Read metadata from the description file (possibly takes long, depending on the filesize) Params: desc_file (str): Pa

For a given audio clip, calculate the log of its Fourier Transform Params: audio_clip(str): Path to the audio clip

Estimate the mean and std of the features from the training set Params: k_samples (int): Use this number of samples for estimatio

Decode a prediction using the highest probable character at each timestep. Then, simply convert the integer sequence to text Params:

Build a testing routine for speech models. Args: model: A keras model (built=True) instance Returns: val_fn (theano.function)

Compute the length of the output sequence after 1D convolution along time. Note that this function is in line with the function used in

Load a model and its weights from a directory Params: load_dir (str): Path the model directory weights_file (str): If this is not

Save the model and costs into a directory Params: save_dir (str): Directory used to store the model model (keras.models.Model)

Validation routine for speech-models Params: model (keras.model): Constructed keras model val_fn (theano.function): A theano func

Build a recurrent network (CTC) for speech with GRU units

Build a function that simply calculates the output of a model Args: model: A keras model (built=True) instance Returns: outpu

Build the CTC training routine for speech models. Args: model: A keras model (built=True) instance Returns: train_fn (theano.

Setup logging. This configures either a console handler, a file handler, or both and adds them to the root logger. Args: console

Plot the training and validation costs over iterations Params: dirs (list(str)): Directories where the model and costs are saved

Compute the spectrogram for a real signal. The parameters follow the naming convention of matplotlib.mlab.specgram Args: sam

Calculate the log of linear spectrogram from FFT energy Params: filename (str): Path to the audio file step (int): Step size in m

Testing routine for speech-models Params: model (keras.model): Constructed keras model test_fn (theano.function): A theano functi

Use a character map and convert text to an integer sequence

Main training routine for speech-models Params: model (keras.model): Constructed keras model train_fn (theano.function): A theano

Get the prediction using the model, and visualize softmax outputs Params: model (keras.models.Model): Trained speech model test_f

Params: step (int): Step size in milliseconds between windows window (int): FFT window size in milliseconds

Featurize a minibatch of audio, zero pad them and return a dictionary Params: audio_paths (list(str)): List of paths to audio fil