Jepa Q2d2
a 1.6 kbps neural speech codec. a jepa (joint-embedding predictive architecture) encoder feeds q2d2, a geometry-aware 2d rhombic-lattice quantizer, and a hifi-gan-style decoder. the main codec is trained without any...
Model source
- Kind
- Other
- File size
- 578–588 MB
- Version
- vRevision ab41ea6
- Updated
- Oct 11, 2026
- Downloads
- 0
- Source
- Hugging Face
Source description
jepa-q2d2
a 1.6 kbps neural speech codec. a jepa (joint-embedding predictive architecture) encoder feeds q2d2, a geometry-aware 2d rhombic-lattice quantizer, and a hifi-gan-style decoder. the main codec is trained without any adversarial loss (a frozen wavlm perceptual loss is used while training the decoder).
- code: https://github.com/anant-004/jepa-q2d2
- paper: extended abstract in the github repo (
paper/sltc2026/main.pdf)
models
each folder has pytorch_model.pt, model.safetensors, and config.json.
| folder | operating point | quantizer | bitrate | quality |
|---|---|---|---|---|
jepa-q2d2-cd64-12.5hz | 12.5 hz, code dim 64 (main) | q2d2 | 1.6 kbps (100 tok/s) | pesq 2.53, estoi 0.80 |
jepa-q2d2-sigreg-cd32-25hz | 25 hz, code dim 32, sigreg | q2d2 | 1.6 kbps (100 tok/s) | pesq 2.33, estoi 0.76 |
teacher-cd128-fsq-12.5hz | 12.5 hz, code dim 128 | fsq | 4.8 kbps (237.5 tok/s) | pesq 3.14, estoi 0.87 |
quality is on a fixed 50-utterance librilight set. estoi is extended stoi, comparable only within this paper's pipeline. the main cd64 model is ahead of encodec at a comparable rate and ahead of mimi on pesq, behind mimi on estoi. the sigreg model is the co-design point: in our runs, the codec at the aggressive 25 hz / cd32 setting collapsed unless the encoder distribution was gaussianized with sigreg. unlike the main model, it was trained with a gan discriminator. the teacher is the higher-rate fsq codec (128 dims with 8 levels each, so 4.8 kbps). a separate cd64 run was distilled from it; the released cd64 model was trained without the distillation loss.
emergent properties
the codec is trained on english read speech only and never sees language or emotion labels. the encoder features still pick up structure it was not supervised on.
cross-lingual: a 5-nn probe on frozen pre-quantization features separates 6 fleurs languages at 0.85 (chance 0.167), ahead of mimi 0.63, encodec 0.61, dac 0.45 and the multilingual ssl models we tried (xls-r 0.75, mhubert-147 0.61). only uncompressed english wav2vec 2.0 base is higher (0.93). cluster nmi against language is higher too (0.26 vs encodec 0.10, dac 0.07). the utterance-disjoint number is the same (0.85).
cross-lingual separability
style and emotion: utterance embeddings separate by speaking style on a hindi 5-class set (angry, excited, neutral, sad, whisper). whisper splits off cleanly and the rest form distinct regions for the jepa encoders, while encodec and dac mix them.
emotion separation
usage
import torch, soundfile as sf
from koe.fast.hf_codec import load_codec_from_hf # from the github repo
model, info = load_codec_from_hf("main", device="cpu") # downloads weights from here
print(info)
wav, sr = sf.read("in.wav") # mono, resampled to 24 khz internally
x = torch.from_numpy(wav).float().view(1, 1, -1)
z_q = model.encode(x)[0] # discrete q2d2 tokens
recon = model.decode(z_q) # 24 khz waveform
model code lives in the github repo. the released checkpoints already include the fine-tuned encoder, so inference needs only these files plus the repo code. the files also hold the stage 1 ema target encoder and predictor (about 121m parameters), which inference does not use.
training data
librilight, english read speech, 24 khz. the cross-lingual structure above is evaluated zero-shot on fleurs.
citation
@misc{shukla2026jepaq2d2,
title = {JEPA-Q2D2: A Low-Bitrate Speech Codec with Emergent Cross-Lingual Structure},
author = {Shukla, Anant and Anand, Aman and Shakya, Suryansh and Bharti, Vatsal},
year = {2026},
howpublished = {\url{https://github.com/anant-004/jepa-q2d2}},
}
license
cc-by-4.0, released for research use.
