LFM3-Audio-2B
LFM3-Audio-2B accepts text or audio and generates text, speech or both.
What the release provides
The checkpoint has a 32K shared window, 16 kHz audio input and 24 kHz speech output. Voice chat currently supports English; transcription, synthesis and translation cover 18 languages. It uses LiquidCodec and an in-house audio encoder.LFM3-Audio-2B official model card
How to use it well
Practical guidance. Evaluate the chosen audio task directly. Multilingual transcription support does not establish equivalent conversational support in every language. Include interruptions, background sound and tool delays in a voice workflow review.
Hosted API contract
The API catalog lists lfm3-audio-2b with a 128,000-token window. Its supported inputs and limits can differ from the downloadable checkpoint.Liquid API model catalog
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API, Downloadable |
| Model identifiers | LiquidAI/LFM3-Audio-2Blfm3-audio-2b |
| Context window | 32,768 tokens Downloadable checkpoint limit; the hosted API has a separate service contract. LFM3-Audio-2B official model card |