Fluentd filter plugin for transcribing audio files using MLX Whisper.
This filter plugin transcribes audio content using MLX Whisper and adds the transcription text to the record.
Note: This plugin is designed to work only on macOS as it relies on MLX, which is optimized for Apple Silicon.
$ gem install fluent-plugin-audio-transcriber
Add the following line to your Gemfile:
gem "fluent-plugin-audio-transcriber"And then execute:
$ bundle
- macOS (the plugin uses MLX which is optimized for Apple Silicon)
- Ruby 3.4 or higher
- Fluentd v1.0.0 or higher
- Python 3.11 or higher
- MLX Whisper installed in a Python environment
You need to set up a Python virtual environment with MLX Whisper installed:
# Install Python using pyenv on macOS
brew install pyenv
pyenv init
pyenv install 3.11
pyenv rehash
pyenv local 3.11
# Create Python virtual environment and install required packages
# Note: The virtual environment must be created in the same directory as your Fluentd configuration
python -m venv myenv
source myenv/bin/activate
pip install mlx-whisper<filter audio.raw>
@type audio_transcriber
# MLX Whisper settings
model mlx-community/whisper-large-v3-turbo # Transcription model
language ja # Language code
initial_prompt This is a transcription of Japanese business meetings and technical discussions. Please accurately recognize Japanese expressions, honorific language, technical terms, and proper nouns.
</filter>
| Parameter | Type | Description | Default |
|---|---|---|---|
| model | string | MLX Whisper model name | mlx-community/whisper-large-v3-turbo |
| language | string | Language code for transcription | ja |
| initial_prompt | string | Initial prompt to guide transcription | A detailed Japanese prompt (see source code) |
The plugin expects records with the following field:
{
"content": <binary data of audio file>,
... (other fields)
}
The plugin processes the record and modifies it to:
{
"transcription": "Transcribed text line 1\nTranscribed text line 2\n...",
"speech_recognition_model": "mlx-community/whisper-large-v3-turbo",
"transcription_language": "ja",
... (other original fields except 'content')
}
Note: The content field is removed from the output record to reduce data size.
Make sure the Python virtual environment (myenv) exists in the same directory where Fluentd is running.
After checking out the repo, run bundle install to install dependencies. Then, run bundle exec rake test to run the tests.
To release a new version, update the version number in the gemspec file, and then run bundle exec rake release, which will create a git tag for the version, push git commits and tags, and push the .gem file to rubygems.org.
Bug reports and pull requests are welcome on GitHub at https://github.com/bash0C7/fluent-plugin-audio-transcriber.