This repo is based on the work done here by OpenAI. This repo allows you use use a mic as demo. This repo copies some of the README from original project.
- Create a venv of your choice.
- Run
pip install -r requirements.txt
There are five model sizes, four with English-only versions, offering speed and accuracy tradeoffs. Below are the names of the available models and their approximate memory requirements and relative speed.
| Size | Parameters | English-only model | Multilingual model | Required VRAM | Relative speed |
|---|---|---|---|---|---|
| tiny | 39 M | tiny.en |
tiny |
~1 GB | ~32x |
| base | 74 M | base.en |
base |
~1 GB | ~16x |
| small | 244 M | small.en |
small |
~2 GB | ~6x |
| medium | 769 M | medium.en |
medium |
~5 GB | ~2x |
| large | 1550 M | N/A | large |
~10 GB | 1x |
For English-only applications, the .en models tend to perform better, especially for the tiny.en and base.en models. We observed that the difference becomes less significant for the small.en and medium.en models.
You can use the model with a microphone using the mic.py program. Use -h to see flag options.
Some of the more important flags are the --model and --english flags.
If you are having issues with the mic.py not running try the following:
sudo apt install portaudio19-dev python3-pyaudio
The code and the model weights of Whisper are released under the MIT License. See their repo for more information.
The code under this repo is under the AGPL license. See LICENSE for further details.
Until recently, access to high performing speech to text models was only available through paid serviecs. With this release, I am excited for the many applications that will come.