LiquidCodec: High-Fidelity Audio Tokenization Optimized for Speed | Blog | Liquid AI

Liquid AI has released LiquidCodec, an in-house streaming audio codec optimized for speed, high-fidelity reconstruction, and on-device deployment. Built on the LFM3 architecture, it combines traditional spectral processing with Grouped Finite Scalar Quantization (GFSQ) to enable parallel token prediction. According to the company, the codec outperforms comparable models in generation speed, naturalness, and token-to-waveform decoding efficiency while maintaining strong multilingual reconstruction performance. Open article →