Liquid AI unveils 73.6 million parameter audio codec

Liquid AI released LiquidCodec, a streaming speech codec with 73.6 million parameters, on November 12. The codec serves as the audio tokenizer for the company's LFM3-Audio-2B and LFM3-Omni-2B models, converting 24kHz speech into discrete tokens and reconstructing audio from them via separate encoder and decoder components. Liquid AI's technical blog rounds the model to roughly 74 million parameters, while the Hugging Face model card lists 73.6 million under the lfm1.0 license, whose terms must be checked separately for commercial use. The release matters because efficient tokenization underpins real-time speech understanding and generation in Liquid AI's small-model lineup. What to watch next is uptake of LiquidCodec by developers building streaming voice applications and clarification of commercial licensing terms.