Implementation:Speechbrain Speechbrain Train SLURP NLU

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

Overview

Concrete tool for text-only Natural Language Understanding (NLU) training on the SLURP dataset provided by the SpeechBrain library.

Description

This recipe defines the SLU class (subclass of sb.Brain) for text-only NLU on the SLURP dataset. It takes golden ASR transcriptions as input and estimates semantics using a seq2seq architecture. The pipeline embeds transcript tokens, encodes them with an SLU encoder, then decodes with an attention-based decoder to produce semantic output sequences. Beam search is used for inference, and evaluation uses SLURP-specific metrics including scenario accuracy, action accuracy, and intent accuracy reported via jsonlines output.

Usage

Use this recipe to train a text-based NLU model on the SLURP dataset using ground truth transcriptions. Requires the SLURP data folder with pre-processed transcripts. Configure with hparams/train.yaml.

Code Reference

Source Location

• Repository: SpeechBrain
• File: recipes/SLURP/NLU/train.py

Signature

class SLU(sb.Brain):
    def compute_forward(self, batch, stage):
        ...
    def compute_objectives(self, predictions, batch, stage):
        ...

Import

python recipes/SLURP/NLU/train.py hparams/train.yaml --data_folder /path/to/SLURP

I/O Contract

Inputs

Name	Type	Required	Description
batch	PaddedBatch	Yes	Batch containing transcript_tokens, semantics_tokens_bos, and semantics_tokens_eos
stage	sb.Stage	Yes	TRAIN, VALID, or TEST

Outputs

Name	Type	Description
predictions	tuple	Log-softmax sequence probabilities, transcript lens, and decoded semantic tokens
loss	torch.Tensor	Sequence-level NLL loss on semantic tokens

Usage Examples

python train.py hparams/train.yaml --data_folder /path/to/SLURP

Related Pages

• Principle:Speechbrain_Speechbrain_Spoken_Language_Understanding