CPCConformerTransfomerSpeakerDiarizationModel

This repository provides the PyTorch implementation of a streaming-capable speaker diarization model based on BW-EDA-EEND.
The model is trained for two-speaker English audio, using a Conformer encoder and CPC (Contrastive Predictive Coding) feature extractor.

👉 Pretrained models and usage examples are available on Hugging Face:
mocomoco-inc/SpeakerDiarizationModel-en-2spk

---

Installation

Clone this repository and install dependencies:

git clone https://github.com/mocomoco-inc/CPCConformerTransfomerSpeakerDiarizationModel.git
cd CPCConformerTransfomerSpeakerDiarizationModel
pip install -e .

---

Repository Structure

.
├── cpc_streaming_diarization
│   ├── config.py   # Model configuration classes and default parameters
│   ├── model.py    # Main diarization model (CPC + Conformer + Transformer)
│   ├── modules     # Submodules used inside the model
│   │   └── ...
│   ├── utils.py    # Helper functions (e.g., device setup, postprocessing)
│   └── ...
├─ examples
│   └── diarize.py  # Example script for running inference
└── ...

---

Pretrained Models

• Hugging Face Hub: mocomoco-inc/SpeakerDiarizationModel-en-2spk

---

License

This project is licensed under the Apache-2.0 License.

---

Contact

For any inquiries, please contact us at:
mocomoco inc. Inada Bldg. 302, 7-20-19 Roppongi,
Minato-ku, Tokyo 106-0032, Japan
contact@mocomoco.ai