Gaussian Error Linear Unit. This is a smoother version of the RELU. Original paper: https://arxiv.org/abs/1606.08415 Args: x:

generate window */

Extract features (mel, energy, f0) and return feat dict feat_dict = {utt_id:{[feat_file, wav_length]}}

generate mask for pretraining Args: acoustic features: i.e F-bank Return:: mask tensor

Image Normalization http://dev.ipol.im/~nmonzon/Normalization.pdf

entry point for model inference, do some preparation work

decode the model

Calculate replicated sequences based on duration sequences Args: phoneme_sequences: sequences of phoneme features, shape: [batch,

decode the model

Args: value_length: the length of value, shape: [batch] max_len: the maximun length Returns: init

Insert blank token between every two label token.

load mean and var

load mel_mean, mel_var, f0_mean, f0_var and energy_mean, energy_var

perform downsampling using conv layer with stride != 1

naive DFT */

Retrieves the next task from the queue

Retrieves the next task from the queue

normalize HKUST transcripts delete unuseful symbols and keep space between English words Args: trans: original transcripts

normalize numbers in text

Parses hyperparameter values from a string into a python map. `values` is a string containing comma-separated `name=value` pairs. For each pa

Custom pipeline for English text, including number and abbreviation expansion.

generate a list of tuples (wav_filename, wav_length_ms, speaker).

loading data

load csv file

generate a list of tuples (wav_filename, wav_length_ms, transcript, speaker).

generate a list of tuples (wav_filename, wav_length_ms, transcript, speaker).

generate a list of tuples (feat_key, speaker).

generate a list of tuples (wav_filename, wav_length_ms, speaker).

generate a list of tuples (wav_filename, wav_offset, wav_length_ms, transcript, label).

generate a list of tuples (audio_feature, wav_length_ms, transcript, duration, speaker).

generate a list of tuples (wav_filename, wav_length_ms, transcript, speaker).

generate a list of tuples (utt_id, transcript).

Generate a list of tuples (feat_key, speaker).

process decode log, generate label file and results files

Convert phonelabeling to phone_dur dict Args: dataset_dir Returns: phone_dur_dict: {utt_id:[phone_str, interval_str]}

Convert phonelabeling to phone dict Args: dataset_dir Returns: phone_dict: {sent_id:[phone_str]}

Downsampling wav and return wav dict

download and process

download and process

download and process

download and process

download and process

remove unnecessary symbols in text

restore from pretrained model

text_list: list of char

segment transcripts according to vocab using Maximum Matching Algorithm Args: vocab_file: vocab file text_file: transcripts f