Remove weight normalization module from all of the layers.

Run training.

Computes the similarity matrix according the section 2.1 of GE2E. :param embeds: the embeddings as a tensor of shape (speakers_per_b

Derives a mel spectrogram ready to be used by the encoder from a preprocessed audio waveform. Note: this not a log-mel spectrogram. wav:

Derives a mel spectrogram ready to be used by the encoder from a preprocessed audio waveform. Note: this not a log-mel spectrogram. wav:

WORLD feature extraction Args: queue (multiprocessing.Queue): the queue to store the file name of utterance wav_list (list): list

Get ndarray for a given key.

Get ndarray for a given key.

Get specified idx items. Args: idx (int): Index of the item. Returns: str: Utterance id (only in return_utt_

Get specified idx items. Args: idx (int): Index of the item. Returns: str: Utterance id (only in return_utt_

Get specified idx items. Args: idx (int): Index of the item. Returns: str: Utterance id (only in return_utt_

Initialize trainer. Args: steps (int): Initial global steps. epochs (int): Initial global epochs. data_lo

Initialize HDF5 scp loader. Args: feats_scp (str): Kaldi-style feats.scp file with hdf5 format. default_hdf5_path (st

Initialize npy scp loader. Args: feats_scp (str): Kaldi-style feats.scp file with npy format.

Initilize PQMF module. The cutoff_ratio and beta parameters are optimized for #subbands = 4. See dicussion in https://github.com/kan-

Initialize 1x1 Conv1d module.

Initialize ResidualBlock module. Args: kernel_size (int): Kernel size of dilation convolution layer. residual_channel

Initialize ResidualBlock module. Args: kernel_size (int): Kernel size of dilation convolution layer. residual_channel

Initialize ResidualStack module. Args: kernel_size (int): Kernel size of dilation convolution layer. channels (int):

Initialize Stretch2d module. Args: x_scale (int): X scaling factor (Time axis in spectrogram). y_scale (int): Y scali

Initialize upsampling network module. Args: upsample_scales (list): List of upsampling scales. nonlinear_activation (

Initialize convolution + upsampling network module. Args: upsample_scales (list): List of upsampling scales. nonlinea

Initialize CausalConvTranspose1d module.

Initialize TFReflectionPad1d module. Args: padding_size (int): Padding size.

Initialize TFConvTranspose1d( module. Args: channels (int): Number of channels. kernel_size (int): kernel size.

Initilize RAdam optimizer.

Initialize customized collater for PyTorch DataLoader. Args: batch_max_steps (int): The maximum length of input signal in batch.

Initialize customized collater for PyTorch DataLoader. Args: batch_max_steps (int): The maximum length of input signal in batch.

Initialize dataset. Args: root_dir (str): Root directory including dumped files. audio_query (str): Query to find aud

Initialize dataset. Args: root_dir (str): Root directory including dumped files. audio_query (str): Query to find aud

Initialize dataset. Args: root_dir (str): Root directory including dumped files. mel_query (str): Query to find featu

Initialize SingerConditional Discriminator module.

Initilize spectral convergence loss module.

Initilize los STFT magnitude loss module.

Initialize Multi resolution STFT loss module. Args: fft_sizes (list): List of FFT sizes. hop_sizes (list): List of ho

Return the iterator of the scp file.

Return the iterator of the scp file.

Return the length of the scp file.

Return the length of the scp file.

Return dataset length. Returns: int: The length of dataset.

Return dataset length. Returns: int: The length of dataset.

Return dataset length. Returns: int: The length of dataset.

Set state.

Adjust the audio and feature lengths. Note: Basically we assume that the length of x and c are adjusted through prepr

merge acoustic features 最终生成的特征为[lf0, lf0与delta的卷积， lf0与acc的卷积， mgc， mgc与delta的卷积， mgc与acc的卷积， bap， bap与delta的卷积， bap与acc的卷积， vuv]

Calculate forward propagation. Args: x (Tensor): Input tensor (B, T, 1, C). Returns: Tensor: Padded tensor (

Calculate forward propagation. Args: x (Tensor): Input tensor (B, T, 1, C). Returns: Tensors: Output tensor

Calculate forward propagation. Args: x (Tensor): Input tensor (B, T, 1, C). Returns: Tensor: Output tensor (

Download pretrained model form google drive. Args: tag (str): Pretrained model tag. download_dir (str): Directory to save downloa

PARAMS ------ C: compression factor

PARAMS ------ C: compression factor used to compress

Computes an embedding for a single utterance. # TODO: handle multiple wavs to benefit from batching on GPU :param wav: a preprocessed (s

Computes an embedding for a single utterance. # TODO: handle multiple wavs to benefit from batching on GPU :param wav: a preprocessed (s

Calculate forward propagation. Args: x (Tensor): Input tensor (B, residual_channels, T). c (Tensor): Local conditioni

Calculate forward propagation. Args: x (Tensor): Input tensor (B, residual_channels, T). c (Tensor): Local conditioni

Calculate forward propagation. Args: c (Tensor): Input tensor (B, channels, T). Returns: Tensor: Output tens

Calculate forward propagation. Args: x (Tensor): Input tensor (B, C, F, T). Returns: Tensor: Interpolated te

Calculate forward propagation. Args: c : Input tensor (B, C, T). Returns: Tensor: Upsampled tensor (B, C, T'

Calculate forward propagation. Args: c : Input tensor (B, C, T'). Returns: Tensor: Upsampled tensor (B, C, T

Calculate forward propagation. Args: x (Tensor): Input tensor (B, in_channels, T). Returns: Tensor: Output t

Calculate forward propagation. Args: x (Tensor): Input tensor (B, in_channels, T_in). Returns: Tensor: Outpu

Calculate forward propagation. Args: x (Tensor): Input noise signal (B, 1, T). Returns: Tensor: Output tenso

Calculate forward propagation. Args: x (Tensor): Input noise signal (B, 1, T). embed (Tensor): Local conditioning aux

Calculate forward propagation. Args: x_mag (Tensor): Magnitude spectrogram of predicted signal (B, #frames, #freq_bins).

Calculate forward propagation. Args: x_mag (Tensor): Magnitude spectrogram of predicted signal (B, #frames, #freq_bins).

Calculate forward propagation. Args: x (Tensor): Predicted signal (B, T). y (Tensor): Groundtruth signal (B, T).

Calculate forward propagation. Args: x (Tensor): Predicted signal (B, T). y (Tensor): Groundtruth signal (B, T).

Computes the embeddings of a batch of utterance spectrograms. :param utterances: batch of mel-scale filterbanks of same duration as

Get hdf5 file path for a given key.