paddlespeech.vector.exps.ge2e.speaker_verification_dataset module

class paddlespeech.vector.exps.ge2e.speaker_verification_dataset.Collate(num_frames)[source]

Bases: object

Methods

__call__(examples)

Call self as a function.

class paddlespeech.vector.exps.ge2e.speaker_verification_dataset.MultiSpeakerMelDataset(dataset_root: Path)[source]

Bases: Dataset

A 2 layer directory that contains mel spectrograms in *.npy format. An Example file structure tree is shown below. We prefer to preprocess raw datasets and organized them like this.

dataset_root/

speaker1/: utterance1.npy utterance2.npy utterance3.npy
speaker2/: utterance1.npy utterance2.npy utterance3.npy

Methods

get_example_by_index

get_example_by_index(speaker_index, utterance_index)[source]

class paddlespeech.vector.exps.ge2e.speaker_verification_dataset.MultiSpeakerSampler(dataset: MultiSpeakerMelDataset, speakers_per_batch: int, utterances_per_speaker: int)[source]

Bases: BatchSampler

A multi-stratal sampler designed for speaker verification task. First, N speakers from all speakers are sampled randomly. Then, for each speaker, randomly sample M utterances from their corresponding utterances.

class paddlespeech.vector.exps.ge2e.speaker_verification_dataset.RandomClip(frames)[source]

Bases: object

Methods

__call__(spec)

Call self as a function.