posts are based on an M4 Max chip and may not apply to other devices. The author confirmed the code and page existence but did not conduct personal device testing for voice conversation quality.
The story so far
2 episodes →- LiquidAI releases browser demo for LFM3-Omni voice AIthis story
Running Voice AI in the Browser: Public Demo of LFM3-Omni
note.com · 30 October 2026
Loading the full article…
This text was published by note.com and written by r3ach｜AIニュースの背景とこれから. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Google announces Gemini 4 Carbon and Genie 4 models with staged rollout for cyber defenders · 7 src
- Anthropic doubles Opus 5.5 fast mode pricing to $8/$40 per million tokens · 1 src
- Datalab releases Surya 2.5 OCR model with 85.9% accuracy on OmniParseBench · 1 src
- Sensenova/SenseNova-Loop-9B-Preview · Hugging Face · 1 src
- TBC launches Mischief video model with neuron-inspired speed boost · 1 src
Comments
via GitHub Discussions