🔢 Number of sentences we need to create a good Chinese STT model

[nukeador (Rubén Martín [❌ taking a break from Mozilla])]

@irvin

We already cover 55% of Chinese pronunciation with only 3600 sentences (and 75% if we don’t consider tone), I really don’t think we need 2M of not-so-fluency sentences over 20K of good-quality easy-reading (and sometimes - funny enough) sentences.

I asked about this and we have been told a few times that 1 sentence per clip needed is what we need, and for 2000hrs that’s 1,8M sentences. Maybe @kdavis or @josh_meyer can chime in here and provide more context.