docker built -t speech-command .
docker run -it -p 8888:8888 \
-v /home/qianhui/Dataset/speech_commands_v0.01/:/data
-v $PWD:/home/workspace
speech-command /bin/bash python bin/preprocess.py \
--data-path /data/ \
--output-path metadata/This will create three metadata csv files (for train, validation and test respectively) inside OUT-DIR, each csv file has two columns: audio file path and label.
Use the following command to train model, make sure you run preprocess.py first.
python bin/train.py \
--train-metadata metadata/metadata_train.csv \
--validation-metadata metadata/metadata_val.csv \
--config configs/config.yaml \
--output outputs/Use --resume to continue training from a previously saved checkpoint.
mlflow uipython bin/predict.py \
--audio /data/cat/0819edb0_nohash_0.wav \
--model models/best-accuracy-229896.pth \
--config configs/config.yaml