Implementation:Speechbrain Speechbrain Hparams KsponSpeech Conformer

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

Overview

Hyperparameter configuration for Conformer ASR training on the KsponSpeech 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 Transformer decoder on KsponSpeech Korean data (965.2 hours). The model uses CTC + KLdiv (label smoothing) losses with unigram tokenization. A pre-trained Transformer language model is loaded for beam search decoding. The configuration supports dynamic batching and evaluates on both eval_clean and eval_other test sets.

Usage

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

Code Reference

Source Location

• Repository: SpeechBrain
• File: recipes/KsponSpeech/ASR/transformer/hparams/conformer_medium.yaml

Key Parameters

seed: 7775
number_of_epochs: 60
batch_size: 48
ctc_weight: 0.4
grad_accumulation_factor: 2
max_grad_norm: 5.0
loss_reduction: 'batchmean'
sorting: random
avg_checkpoints: 5
lr_adam: 0.001

# Feature parameters
sample_rate: 16000
n_fft: 400
n_mels: 80

# Pre-trained LM
pretrained_lm_tokenizer_path: ddwkim/asr-conformer-transformerlm-ksponspeech

I/O Contract

Inputs

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

Outputs

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

Usage Examples

python train.py hparams/conformer_medium.yaml --data_folder /path/to/KsponSpeech

Related Pages

• Principle:Speechbrain_Speechbrain_HyperPyYAML_Configuration