Sources
1 source
Sources
1 sourceVerified Oct 11
Download files
Model artifacts
6 artifacts
Model artifacts
6 artifactsjepa-q2d2-cd64-12.5hz/model.safetensors
safetensors · 587 MB · SHA-256 2c8f74b5fc15…2df1 · Hugging Face
Download filejepa-q2d2-cd64-12.5hz/pytorch_model.pt
pt · 588 MB · SHA-256 e75f2841df33…9b4b · Hugging Face
Download filejepa-q2d2-sigreg-cd32-25hz/model.safetensors
safetensors · 578 MB · SHA-256 34d6b47f9d13…0c52 · Hugging Face
Download filejepa-q2d2-sigreg-cd32-25hz/pytorch_model.pt
pt · 579 MB · SHA-256 0a31341ac9c1…48eb · Hugging Face
Download fileteacher-cd128-fsq-12.5hz/model.safetensors
safetensors · 588 MB · SHA-256 aa513fc3af33…3231 · Hugging Face
Download fileteacher-cd128-fsq-12.5hz/pytorch_model.pt
pt · 588 MB · SHA-256 75b3dbd0c98c…72ec · Hugging Face
Download fileSource excerpts
3 excerpts
Source excerpts
3 excerpts--- license: cc-by-4.0 library_name: pytorch pipeline_tag: audio-to-audio tags: - speech-codec - neural-audio-codec - jepa - low-bitrate - speech language: - en --- # jepa-q2d2 a 1.6 kbps neural speech codec. a jepa (joint-embedding predictive architecture) encoder feeds q2d2, a geometry-aware 2d rhombic-lattice quantizer, and a hifi-gan-style decoder. the main codec is trained without any adversarial loss (a frozen wavlm perceptual loss is used while training the decoder). - code: https://github.com/anant-004/jepa-q2d2 - paper: extended abstract in the github repo (`paper/sltc2026/main.pdf`) ## models each folder has `pytorch_model.pt`, `model.safetensors`, and `config.json`. | folder | operating point | quantizer | bitrate | quality | |--------|-----------------|-----------|---------|---------| | `jepa-q2d2-cd64-12.5hz` | 12.5 hz, code dim 64 (main) | q2d2 | 1.6 kbps (100 tok/s) | pesq 2.53, estoi 0.80 | | `jepa-q2d2-sigreg-cd32-25hz` | 25 hz, code dim 32, sigreg | q2d2 | 1.6 kbps (100 tok/s) | pesq 2.33, estoi 0.76 | | `teacher-cd128-fsq-12.5hz` | 12.5 hz, code dim 128 | fsq | 4.8 kbps (237.5 tok/s) | pesq 3.14, estoi 0.87 | quality is on a fixed 50-utterance librilight set. estoi is extended stoi, comparable only within this paper's pipeline. the main cd64 model is ahead of encodec at a comparable rate and ahead of mimi on pesq, behind mimi on estoi. the sigreg model is the co-design point: in our runs, the codec at the aggressive 25 hz / cd32 setting collapsed unless the encoder distribution was gaussianized with sigreg. unlike the main model, it was trained with a gan discriminator. the teacher is the higher-rate fsq codec (128 dims with 8 levels each, so 4.8 kbps). a separate cd64 run was distilled from it; the released cd64 model was trained without the distillation loss. ## emergent properties the codec is trained on english read speech only and never sees language or emotion labels. the encoder features still pick up structure it was not supervised on. cross-lingual: a 5-nn probe on frozen pre-quantization features separates 6 fleurs languages at 0.85 (chance 0.167), ahead of mimi 0.63, encodec 0.61, dac 0.45 and the multilingual ssl models we tried (xls-r 0.75, mhubert-147 0.61). only uncompressed english wav2vec 2.0 base is higher (0.93). cluster nmi against language is higher too (0.26 vs encodec 0.10, dac 0.07). the utterance...
--- license: cc-by-4.0 library_name: pytorch pipeline_tag: audio-to-audio tags: - speech-codec - neural-audio-codec - jepa - low-bitrate - speech language: - en --- # jepa-q2d2 a 1.6 kbps neural speech codec. a jepa (joint-embedding predictive architecture) encoder feeds q2d2, a geometry-aware 2d rhombic-lattice quantizer, and a hifi-gan decoder. the codec objective is trained without any adversarial loss (a frozen wavlm perceptual loss is used while training the decoder). - code: https://github.com/anant-004/jepa-q2d2 - paper: apsipa asc 2026 ## models each folder has `pytorch_model.pt`, `model.safetensors`, and `config.json`. | folder | operating point | quantizer | bitrate | quality | |--------|-----------------|-----------|---------|---------| | `jepa-q2d2-cd64-12.5hz` | 12.5 hz, code dim 64 (main) | q2d2 | 1.6 kbps (100 tok/s) | pesq 2.53, estoi 0.80 | | `jepa-q2d2-sigreg-cd32-25hz` | 25 hz, code dim 32, sigreg | q2d2 | 1.6 kbps (100 tok/s) | estoi 0.79 | | `teacher-cd128-fsq-12.5hz` | 12.5 hz, code dim 128 | fsq | ~2.85 kbps (237.5 tok/s) | pesq ~2.91 | quality is on a fixed 50-utterance librilight set. estoi is extended stoi, comparable only within this paper's pipeline. the main cd64 model is ahead of encodec at a comparable rate and ahead of mimi on pesq. the sigreg model is the co-design point: at the aggressive 25 hz / cd32 setting the codec collapses unless the encoder distribution is gaussianized with sigreg. the teacher is the higher-rate fsq codec used to distill the cd64 student. ## emergent properties the codec is trained on english read speech only and never sees language or emotion labels. the encoder features still pick up structure it was not supervised on. cross-lingual: a 5-nn probe on frozen pre-quantization features separates 6 fleurs languages at 0.85 (chance 0.167), ahead of mimi 0.63, encodec 0.61, dac 0.45, and close to multilingual ssl models. cluster nmi against language is higher too (0.26 vs encodec 0.10, dac 0.07). the utterance-disjoint number is the same (0.85).  style and emotion: utterance embeddings separate by speaking style on a hindi 5-class set (angry, excited, neutral, sad, whisper). whisper splits off cleanly and the rest form distinct regions for the jepa encoders, while encodec and dac mix them. 