Whisper – Batch transcription (CPU & GPU)

Minimal project to transcribe audio files from the audio_ref/ folder to text using Whisper (OpenAI open source), with two scripts:

• run_whisper.py (CPU)
• run_whisper_gpu.py (GPU CUDA)

Transcriptions and status are printed to the terminal and logged to files (whisper.log, whisper_gpu.log). Models are downloaded/stored under the project model/ folder.

Useful links

• Whisper (OpenAI, open source): https://github.com/openai/whisper
• PyTorch CUDA 12.8 (cu128) index: https://download.pytorch.org/whl/cu128
• FFmpeg for Windows (Gyan builds): https://www.gyan.dev/ffmpeg/builds/

Prerequisites

• Windows 10/11 (PowerShell) with Python 3.10+ installed and on PATH
• FFmpeg available on PATH (audio decoding)
• Optional (GPU): NVIDIA GPU (e.g., RTX 5080) + up-to-date NVIDIA drivers + CUDA runtime via PyTorch cu128

Quick install (CPU)

cd D:\TTS\Whisper
python -m venv .venv
Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass -Force
.\.venv\Scripts\Activate.ps1
python -m pip install -U pip setuptools wheel
python -m pip install -U openai-whisper
winget install -e --id Gyan.FFmpeg --accept-package-agreements --accept-source-agreements

Installation (GPU – CUDA)

Install PyTorch with CUDA 12.8 (cu128), then Whisper:

cd D:\TTS\Whisper
Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass -Force
.\.venv\Scripts\Activate.ps1
python -m pip uninstall -y torch torchvision torchaudio
python -m pip install --index-url https://download.pytorch.org/whl/cu128 torch torchvision torchaudio -U
python -m pip install -U openai-whisper
python - << "PY"
import torch
print('cuda available:', torch.cuda.is_available())
print('gpu:', torch.cuda.get_device_name(0) if torch.cuda.is_available() else 'CPU')
PY

Whisper models (local model/)

• Weights are downloaded automatically on first run into model/.
• You can also pre-download:

.\.venv\Scripts\Activate.ps1
python - << "PY"
import whisper
from pathlib import Path
Path('model').mkdir(exist_ok=True)
for name in ['tiny','base','small','medium']:
    whisper.load_model(name, download_root='model')
print('OK')
PY

Usage (CPU)

.\.venv\Scripts\Activate.ps1
python run_whisper.py --phrase "Appeler Matt" --model small --language fr --verbose

Usage (GPU)

.\.venv\Scripts\Activate.ps1
python run_whisper_gpu.py --phrase "Appeler Matt" --model medium --language en --verbose

What the scripts do

• List all audio files in audio_ref/ (common formats: wav, mp3, m4a, flac, ogg, webm, aac, wma, opus, mkv, mp4)
• Transcribe each file with Whisper (configurable model and language)
• Print and log the full transcription to the terminal and log files
• Check if a target phrase (default "Appeler Matt") is present (accent/case-insensitive)
• Final summary: number of files and matches

Main CLI options

• --dir: audio folder (default: audio_ref)
• --model: tiny | base | small | medium | large (default: small for CPU, medium for GPU)
• --language: Whisper language code (e.g., fr, en) – CPU default: fr, GPU default: en
• --phrase: phrase to search in the transcription
• --verbose: detailed logs

Output & logs

• Terminal: progress, full transcription per file, and phrase search result
• Log files: whisper.log (CPU) and whisper_gpu.log (GPU)
• Models: model/ (e.g., tiny.pt, base.pt, medium.pt)

Structure

Whisper/
  audio_ref/            # Put your audio files here
  model/                # Local model weights (.pt)
  run_whisper.py        # CPU version
  run_whisper_gpu.py    # GPU (CUDA) version
  whisper.log           # CPU logs
  whisper_gpu.log       # GPU logs
  .venv/                # Virtual environment (git-ignored)
  .gitignore
  README.md

Troubleshooting

• AttributeError: module 'whisper' has no attribute 'load_model': do not name your script whisper.py (name clash). Use run_whisper*.py.
• FFmpeg not found: install FFmpeg and add it to PATH (see link above).
• CUDA not available: check PyTorch cu128 install, NVIDIA drivers, and that torch.cuda.is_available() is True.

Publish to GitHub

.gitignore is included to ignore .venv/ and all *.pt model files. Initialize a repo, commit, and push:

git init
git add .
git commit -m "Init Whisper CPU/GPU + scripts + README"
git branch -M main
git remote add origin <REPO_URL>
git push -u origin main