Speech to Text
Based on: https://github.com/buriburisuri/speech-to-text-wavenet
I have included the asset folder with pre-trained model which is not included in original repository.
The pre-trained model is from here:
https://github.com/buriburisuri/speech-to-text-wavenet#pre-trained-models
The model was trained on the CSTR VCTK Corpus:
http://homepages.inf.ed.ac.uk/jyamagis/page3/page58/page58.html
Dependencies
-
The original dependancies are not 100% correct, as described here:
https://github.com/buriburisuri/speech-to-text-wavenet#dependencies
It seems to break with newer versions of tensorflow or sugartensor.
My Updated Dependancies File: https://github.com/EN10/STT/blob/master/requirements.txt
Working Dependancies
-
Works with:
pandas 0.19.2 (latest)
librosa to 0.5.0 (latest)
tqdm to 4.11.2 (latest)
tensorflow 1.0.0, 0.12.1 & 0.12.0 doesn’t work, only tensorflow 0.11.0.
sugartensor version > 0.0.1.9 doesn’t work, only 0.0.1.9 does.
Changing Dependancies
-
To see which version installed use:
pip freezepip show tensorflow
If a newer version is installed then uninstall:
sudo pip uninstall sugartensor
Then install correct version:
sudo pip install sugartensor==0.0.1.9
To install correct version of tensorflow:
sudo pip install https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.11.0-cp27-none-linux_x86_64.whl
Run
-
Use recognise using test file:
python recognize.py --file test.wav
Other Issues
-
ImportError: No module named
sudo -H pip install
Convert Audio:
http://superuser.com/questions/23930/how-to-decode-aac-m4a-audio-files-into-wav