runngezhang / speech_recognition

"# speech_recognition"

Summary:

During the work two models were observed: Simple Audio Recognition from Tensorflow and the model similar to introduced in https://yerevann.github.io/2015/10/11/spoken-language-identification-with-deep-convolutional-networks/. Both of them were trained using dataset used in speech_commands example from Tensorflow.

Steps to reproduce:

For the first model:

1. Install Docker on your machine
2. Pull Tensorflow Docker image using following command: docker pull tensorflow/tensorflow
3. Run container: docker run -it -p 8888:8888 tensorflow/tensorflow
4. Go through the steps from the tutorial: https://www.tensorflow.org/versions/master/tutorials/audio_recognition

For the second model:

1. Install Anaconda2 package from https://www.anaconda.com/download/
2. Go to the Scripts directory and run: conda install -c anaconda theano
   conda install -c anaconda lasagne
3. Download the files from the repository
4. Copy the dataset from the Docker container to Host
5. Change the png_folder and listfile paths in theano/main.py to your ones
6. Create spectrograms for all of the files from dataset using create_spectrograms.py
7. Run theano/main.py.

Link to presentation: https://drive.google.com/open?id=0BzosZ0Y6TKpHT0thYlB0NlVESTQ