closed-caption-app/INSTALL_WHISPER.md

Installing Whisper for Local Transcription

This guide explains how to install OpenAI's Whisper for local speech-to-text transcription.

Prerequisites

• Python 3.8 or newer
• pip (Python package manager)
• ffmpeg

Installation Steps

1. Install Python Dependencies

# Install Whisper
pip install openai-whisper

# Or install with specific version
pip install openai-whisper==20230918

2. Install ffmpeg

Ubuntu/Debian:

sudo apt update
sudo apt install ffmpeg

macOS (using Homebrew):

brew install ffmpeg

Windows:

1. Download ffmpeg from https://ffmpeg.org/download.html
2. Extract the archive
3. Add the bin folder to your PATH

3. Verify Installation

# Test Whisper installation
whisper --help

# Test with an audio file
whisper audio.mp3 --model base

Available Models

Whisper offers several models with different speed/accuracy tradeoffs:

Model	Parameters	Relative Speed	Required VRAM
tiny	39 M	~32x	~1 GB
base	74 M	~16x	~1 GB
small	244 M	~6x	~2 GB
medium	769 M	~2x	~5 GB
large	1550 M	1x	~10 GB

Configuration

Set your preferred model in the .env file:

WHISPER_MODEL=base  # Options: tiny, base, small, medium, large

GPU Acceleration Setup

NVIDIA GPUs (CUDA)

# Install PyTorch with CUDA support
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

# Verify CUDA is available
python -c "import torch; print(f'CUDA available: {torch.cuda.is_available()}')"

AMD GPUs (ROCm) - Linux Only

# Install ROCm (Ubuntu/Debian)
wget -q -O - https://repo.radeon.com/rocm/rocm.gpg.key | sudo apt-key add -
echo 'deb [arch=amd64] https://repo.radeon.com/rocm/apt/debian/ ubuntu main' | sudo tee /etc/apt/sources.list.d/rocm.list
sudo apt update
sudo apt install rocm-dkms

# Install PyTorch with ROCm support
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm5.4.2

# Add user to render/video groups
sudo usermod -a -G render,video $USER
# Logout and login for group changes to take effect

Intel GPUs (OpenVINO)

# Install OpenVINO toolkit
pip install openvino-dev[pytorch]

# For Arc GPUs, you may need the compute runtime
# Ubuntu/Debian:
wget -qO - https://repositories.intel.com/graphics/intel-graphics.key | sudo apt-key add -
sudo apt-add-repository 'deb [arch=amd64] https://repositories.intel.com/graphics/ubuntu focal main'
sudo apt update
sudo apt install intel-opencl-icd intel-level-zero-gpu level-zero

# Note: You may need a Whisper fork with OpenVINO support
pip install git+https://github.com/openvinotoolkit/whisper.git

Troubleshooting

Issue: "whisper: command not found"

• Make sure Python's Scripts directory is in your PATH
• On Windows: C:\Users\[Username]\AppData\Local\Programs\Python\Python3X\Scripts
• On macOS/Linux: ~/.local/bin

Issue: GPU not detected

• Use the "Detect GPU" button in the app's settings page
• For NVIDIA: Ensure nvidia-smi command works
• For AMD: Ensure rocm-smi command works
• For Intel: Check lspci | grep -i intel shows your GPU

Issue: CUDA out of memory

• Try using a smaller Whisper model (tiny or base)
• Reduce batch size or use CPU for very long audio

Issue: Audio Format Not Supported

Whisper works best with WAV, MP3, and M4A files. The app automatically converts audio to a compatible format.

Alternative: Docker Installation

You can also run Whisper in a Docker container:

docker pull openai/whisper
docker run -it -v $(pwd):/app openai/whisper audio.mp3 --model base