tips / recipe guidelines - FLEURS xtreme_s or others

[flutter-painter]

Could you please provide a recipe for FLEURS / xtreme_s ?

[pzelasko]

Could you take a look at the existing recipes in lhotse/recipes and see if you can use any of them as a basis to write your own? If you'd be willing to make a PR, I'm happy to review it.

[flutter-painter]

Thank you for your quick reply,
I looked but lack practice yet
I am gathering more audio first and will dig deeper in a few months

[flutter-painter]

Hi,
I looked again at the recipes but stay puzzled.
I also edited the issue name since there is a wider dataset for fula that goes beyond fleurs here :

• https://huggingface.co/datasets/cawoylel/FulaSpeechCorpora/viewer/default/pular

It is structured in three fields : audio, transcription, dialect
Are there any guidelines as to which recipe could be used this kind of dataset ?
Should I try them randomly until I find one that looks similar to the dataset I intend to use ?

[desh2608]

A "recipe" in Lhotse simply means a script that creates standard manifests (recording manifest and supervisions manifest out of any dataset of your choice. You can take a look at lhotse/recipes to see several examples of how this is done, and then write a recipe yourself for the FLEURS dataset.

[flutter-painter]

Hi Desh,
Thank you for pointing me again at the same link. Could you please name one recipe that you consider a good starting point ?

[desh2608]

You can look at the AISHELL recipe for an example. But keep in mind that each recipe will be different depending on how your data is structured. If you are familiar with Kaldi, think of this as the prepare_data.sh scripts.