Append new item to the BeamState.

Load the model from a file. Args: filepath: the path of the file.

Save the model to a file. Args: filepath: the path of the file.

Encodes a sentence to token IDs Args: sentence (str): Sentence in text form bos_token (bool, op

Fit UISRNN model. Args: train_sequences: Either a list of training sequences, or a single concatenated training s

Args: x - Tensor of raw waveform [batch, length]

Decodes a sequence of IDs into a string Args: tokens (Union[list, torch.LongTensor]): Tokens sequence

Print a message if level is not higher than verbosity. Args: level: the level of this message, smaller value means more important

Generates text from encoder features. Args: audio_x: Input audio (wav). [batch, seq len] generated: The i

Checks if a single utterance is valid

Loads an audio segment and uses local cache

Predict labels for a single test sequence using UISRNN model. Args: test_sequence: the test observation sequence, which is 2-dim

Update a beam state given a look ahead sequence and known cluster assignments. Args: beam_state: A BeamState object.

Compute the accuracy between two sequences by finding optimal matching. Args: sequence1: A list of integers or strings. sequenc

Converts an audio file to consistent specifications Args: in_file (str): Input file location (raw) out_file (str): Output fi

Input - list of tuples: (utterance string, speaker ID) OR (utterance string, speaker vectors) Return - list of t

12/4/2019 - Shuyang Li Debug logging of an object. Usage: from functools import partial debug_fxn = partial(debug_log, debug=True)

Decode a list of IDs Args: tokens (list): List of token IDs Returns: str: Ou

Args: y_prev: [batch, seq_len]

Generates audio frames from PCM audio data. Takes the desired frame duration in milliseconds, the PCM data, and the sample rate. Yields

Get audio information for an audio file Args: audio_loc (str): Target audio file path audio_fmt (str): If 'wav', process dif

Get value to position index from a list of unique ids. Args: unique_ids: A list of unique integers of strings. Returns: result

Get value to position index from a list of unique ids. Args: unique_ids: A list of unique integers of strings. Returns: result

Get value to position index from a list of unique ids. Args: unique_ids: A list of unique integers of strings. Returns: re

Get value to position index from a list of unique ids. Args: unique_ids: A list of unique integers of strings. Returns: re

Get value to position index from a list of unique ids. Args: unique_ids: A list of unique integers of strings. Returns: result: a dict f

Retrieve a trained Wav2Vec model from local cache Args: cache_path (str): Cache directory Returns: Wav2VecModel: Traine

Given edits, speakers, and roles, compute WDER (and attribute WDER to roles!)

Given edits, speakers, and roles, compute WDER (and attribute WDER to roles!)

Given edits, speakers, and roles, compute WDER (and attribute WDER to roles!)

Predict labels for a single or many test sequences using UISRNN model. Args: test_sequences: Either a list of test sequences, or

Run a single experiment Args: train_sequence (np.array): Sequence of T x H arrays (audio features) train_cluster_id (np.arra

Performs a single optimization step. Arguments: closure (callable, optional): A closure that reevaluates the model

Args: tokenize_speaker: Function to tokenize speaker

Filters out non-voiced audio frames. Given a webrtcvad.Vad and a source of audio frames, yields only the voiced audio. Uses a padded, s

Writes a .wav file. Takes path, PCM audio data, and sample rate.

Construct the UISRNN object. Args: args: Model configurations. See `arguments.py` for details.

Calculate negative log likelihoods for all possible state allocations of a look ahead sequence, according to the current beam state.

Encodes a sentence to token IDs Args: sentence (str): Sentence in text form Returns:

Get optimizer for UISRNN. Args: optimizer: string - name of the optimizer. learning_rate: - learning rate for the ent

Feature aggregation network RF: 141 frames (1.41s) or 61 extracted features.

https://arxiv.org/pdf/1907.05337.pdf WDER = ( (# ASR substitutions, corect speakers) + (# correct ASR, correct speakers) ) ----

Compute the accuracy between two sequences by finding optimal matching. Args: sequence1: A list of integers or strings. sequence2:

Compute the accuracy between two sequences by finding optimal matching. Args: sequence1: A list of integers or strings. sequence2:

Compute the accuracy between two sequences by finding optimal matching. Args: sequence1: A list of integers or strings. sequenc

Converts an audio file to consistent specifications Args: in_file (str): Input file location (raw) out_file (str): Outp

Converts a .sph file into a normal output format Args: sph2pipe_bin (str): Name with which to invoke sph2pipe utility in_fil

Args: y_prev: [batch, seq_len]

Decodes a sequence of tokens. Tokens outside of the tokenizer range are treated as speaker IDs. Returns: [(s

Experiment pipeline. Load data --> train model --> test model --> output result Args: model_args: model configurations training_args

Experiment pipeline. Load data --> train model --> test model --> output result Args: model_args: model configurations training_args