Mamba-based Segmentation Model for Speaker Diarization

Alexis Plaquet, Naohiro Tawara, Marc Delcroix, Shota Horiguchi, Atsushi Ando, and Shoko Araki

Mamba is a newly proposed architecture which behaves like a recurrent neural network (RNN) with attention-like capabilities. These properties are promising for speaker diarization, as attention-based models have unsuitable memory requirements for long-form audio, and traditional RNN capabilities are too limited. In this paper, we propose to assess the potential of Mamba for diarization by comparing the state-of-the-art neural segmentation of the pyannote.audio pipeline with our proposed Mamba-based variant. Mamba's stronger processing capabilities allow usage of longer local windows, which significantly improve diarization quality by making the speaker embedding extraction more reliable. We find Mamba to be a superior alternative to both traditional RNN and the tested attention-based model. Our proposed Mamba-based system achieves state-of-the-art performance on three widely used diarization datasets.

• 🌐 Read the paper on IEEE Xplore
• 📄 Read the preprint on arXiv

Citations

@INPROCEEDINGS{plaquet2025mambabasedsegmentationmodel,
  author={Plaquet, Alexis and Tawara, Naohiro and Delcroix, Marc and Horiguchi, Shota and Ando, Atsushi and Araki, Shoko},
  booktitle={ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, 
  title={Mamba-based Segmentation Model for Speaker Diarization}, 
  year={2025},
  pages={1-5},
  keywords={Pipelines;Memory management;Bidirectional long short term memory;Signal processing;Acoustics;Reliability;Speech processing;Speaker diarization;end-to-end neural diarization;Mamba;state-space model},
  doi={10.1109/ICASSP49660.2025.10889446}}

Repository contents

• The source code relevant for running and evaluating Mamba-based models in src/
• Tutorial notebooks in tutorials/:
  1. Setting up the environment
  2. Training a Mamba-based segmentation model from scratch
• The subset splits used for each dataset in databases/ (in pyannote.audio format)
• The predictions outputted by the model (in eval_rttms.zip files) and the detail of all computed metrics in .csv files, all contained in results/

Installation

You can install the plaqntt package using pip.

1. Clone this repository and open a terminal in the same folder as this file.
2. Run pip install -e .

License

Please refer to the LICENSE file for details.