Whisper websocket server

whisper_echo is a server service repository, tailored for voice-to-text transcription using the whisper-cpp. The service takes in audio data, processes it through the Whisper system, and returns the corresponding text transcription. Implemented in C++20, specifically designed for Nix environments.

Getting Started

These instructions will guide you on how to clone, build, and test this project in your local Nix environment.

Development

Prerequisites

• Nix package manager installed and configured on your system.

Installing Dependencies

This project utilizes Nix, a potent package manager that simplifies dependency management. Run the following command to install all necessary dependencies:

nix develop

Building

git clone https://github.com/faker2048/whisper_echo.git
cd whisper_echo
mkdir build && cd build
cmake ..
make

Upon successful build, the resulting binary can be found in the build directory.

Run

./bin/whisper_echo ./models/ggml-base.en.bin

Instead ./models/ggml-base.en.bin use your own model_path

Client

Here is a usable Python version of the client example. Run nix develop to enter the Python environment, and then execute python ./whisper_echo/examples/py_client/client.py --file path/to/your_audio.wav.

License

This project is licensed under the MIT License.