torchaudio.prototype.pipelines

The pipelines subpackage contains APIs to models with pretrained weights and relevant utilities.

RNN-T Streaming/Non-Streaming ASR

Pretrained Models

EMFORMER_RNNT_BASE_MUSTC	Pre-trained Emformer-RNNT-based ASR pipeline capable of performing both streaming and non-streaming inference.
EMFORMER_RNNT_BASE_TEDLIUM3	Pre-trained Emformer-RNNT-based ASR pipeline capable of performing both streaming and non-streaming inference.

HiFiGAN Vocoder

Interface

HiFiGANVocoderBundle defines HiFiGAN Vocoder pipeline capable of transforming mel spectrograms into waveforms.

HiFiGANVocoderBundle

Data class that bundles associated information to use pretrained HiFiGANVocoder.

Pretrained Models

HIFIGAN_VOCODER_V3_LJSPEECH

HiFiGAN Vocoder pipeline, trained on The LJ Speech Dataset [Ito and Johnson, 2017].

Squim Objective

Interface

SquimObjectiveBundle defines speech quality and intelligibility measurement (SQUIM) pipeline that can predict objecive metric scores given the input waveform.

SquimObjectiveBundle

Data class that bundles associated information to use pretrained SquimObjective model.

Pretrained Models

SQUIM_OBJECTIVE

SquimObjective pipeline trained using approach described in [Kumar et al., 2023] on the DNS 2020 Dataset [Reddy et al., 2020].

Squim Subjective

Interface

SquimSubjectiveBundle defines speech quality and intelligibility measurement (SQUIM) pipeline that can predict subjective metric scores given the input waveform.

SquimSubjectiveBundle

Data class that bundles associated information to use pretrained SquimSubjective model.

Pretrained Models

SQUIM_SUBJECTIVE

SquimSubjective pipeline trained as described in [Manocha and Kumar, 2022] and [Kumar et al., 2023] on the BVCC [Cooper and Yamagishi, 2021] and DAPS [Mysore, 2014] datasets.