This is my solution for the tensorflow speech recognition challenge on kaggle.
- tensorflow v1.4
- scipy for wave loading
- librosa for audio manipulations
- numpy, pandas
- pyaudio for microphone
The model can be trained using the train.py, all the parameters that can be defined can be found in the parser settings or using python train.py --help.
The file settings1.csv contains a definition of the parameters that will be used to run a particular experiment.
Once the model has been trained, it can be used to predict samples using predict.py, passing the previously generated checkpoints.
The file mic.py is used for real-time recognition, using the microphone and the pyAudio library.