VoxTranscript

A Windows desktop application for offline audio transcription with speaker diarization, built on top of WhisperX, faster-whisper, and pyannote.audio.

Installation

Download the latest installer from the Releases page and run VoxTranscript_Setup.exe. The installer requires no administrator privileges and installs to %LOCALAPPDATA%\VoxTranscript\.

The installer is split into multiple .bin slices to fit GitHub's per-file size limit. Download all files (VoxTranscript_Setup.exe and every VoxTranscript_Setup-*.bin) into the same folder before running the .exe.

Development setup

Requirements: Python 3.10, Inno Setup 7 (for building the installer).

py -3.10-m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r requirements.txt
python app.py

Models

The bundled installer ships with the models below pre-installed. For a development checkout, download each into its corresponding subfolder under models/:

• faster-whisper → models/faster_whisper/ — download from Systran on HuggingFace
• pyannote segmentation 3.0 → models/pyannote/segmentation-3.0/ — download from HuggingFace (gated, requires accepting terms)
• pyannote wespeaker-voxceleb-resnet34-LM → models/pyannote/wespeaker-voxceleb-resnet34-LM/ — download from HuggingFace (gated, requires accepting terms)
• wav2vec2 → models/wav2vec2_base/ — download from download.pytorch.org/models/wav2vec2_fairseq_base_ls960_asr_ls960.pth
• SpeechBrain ECAPA-TDNN → models/speechbrain_models/ — downloaded automatically on first run

Building the installer

pyinstaller app.spec --clean
&"C:\Program Files\Inno Setup 7\ISCC.exe" VoxTranscriptSetup.iss

Output lands in Output\ as a VoxTranscript_Setup.exe plus one or more .bin slice files.

Acknowledgements

VoxTranscript bundles and uses the following third-party models and software. Their LICENSE and README files are preserved inside the installed models/ directory.

Component	Authors	License	Source
WhisperX (transcription pipeline)	Max Bain et al.	BSD-4-Clause	m-bain/whisperX
faster-whisper (CTranslate2 Whisper inference)	Systran	MIT	SYSTRAN/faster-whisper
OpenAI Whisper (underlying speech recognition model)	OpenAI	MIT	openai/whisper
pyannote.audio (speaker diarization toolkit)	Hervé Bredin / CNRS	MIT	pyannote/pyannote-audio
pyannote/segmentation-3.0 (speaker segmentation model)	Alexis Plaquet, Hervé Bredin	MIT	HF: pyannote/segmentation-3.0
pyannote/wespeaker-voxceleb-resnet34-LM (speaker embedding model)	Wenet/WeSpeaker team; pyannote.audio wrapper by H. Bredin	CC-BY 4.0 (inherited from VoxCeleb dataset)	HF: pyannote/wespeaker-voxceleb-resnet34-LM
wav2vec2 (forced-alignment model)	Meta AI / Facebook AI Research	BSD-style (via torchaudio)	pytorch/audio
SpeechBrain spkrec-ecapa-voxceleb (speaker embedding model)	SpeechBrain team	Apache 2.0	HF: speechbrain/spkrec-ecapa-voxceleb
FFmpeg (audio decoding)	FFmpeg contributors	LGPL/GPL	ffmpeg.org

Citations

If you use VoxTranscript in academic work, please also cite the underlying research:

@inproceedings{bain2023whisperx,
title={WhisperX: Time-Accurate Speech Transcription of Long-Form Audio},
author={Bain, Max and Huh, Jaesung and Han, Tengda and Zisserman, Andrew},
booktitle={INTERSPEECH 2023},
year={2023}
}
@inproceedings{Plaquet23,
title={Powerset multi-class cross entropy loss for neural speaker diarization},
author={Plaquet, Alexis and Bredin, Hervé},
booktitle={Proc. INTERSPEECH 2023},
year={2023}
}
@inproceedings{Bredin23,
title={{pyannote.audio 2.1 speaker diarization pipeline: principle, benchmark, and recipe}},
author={Bredin, Hervé},
booktitle={Proc. INTERSPEECH 2023},
year={2023}
}
@inproceedings{Wang2023,
title={Wespeaker: A research and production oriented speaker embedding learning toolkit},
author={Wang, Hongji and Liang, Chengdong and Wang, Shuai and Chen, Zhengyang and Zhang, Binbin and Xiang, Xu and Deng, Yanlei and Qian, Yanmin},
booktitle={ICASSP 2023},
year={2023}
}

Each bundled model also retains its own LICENSE and README inside the install's models/ subdirectory.