2 ms·

I'm looking forward to great speech-to-text technology. I help run fan-driven transcripts for a podcast, and I recently tried using Google's; it returned someth

I'm looking forward to great speech-to-text technology. I help run fan-driven transcripts for a podcast, and I recently tried using Google's; it returned something like the hosts were concussed, or Markov models of themselves. Edit: the word salad result: http://mefiwiki.com/wiki/Podcast_107_Transcript_%28automated%29 http://mefiwiki.com/wiki/Podcast_107_Transcript_%28automated...

caseysoftware 11y ago
That's partially due to Google's approach. It appears that they're building a single model for every acoustic model, language, and domain. While that works as a general purpose tool, it's less useful as you go down the rabbit hole. Side question.. how do you host your podcast? I'd love to chat if you get the chance. My email is in my profile.