Implementation:Speechbrain Speechbrain Hparams Switchboard Transformer LM

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

Overview

Hyperparameter configuration for Transformer ASR with finetuned language model on the Switchboard dataset.

Description

HyperPyYAML configuration file that defines the model architecture, training schedule, and data processing pipeline for end-to-end ASR with a Transformer encoder-decoder and a finetuned Transformer language model on Switchboard conversational telephone speech. This configuration is similar to the base Transformer configuration but is designed for finetuning with a language model, using a reduced learning rate (lr_adam: 0.001), fewer epochs (60), and smaller batch size (32). The model uses CTC + KLdiv (label smoothing) losses with unigram tokenization.

Usage

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

Code Reference

Source Location

• Repository: SpeechBrain
• File: recipes/Switchboard/ASR/transformer/hparams/transformer_finetuned_LM.yaml

Key Parameters

seed: 1312
number_of_epochs: 60
batch_size: 32
ctc_weight: 0.3
grad_accumulation_factor: 2
max_grad_norm: 5.0
loss_reduction: batchmean
sorting: random
avg_checkpoints: 5
lr_adam: 0.001

# Transcript normalization
normalize_words: True
max_utt: 300

I/O Contract

Inputs

Name	Type	Required	Description
--data_folder	str	Yes	Path to Switchboard dataset
--pretrained_lm_tokenizer_path	str	Yes	Path to pre-trained finetuned LM and tokenizer

Outputs

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

Usage Examples

python train.py hparams/transformer_finetuned_LM.yaml --data_folder /path/to/Switchboard

Related Pages

• Principle:Speechbrain_Speechbrain_HyperPyYAML_Configuration