From the course: Microsoft Azure AI Engineer Associate (AI-102) Cert Prep by Microsoft Press
Unlock this course with a free trial
Join today to access over 25,100 courses taught by industry experts.
Optimize transcription models for accuracy - Azure Tutorial
From the course: Microsoft Azure AI Engineer Associate (AI-102) Cert Prep by Microsoft Press
Optimize transcription models for accuracy
- [Instructor] As I said, there is some capacity for you to train on custom speech. Where you can train on your pre-configured, pre-process labeled audio and text. Phrase hints allows you to pass in additional labels, both single terms, as well as phrases, that can help the service recognize your industry specific jargon. Think of phrase hints as a cheat sheet to help the Microsoft hosted ML model. Language identification, another benefit of these Azure Cognitive Services is that they're so mature, they've been trained on many of the world's languages, multi-channel, diarization. We're getting into the high weeds slinging around the lingo here, aren't we? The terms are really important. Actually, multi-channel refers basically, to multiple speakers. Is it a podcast where you have a host and a guest or is it a whole panel with 10 speakers, you know? And diarization is who said, what, when. Word error rate is WER, and that's the main metric that you're using to evaluate the accuracy of…