This function reads audio file and return data in numpy.float32 array. "lru_cache" holds recently loaded audio so that can be called

PIT loss over mini-batch. Args: ys: B-length list of predictions ts: B-length list of labels Returns: loss: (1,)-shap

Calculates diarization error stats for reporting. Args: pred (torch.FloatTensor): (T,C)-shaped pre-activation values label (torc

returns dictionary { recid: duration }

returns dictionary { spkid: list of uttids }

returns dictionary { uttid: spkid }

return dictionary { rec: wav_rxfilename }

Permutation-invariant training (PIT) cross entropy loss function. Args: pred: (T,C)-shaped pre-activation values label: (T,C)-s

Print variable assignment statements from yaml object. - { key: value } -> key=value - { parent: { child: value } } -> parent_child=v

Reports diarization errors Should be called with torch.no_grad Args: ys: B-length list of predictions (torch.FloatTensor) la

Compute STFT features Args: data: audio signal (n_samples,)-shaped np.float32 array frame_size: number of samples in

Training model with pytorch backend. This function is called from eend/bin/train.py with parsed command-line arguments.

Self-attention-based diarization model. Args: n_speakers (int): Number of speakers in recording in_size (int): Dimension

This function returns generator of segmented audio as (utterance id, numpy.float32 array) TODO?: sampling rate is not converted.

Get frame-aligned labels of given recording Args: kaldi_obj (KaldiData) rec (str): recording id start (int): start frame

Extracts STFT and corresponding labels Extracts STFT and corresponding diarization labels for given recording id and start/end times Ar

load segments file as array

This function returns preprocessed wav_rxfilename Args: wav_rxfilename: input process: command which can be connected via pipe,

Frame splicing Args: Y: feature (n_frames, n_featdim)-shaped numpy array context_size: number of frames

Frame subsampling

Transform STFT feature Args: Y: STFT (n_frames, n_bins)-shaped np.complex array transform_type: None, "l