Liquid AI Released LFM2-Audio-1.5B: An End-to-End Audio Foundation Model with Sub-100 ms Response Latency

Liquid AI has introduced LFM2-Audio-1.5B, a compact, end-to-end audiolanguage foundation model designed for real-time, low-latency speech and text understanding and generation on resource-constrained devices. The key innovation lies in its unified backbone that treats audio and text as first-class sequence tokens, utilizing a disentangled approach where raw waveform chunks are projected into continuous embeddings for input, and discrete audio codes are generated for output, thereby reducing artifacts and enabling autoregressive processing across modalities. This model extends the LFM2 family by incorporating a FastConformer-based audio encoder