Method __init__

wavenet/model.py:46–115

Initializes the WaveNet model. Args: batch_size: How many audio files are supplied per batch (recommended: 1). dilations: A list with the dilation factor for each layer. filter_width: The samples that are included in each convolution,

(self,
                 batch_size,
                 dilations,
                 filter_width,
                 residual_channels,
                 dilation_channels,
                 skip_channels,
                 quantization_channels=2**8,
                 use_biases=False,
                 scalar_input=False,
                 initial_filter_width=32,
                 histograms=False,
                 global_condition_channels=None,
                 global_condition_cardinality=None)

    '''

    def __init__(self,
                 batch_size,
                 dilations,
                 filter_width,
                 residual_channels,
                 dilation_channels,
                 skip_channels,
                 quantization_channels=2**8,
                 use_biases=False,
                 scalar_input=False,
                 initial_filter_width=32,
                 histograms=False,
                 global_condition_channels=None,
                 global_condition_cardinality=None):
        '''Initializes the WaveNet model.

        Args:
            batch_size: How many audio files are supplied per batch
                (recommended: 1).
            dilations: A list with the dilation factor for each layer.
            filter_width: The samples that are included in each convolution,
                after dilating.
            residual_channels: How many filters to learn for the residual.
            dilation_channels: How many filters to learn for the dilated
                convolution.
            skip_channels: How many filters to learn that contribute to the
                quantized softmax output.
            quantization_channels: How many amplitude values to use for audio
                quantization and the corresponding one-hot encoding.
                Default: 256 (8-bit quantization).
            use_biases: Whether to add a bias layer to each convolution.
                Default: False.
            scalar_input: Whether to use the quantized waveform directly as
                input to the network instead of one-hot encoding it.
                Default: False.
            initial_filter_width: The width of the initial filter of the
                convolution applied to the scalar input. This is only relevant
                if scalar_input=True.
            histograms: Whether to store histograms in the summary.
                Default: False.
            global_condition_channels: Number of channels in (embedding
                size) of global conditioning vector. None indicates there is
                no global conditioning.
            global_condition_cardinality: Number of mutually exclusive
                categories to be embedded in global condition embedding. If
                not None, then this implies that global_condition tensor
                specifies an integer selecting which of the N global condition
                categories, where N = global_condition_cardinality. If None,
                then the global_condition tensor is regarded as a vector which
                must have dimension global_condition_channels.

        '''
        self.batch_size = batch_size
        self.dilations = dilations
        self.filter_width = filter_width
        self.residual_channels = residual_channels
        self.dilation_channels = dilation_channels
        self.quantization_channels = quantization_channels