Data extraction
#3
by paksaz - opened
Newbie here. how did data extraction works here? did you use subtitles? how about dub movies? those are also great data with clean voice. But I don't know how to extract them. No ASR also give me 100% confidence and subtitles are not linked to dubs completely. most of them have tiny differences.
In short, how do you labeled the unlabeled sentences?
Thank you
Both filimo-ASR and youtube-ASR are datasets for transcriptions not translations of audio. Chunking is done based on the timing of sentences in a subtitle, so each chunk has an audio segment and a corresponding sentence. There is no labeling.