Implementation:Speechbrain Speechbrain Train TimersAndSuch Decoupled

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

Overview

Concrete tool for decoupled (speech to ASR to text to NLU to semantics) SLU training on the Timers-and-Such dataset provided by the SpeechBrain library.

Description

This recipe defines the SLU class (subclass of sb.Brain) for decoupled spoken language understanding on the Timers-and-Such dataset. The NLU component is trained on ground truth transcripts during training and validation, while at test time an ASR model first transcribes the audio and then the transcript is fed to the NLU. ASR tokens are encoded with an input embedding, passed through an SLU encoder, and decoded with a seq2seq decoder to produce semantic outputs. This decoupled approach allows separate optimization of ASR and NLU components.

Usage

Use this recipe to train a decoupled SLU model on the Timers-and-Such dataset. Requires the dataset and a pre-trained ASR model (used only at test time). Configure with hparams/train.yaml.

Code Reference

Source Location

• Repository: SpeechBrain
• File: recipes/timers-and-such/decoupled/train.py

Signature

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

Import

python recipes/timers-and-such/decoupled/train.py hparams/train.yaml --data_folder /path/to/timers-and-such

I/O Contract

Inputs

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

Outputs

Name	Type	Description
predictions	tuple	Log-softmax sequence probabilities, asr_tokens_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/timers-and-such

Related Pages

• Principle:Speechbrain_Speechbrain_Spoken_Language_Understanding