Liquid AI Releases LFM3-Audio-2B Compact Local Audio Model
Liquid AI has released LFM3-Audio-2B, a compact 2-billion parameter audio model designed to run locally on laptops and edge devices. It supports multimodal input and output, streaming voice chat, reasoning, multilingual translation, and tool calling. This model brings advanced multimodal audio capabilities directly to consumer hardware without requiring a cloud connection. It empowers developers and users to build private, low-latency voice assistants and edge applications. LFM3-Audio-2B features an in-house audio encoder and LiquidCodec audio tokenizer, improving speech quality over previous versions across 18 languages. It can accept text or audio inputs and respond with text, speech, or both.
## BACKGROUND
Liquid AI is an MIT spin-off that develops liquid foundation models capable of running directly on devices with smaller memory and computing footprints. Tool calling, or function calling, is a capability that allows generative AI systems to interact with external tools or APIs by generating structured calls.