Instructions to use nwhite872/CHATWhisper-en-mlx with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use nwhite872/CHATWhisper-en-mlx with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir CHATWhisper-en-mlx nwhite872/CHATWhisper-en-mlx
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
Batchalign CHATWhisper - MLX Variant
CHATWhisper is a series of ASR models specifically designed for the task for Language Sample Analysis (LSA) released by the TalkBank project. This unaffiliated set of models was created as a conversion of their models to an MLX version.
The model is based on talkbank/CHATWhisper-en converted using MLX-Communities mlx.convert. This is going to be an unmaintained project as it worked as a one off and is sufficient for my needs.
Usage
The models can be used directly as a Whisper-class ASR model following similar instructions as to those used with the MLX-Whisper package. To get the full analysis possible with the model, it is best combined with my adaptation to the TalkBank Batchalign suite of analysis software.
The original software, available here, is high quality and well documented. My adaptation to the TalkBank Batchalign suite uses batch align as a basis point but adds MLX variants of the pipeline and model, available here.
Data
The models were adapted using mlx-community resources. No further training or modifications were done to the model.
Use
Free to use, no guarantee of maintenance or reliability; however as it is simply a local whisper model reformatted for MPS, it should be sufficient.
In my preliminary testing I had about a 50% to 75% reduction (45 seconds to 8 seconds) in processing time.
- Downloads last month
- -
Quantized
Model tree for nwhite872/CHATWhisper-en-mlx
Base model
talkbank/CHATWhisper-en