🔀 DyCAST

A variable-frame-rate 16 kHz speech codec based on FocalCodec.

• 📜 Papers:

  • Beyond Fixed Frames: Dynamic Character-Aligned Speech Tokenization

  • FocalCodec-Stream: Streaming Low-Bitrate Speech Coding via Causal Distillation

  • FocalCodec: Low-Bitrate Speech Coding via Focal Modulation Networks

• 🔊 Downstream Tasks: https://github.com/lucadellalib/audiocodecs

---

📌 Available Checkpoints

Checkpoint	Sample Rate (kHz)	Frame Rate (Hz)	Codebooks	Bitrate (kbps)	Dataset
lucadellalib/dycast	16	6.2 - 17.5	1 x (4^32)	0.40 - 1.12	LibriTTS-960

---

🛠️️ Installation

First of all, install Python 3.8 or later. Then, open a terminal and run:

pip install faiss-cpu huggingface-hub numpy safetensors soundfile torch torchaudio torchcodec transformers

---

▶️ Quickstart

NOTE: the audios directory contains audio samples that you can download and use to test the codec.

You can easily load the model using torch.hub without cloning the repository:

import torch
import torchaudio

# Load DyCAST model
codec = torch.hub.load(
    repo_or_dir="lucadellalib/dycast",
    model="dycast",
    config="lucadellalib/dycast",
    force_reload=True,  # Fetch the latest version from Torch Hub
)
codec.eval().requires_grad_(False)

# Load and preprocess the input audio
audio_file = "audios/librispeech-dev-clean/251-118436-0003.wav"
sig, sample_rate = torchaudio.load(audio_file)
sig = torchaudio.functional.resample(sig, sample_rate, codec.sample_rate_input)

# Forward
toks, pcodes, rec_sig = codec(sig)
print("Tokens:")
print(toks.shape)
print(toks)

print("Pooled codes:")
print(pcodes.shape)
print(pcodes)

# Save the reconstructed audio
rec_sig = torchaudio.functional.resample(rec_sig, codec.sample_rate_output, sample_rate)
torchaudio.save("reconstruction.wav", rec_sig, sample_rate)

Alternatively, you can install DyCAST as a standard Python package using pip:

pip install dycast@git+https://github.com/lucadellalib/dycast.git@main#egg=dycast

Once installed, you can import it in your scripts:

import dycast

config = "lucadellalib/dycast"
codec = dycast.DyCAST.from_pretrained(config)

Check the code documentation for more details on model usage and available configurations.

---

🎤 Running the Demo

Clone or download and extract the repository, navigate to <path-to-repository>, open a terminal and run:

python dycast/codec.py audios/librispeech-dev-clean/251-118436-0003.wav

Reconstructed audio samples using different inference modes can be found in the reconstructions directory.

---

@ Citing

@article{dellalibera2026dycast,
    title   = {Beyond Fixed Frames: Dynamic Character-Aligned Speech Tokenization},
    author  = {Luca {Della Libera} and Cem Subakan and Mirco Ravanelli},
    journal = {arXiv preprint arXiv:2601.23174},
    year    = {2026},
}

@article{dellalibera2025focalcodecstream,
    title   = {{FocalCodec-Stream}: Streaming Low-Bitrate Speech Coding via Causal Distillation},
    author  = {Luca {Della Libera} and Cem Subakan and Mirco Ravanelli},
    journal = {arXiv preprint arXiv:2509.16195},
    year    = {2025},
}

@inproceedings{dellalibera2025focalcodec,
    title     = {{FocalCodec}: Low-Bitrate Speech Coding via Focal Modulation Networks},
    author    = {Luca {Della Libera} and Francesco Paissan and Cem Subakan and Mirco Ravanelli},
    booktitle = {Advances in Neural Information Processing Systems},
    year      = {2025},
}

---

📧 Contact

luca.dellalib@gmail.com

---