# Create new conda environment
conda create -y --prefix ./env python=3.10 --no-default-packages
# Activate environment
conda activate ./env
# Install requirements
pip install -r requirements.txt
Start a JupyterLab server:
# Activate environment if necessary
# conda activate ./env
jupyter lab
In your browser open the Jupyter Lab web interface and open an ipynb file (e.g. prep-from-eafs_mixtec.ipynb
).
Included in the repo is a copy of the NVIDIA NeMo Speech Data Explorer
pip install -r requirements-sde.txt
Given a manifest JSON file such as (where text
is the human/reference transcription and pred_text
is the machine transcription):
{"audio_filepath": "data/processed/20230818_mixtec/clips/SMD-0049-Medicinas6_00h02m45s110.wav", "duration": 1.2, "text": "ujum", "pred_text": "ujum ujum"}
{"audio_filepath": "data/processed/20230818_mixtec/clips/SMD-0054-Maestro_00h35m17s082.wav", "duration": 1.1, "text": "su si ko an", "pred_text": "txn ixin ko'o o va"}
python nvidia-sde.py tmp/mixtec-manifest-demo.json
You can then browse to the Data Explorer running on the URL reported by the script (e.g. http://0.0.0.0:8050/
):