Implementation:Speechbrain Speechbrain Train SLURP Direct

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

Overview

Concrete tool for direct (speech to semantics) SLU training with ASR-based transfer learning on the SLURP dataset provided by the SpeechBrain library.

Description

This recipe defines the SLU class (subclass of sb.Brain) for direct spoken language understanding on the SLURP dataset. Input waveforms are encoded into features using a pre-trained ASR model (from LibriSpeech), then the features are passed through an SLU encoder and a seq2seq decoder to map directly to semantic outputs. The ASR encoder is frozen during training. Supports waveform augmentation with label replication for augmented samples, beam search decoding, and SLURP-specific evaluation metrics (scenario, action, intent accuracy).

Usage

Use this recipe to train a direct SLU model on the SLURP dataset using ASR-based transfer learning. Requires the SLURP data folder and a pre-trained ASR model checkpoint. Configure with hparams/train.yaml.

Code Reference

Source Location

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

Signature

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

Import

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

I/O Contract

Inputs

Name	Type	Required	Description
batch	PaddedBatch	Yes	Batch containing sig (waveforms), tokens_bos, and tokens_eos (semantic tokens)
stage	sb.Stage	Yes	TRAIN, VALID, or TEST

Outputs

Name	Type	Description
predictions	tuple	Log-softmax sequence probabilities, wav_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