Liquid AI has released LiquidCodec, the audio codec behind its LFM3-Audio-2B and LFM3-Omni-2B models. The LiquidCodec audio codec is a ~74M-parameter streaming speech tokenizer. It turns speech into tokens a language model can predict, then converts those tokens back into audio. Liquid AI had 2 goals: easy-to-predict tokens and fast conversion back to speech. The weights are on Hugging Face.
TL;DR
- Size: ~74M parameters (73.6M on the model card); 24 kHz audio, 12.5 Hz frames, 12 codebooks.
- Runs on: a single CPU thread (tested on an Apple M4 Max laptop, FP32); CUDA GPUs via Transformers; Q4_0, Q4_K_M and Q8_0 GGUF builds for llama.cpp.
- Performance: best ViSQOL and lowest mel distance on all 8 test sets versus X-Codec 2, NanoCodec and Mimi.
- Best: ~361x real-time encoding on 1 CPU thread (encode RTF 0.0028 vs Mimi’s 0.0361).
What is LiquidCodec?
LiquidCodec is a Liquid AI speech codec that compresses 24 kHz audio into 12 token streams. Each 80 ms frame becomes 12 discrete tokens, per the LFM3-Audio-2B model card. It replaces Kyutai’s Mimi, which Liquid AI used in LFM2-Audio-1.5B. Mimi has 96.2M parameters.
How does LiquidCodec work?
The pipeline is encode, quantize, decode. Liquid AI uses blocks from its LFM3 architecture in both the encoder and decoder. The 2 halves are symmetric, and the encoder is fully causal, which suits streaming.
Quantization uses grouped finite scalar quantization (GFSQ) instead of residual vector quantization (RVQ). In RVQ, each codebook encodes what the previous one missed. GFSQ groups are independent, so all 12 codebooks can be predicted in parallel by MaskGIT-style transformers.
The decoder predicts spectral coefficients directly. This avoids learned transposed-convolution upsampling, a common cost in waveform decoders.
How was it trained?
Training ran in 2 stages. Stage 1 used 264,000 hours of speech across 107 languages. It applied waveform and spectral losses, without adversarial losses. Stage 2 fine-tuned on 55,000 hours of full-band speech (24 kHz or higher) with adversarial critics. Liquid AI says this removed metallic artifacts left after stage 1. A final decoder-only quantization-aware training (QAT) step targets edge inference.
How fast is LiquidCodec?
Liquid AI measured speed in FP32 on 1 CPU thread of an Apple M4 Max MacBook Pro. LiquidCodec ran at ~361x real time for encoding and ~87x for decoding. These are self-reported figures.
Real-time factor (RTF) is processing time divided by audio length, so lower is better. LiquidCodec’s encode RTF is 0.0028, against 0.0361 for Mimi, 0.2145 for NanoCodec and 0.7359 for X-Codec 2. Decode RTF is 0.0115, against 0.0329, 0.2543 and 0.0379. That is about 13x lower encode RTF and 2.9x lower decode RTF than Mimi.
How does LiquidCodec compare on quality benchmarks?
Liquid AI used 8 test sets. These were LibriSpeech test-clean, Seed-TTS English and Mandarin, and 5 FLEURS languages (German, French, Mandarin, Japanese, Korean). Baselines were X-Codec 2, NanoCodec and Mimi at 12 codebooks.
- Reconstruction: LiquidCodec had the best ViSQOL and lowest mel distance on all 8 sets. On LibriSpeech, ViSQOL was 4.319 versus 4.219 (NanoCodec), 4.119 (X-Codec 2) and 4.031 (Mimi).
- Text-to-speech: Liquid AI trained identical TTS models on each codec’s tokens. Data (101,000 hours, 6 languages), architecture, compute and batch order were fixed. LiquidCodec scored the highest UTMOSv2 on all 8 sets and the best WER/CER on 4 of 8. LibriSpeech WER was 3.15%, against 6.27% for Mimi. Results held with a Qwen3-1.7B backbone as well as LFM3-1.7B.
- Where it trails: NanoCodec scored higher PESQ (WB) on 7 of 8 sets. It also led on LibriSpeech speaker similarity (SECS 0.836 vs 0.785).
‘+it.v+’
