Implementation:Speechbrain Speechbrain Train GigaSpeech Transducer

Knowledge Sources	SpeechBrain
Domains	ASR, Training
Last Updated	2026-02-09 00:00 GMT

Overview

Concrete tool for training a Transducer ASR model on the GigaSpeech dataset provided by the SpeechBrain library.

Description

This recipe defines the ASR class (subclass of sb.Brain) for Conformer-Transducer speech recognition on GigaSpeech. The architecture uses a Conformer encoder with an LSTM-based transducer decoder and a joint network. Dynamic Chunk Training (streaming support) is optionally enabled. The model is jointly trained with transducer loss, CTC, and optional cross-entropy losses. Beam search decoding coupled with an RNN language model is supported. BPE tokenization via SentencePiece is used for subword units.

Usage

Use this recipe to train a Conformer-Transducer ASR model on the GigaSpeech dataset (supporting XS through XL splits). Requires the corresponding hyperparameter YAML file and data preparation script.

Code Reference

Source Location

• Repository: SpeechBrain
• File: recipes/GigaSpeech/ASR/transducer/train.py

Signature

class ASR(sb.Brain):
    def compute_forward(self, batch, stage):
        """Forward computations from the waveform batches to the output probabilities."""
        ...
    def compute_objectives(self, predictions, batch, stage):
        """Computes the loss given predictions and targets."""
        ...

Import

# Run as recipe script
python recipes/GigaSpeech/ASR/transducer/train.py hparams/conformer_transducer.yaml --data_folder /path/to/GigaSpeech

I/O Contract

Inputs

Name	Type	Required	Description
batch.sig	torch.Tensor	Yes	Input waveform signal
batch.tokens_bos	torch.Tensor	Yes	Target token sequence with BOS prefix
batch.tokens	torch.Tensor	Yes	Target token sequence

Outputs

Name	Type	Description
p_transducer	torch.Tensor	Transducer joint network log-probabilities
p_ctc	torch.Tensor	CTC log-probabilities from encoder (optional)
p_ce	torch.Tensor	Cross-entropy predictions from decoder (optional)
wav_lens	torch.Tensor	Relative waveform lengths
hyps	list	Beam search hypotheses (at validation/test)

Usage Examples

python train.py hparams/conformer_transducer.yaml --data_folder /path/to/GigaSpeech

Related Pages

• Principle:Speechbrain_Speechbrain_Transducer_ASR_Training