Breeze ASR Quick Start (Windows)

• Create venv: Set up an isolated Python environment.
• Install requirements: Install dependencies from requirements.txt.
• Run script: Execute breeze_acr.py to transcribe an audio file.

Prerequisites

• Python: 3.9–3.13 (workspace uses 3.13)
• Git/Shell: Windows PowerShell
• Audio file: Provide a local .mp3 or .wav path
• Optional: NVIDIA GPU with CUDA for faster inference

1) Create and activate virtual environment

python -m venv .venv

.\.venv\Scripts\activate

2) Install dependencies

pip install torch==2.9.1+cu126 torchvision==0.24.1+cu126 torchaudio==2.9.1 --index-url https://download.pytorch.org/whl/cu126

pip install -r requirements.txt

3) Prepare an audio file

• Place or reference an audio file path you want to transcribe.
• Edit breeze_acr.py to set audio_path to your file, e.g. C:\path\to\audio.mp3.

4) Run the transcription script

python breeze_acr.py

Notes

• The script auto-selects CPU/GPU and resamples to 16 kHz if needed.
• Stereo inputs are converted to mono automatically.
• Model used: MediaTek-Research/Breeze-ASR-25 via Hugging Face Transformers.

Model and License

This project uses the Breeze-ASR-25 model from MediaTek Research: https://huggingface.co/MediaTek-Research/Breeze-ASR-25

Model License: Apache-2.0