Summarize the error_rate and return relevant statistics. * See MetricStats.summarize()

Write all relevant info (e.g., error rate alignments) to file. * See MetricStats.write_stats()

Given an input batch it computes the phoneme probabilities.

Given an input batch it computes the phoneme probabilities.

Given the network predictions and targets computed the NLL loss.

Given the network predictions and targets computed the NLL loss.

Normalize phoneme labels to lower case, stress-free form. This will also deal with L2-ARCTIC annotations. Args: s: A phoneme annotatio

a, b are two aligned lists (i.e. same length) gap_token is the artificial token placeholder used in L2Arctic annotation. In this case is a `s

Iterate epochs and datasets to improve objective. Relies on the existence of multiple functions that can (or should) be overridden. Th

wav_lens: relative lengths(i.e. 0-1) of a batch. shape: (bs, ) return a tensor of shape (bs, seq_len), representing mask on allowed positions

Normalize the marks of an IntervalTier. Refer to the code for supported modes. Args: tier: An IntervalTier object. mode: The f

Gets called when a stage (either training, validation, test) starts.

Given an input batch it computes the phoneme probabilities.

Given an input batch it computes the phoneme probabilities.

Given the network predictions and targets computed the NLL loss.

Simply compute the CTC loss

This function prepares the datasets to be used in the brain class. It also defines the data processing pipeline through user-defined functions.

This function prepares the datasets to be used in the brain class. It also defines the data processing pipeline through user-defined functions.

This function prepares the datasets to be used in the brain class. It also defines the data processing pipeline through user-defined functions.

This function prepares the datasets to be used in the brain class. It also defines the data processing pipeline through user-defined functions.

Computations needed for validation/test batches

Fit one batch, override to do multiple updates. The default implementation depends on a few methods being defined with a particular b

Initializes the wav2vec2 optimizer and model optimizer

Initializes the wav2vec2 optimizer and model optimizer

Test if the input string represents silence. Args: s: A phoneme label. Returns: True if is silence, otherwise False.

wav_lens: relative lengths(i.e. 0-1) of a batch. shape: (bs, ) return a tensor of shape (bs, seq_len), representing mask on allowed positions

wav_lens: relative lengths(i.e. 0-1) of a batch. shape: (bs, ) return a tensor of shape (bs, seq_len), representing mask on allowed positions

wav_lens: relative lengths(i.e. 0-1) of a batch. shape: (bs, ) return a tensor of shape (bs, seq_len), representing mask on allowed positions

check whether the wav is presented in labled_data we only keep those wav that were not labled.

schema: [(operator, idx_i(None), idx_j(None))] c: canonical p: perceived h: hypothesis

Gets called at the beginning of ``fit()``, on multiple processes if ``distributed_count > 0`` and backend is ddp. Default implementat

Gets called at the end of a epoch.

Computations needed for validation/test batches

Computations needed for validation/test batches

Fit one batch, override to do multiple updates. The default implementation depends on a few methods being defined with a particular b

word_tier: [(minTime, maxTime, word1), (minTime, maxTime, word2), ...] phone_tier: [(minTime, maxTime, phn1), (minTime, maxTime, phn2), ...]

Gets called at the beginning of ``fit()``, on multiple processes if ``distributed_count > 0`` and backend is ddp. Default implementat

Gets called at the end of a epoch.

Gets called at the end of a epoch.

Gets called when a stage (either training, validation, test) starts.

Gets called when a stage (either training, validation, test) starts.