Implementation:Speechbrain Speechbrain Hparams GigaSpeech Conformer Transducer

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

Overview

Hyperparameter configuration for Conformer Transducer ASR training on the GigaSpeech dataset.

Description

HyperPyYAML configuration file that defines the model architecture, training schedule, and data processing pipeline for end-to-end ASR with a Conformer encoder and LSTM transducer decoder with RNN language model on GigaSpeech. The model uses Transducer + CTC + optional CE losses. It supports data splits from XS to XL, HuggingFace-based downloading, and Dynamic Chunk Training for streaming. Training is limited by an optimizer step limit of 500,000 steps. Mixed precision (fp16) is enabled by default.

Usage

Pass this YAML file as the first argument to the corresponding training script.

Code Reference

Source Location

• Repository: SpeechBrain
• File: recipes/GigaSpeech/ASR/transducer/hparams/conformer_transducer.yaml

Key Parameters

seed: 1986
number_of_epochs: 40
optimizer_step_limit: 500000
warmup_steps: 30000
lr: 0.0008
weight_decay: 0.01
ctc_weight: 0.3
ce_weight: 0.0
precision: fp16
batch_size: 8
splits: ["XL", "DEV", "TEST"]

# Feature parameters
sample_rate: 16000
n_fft: 512
n_mels: 80

# Streaming & Dynamic Chunk Training
streaming: True

I/O Contract

Inputs

Name	Type	Required	Description
--data_folder	str	Yes	Path to GigaSpeech dataset

Outputs

Name	Type	Description
Instantiated objects	Python objects	Model, optimizer, scheduler, etc.

Usage Examples

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

Related Pages

• Principle:Speechbrain_Speechbrain_HyperPyYAML_Configuration