Nendo Plugin Transcribe Whisper
A nendo plugin for speech transcription, based on Whisper by OpenAI.
Features
- Fast speech transcription with optional word-level timestamps.
Requirements
Since we depend on transformers, please make sure that you fulfill their requirements.
You also need Pytorch installed on your system, please refer to the pytorch installation instructions.
Installation
- Install Nendo
pip install nendo-plugin-transcribe-whisper
If you have a cuda GPU on your machine you can also install flash-attn to get an additional speedup:
pip install flash-attn --no-build-isolation
Then set ATTN_IMPLEMENTATION=flash_attention_2 in your environment variables.
Usage
>>> from nendo import Nendo
>>> nd = Nendo(plugins=["nendo_plugin_transcribe_whisper"])
>>> track = nd.library.add_track(file_path="path/to/file.mp3")
>>> nd.plugins.transcribe_whisper(track=track)
>>> track.get_plugin_value("transcription")Contributing
Visit our docs to learn all about how to contribute to Nendo: Contributing
License
Nendo: MIT License
Pretrained models: The weights are released under the Apache 2.0 license.
