Open mohsenoon opened 1 year ago
Hi, [tmp note : This answer will be updated as much as I can. when todo is completed this line will be removed]
I'll write here my todo list to update and edit repo as soon as possible:
Thanks to @mohsenoon for analysing data
Update 1 : 3/6/2023
new Stats for data is ready, these stats are not 100% accurate but they are accurate enough. you can trust these numbers :
Total Hours : 1697.1423399942473 Hour Total size : 195510797567.33728 bytes duration mean : 4.834937608942991 second size : 154718.00348617567 bytes
Hi Masoud, I have downloaded all parts of version 2, but after removing duplicated metadata from CSVs, the remaining dataset consists of only 625h of audio clips, not 1697h. What do you think is the problem?
@hamjam hi, sorry for the late response, can you add the duration of data v1 too, and say how much is data when both versions are added?
then if my information is wrong, send a pull request for reamdefile and correct it.
thanks for your attention
Apart from data size, please publish other information such as duration, number of files, average length of files, quality, number of speakers, source, and method of collection.
Also, since these data are Google's speech-to-text transcriptions, it is better to report this issue and its approximate error. The raw outputs of a speech-to-text model can be used with some considerations to train other models, but it certainly cannot be introduced as a speech-to-text dataset.