Datasets:

---

FBK-MT

/

mosel

like 72

Follow

FBK-MT 27

modality:

• text
• audio

Does not work, you should leave them as tags, check here: https://huggingface.co/docs/hub/en/datasets-cards#force-set-a-dataset-modality

Hi!
Files looks good,
few notes, about readme:

1. Seems that Dataset Statistics not updated
2. Dataset creation section lacks of SDP (speech data processor) link, that was used for creation
3. in changelog mentioned YTC addition, but in dataset statistics - no mention (probably related to 1st)

[I can calculate statistics if needed]

Thanks for your comment @Gorgarik .

Dataset statistics are about all the data presented in the MOSEL paper, so they include not only the data transcribed here, but also other data (e.g. ASR datasets already having the transcripts or unlabelled data). We can probably clarify this in the readme and provide updated statistics on the data present in this repo. In this case, if you already computed the number of hours for each section and you can share that would be great.

Do you have a link for the SDP I can refer to?

Thanks.

Hello @mgaido91

https://github.com/NVIDIA/NeMo-speech-data-processor - link to sdp

About statistics - yes, probably better to explain it, cause it can leand to some misunderstanding

Will share duration as a table in an email

Thanks!