pywhispercpp

To enable CoreML support (typically for macOS hardware acceleration), set the WHISPER_COREML=1 environment variable during the pip installation from source.

WHISPER_COREML=1 pip install git+https://github.com/absadiki/pywhispercpp

To enable Vulkan support, set the GGML_VULKAN=1 environment variable during the pip installation from source.

GGML_VULKAN=1 pip install git+https://github.com/absadiki/pywhispercpp

To transcribe audio files other than .wav, you must have ffmpeg installed on your system. Use the command corresponding to your operating system:

# on Ubuntu or Debian
sudo apt update && sudo apt install ffmpeg

# on Arch Linux
sudo pacman -S ffmpeg

# on MacOS using Homebrew
brew install ffmpeg

# on Windows using Chocolatey
choco install ffmpeg

# on Windows using Scoop
scoop install ffmpeg

To use OpenVINO, follow the official whisper.cpp OpenVINO setup steps to download the toolkit. Then, initialize the environment and install with WHISPER_OPENVINO=1.

source ~/l_openvino_toolkit_ubuntu22_2023.0.0.10926.b4452d56304_x86_64/setupvars.sh 
WHISPER_OPENVINO=1 pip install git+https://github.com/absadiki/pywhispercpp --no-cache --force-reinstall

For optimal performance, install the package directly from the GitHub repository. This allows the build process to optimize for your specific hardware and backend requirements.

pip install git+https://github.com/absadiki/pywhispercpp

If OpenBLAS is installed on your system, you can use GGML_BLAS=1. It is recommended to use --no-cache and --force-reinstall to ensure a fresh build with the correct flags.

GGML_BLAS=1 pip install git+https://github.com/absadiki/pywhispercpp --no-cache --force-reinstall -v

A graphical interface built with PyQt5 is available. First, install the GUI dependencies, then run the pwcpp-gui command.

# Install GUI dependencies
pip install pywhispercpp[gui]

# Launch the GUI
pwcpp-gui

Use the Model class to load a Whisper model and transcribe audio. Models are downloaded automatically if not found locally. You can pass any whisper.cpp parameter as a keyword argument to the Model constructor or the transcribe method.

from pywhispercpp.model import Model

# Initialize model (e.g., 'base.en')
model = Model('base.en')

# Transcribe a file
segments = model.transcribe('file.wav')
for segment in segments:
    print(segment.text)

You can install basic CPU-only wheels from PyPI. If you want to run the provided examples, install the [examples] extra to include necessary dependencies.

# Basic CPU installation
pip install pywhispercpp

# Installation with example dependencies
pip install pywhispercpp[examples]

To enable CUDA support, ensure you have NVIDIA CUDA installed, then set the GGML_CUDA=1 environment variable during the pip installation from source.

GGML_CUDA=1 pip install git+https://github.com/absadiki/pywhispercpp

The Assistant class uses a Voice Activity Detector (VAD) to detect speech and trigger transcription. You can use it programmatically or via the pwcpp-assistant CLI.

from pywhispercpp.examples.assistant import Assistant

# Initialize assistant with a callback for commands
my_assistant = Assistant(commands_callback=print, n_threads=8)
my_assistant.start()

You can provide a new_segment_callback to the transcribe method to handle segments as they are produced (e.g., for real-time printing).

from pywhispercpp.model import Model

model = Model('base.en', print_realtime=False, print_progress=False)
segments = model.transcribe('file.mp3', new_segment_callback=print)