Implementation:Speechbrain Speechbrain Prepare MEDIA

Knowledge Sources	SpeechBrain
Domains	Spoken_Language_Understanding, Data_Preparation
Last Updated	2026-02-09 00:00 GMT

Overview

Concrete tool for preparing the MEDIA dataset for spoken language understanding provided by the SpeechBrain library.

Description

This script creates CSV data manifest files for the MEDIA dataset, a French spoken language understanding corpus for hotel reservation dialogues. It processes ELRA-distributed transcription/annotation archives (S0272) and audio archives (E0024), parses XML annotation files, extracts concept labels, and generates train/dev/test CSV splits. The script supports both full SLU and relaxed concept annotation schemes, and can optionally process a secondary test set (test2).

Usage

Use this when preparing the MEDIA dataset for spoken language understanding training with SpeechBrain recipes.

Code Reference

Source Location

• Repository: SpeechBrain
• File: recipes/MEDIA/media_prepare.py

Signature

def prepare_media(
    data_folder,
    save_folder,
    channels_path,
    concepts_path,
    skip_wav=True,
    method="slu",
    task="full",
    skip_prep=False,
    process_test2=False,
):

Import

from recipes.MEDIA.media_prepare import prepare_media

I/O Contract

Inputs

Name	Type	Required	Description
data_folder	str	Yes	Path where ELRA folders S0272 and E0024 are stored
save_folder	str	Yes	Path where CSVs and preprocessed WAVs will be stored
channels_path	str	Yes	Path to the channels.csv file
concepts_path	str	Yes	Path to the concepts_full_relax.csv file
skip_wav	bool	No	Skip WAV file extraction if already done (default: True)
method	str	No	Annotation method: "full" or "relax" (default: "slu")
task	str	No	Task type: "full" or other variants (default: "full")
skip_prep	bool	No	If True, skip data preparation (default: False)
process_test2	bool	No	Whether to process the secondary test set (default: False)

Outputs

Name	Type	Description
train.csv	CSV	Training split manifest with audio paths, transcriptions, and concept annotations
dev.csv	CSV	Development split manifest
test.csv	CSV	Test split manifest
test2.csv	CSV	Optional secondary test split manifest (if process_test2=True)

Usage Examples

from recipes.MEDIA.media_prepare import prepare_media

prepare_media(
    data_folder="/path/to/MEDIA",
    save_folder="/path/to/output",
    channels_path="/path/to/channels.csv",
    concepts_path="/path/to/concepts_full_relax.csv",
    skip_wav=False,
    task="full",
)

Related Pages

• Principle:Speechbrain_Speechbrain_Dataset_Specific_Data_Preparation