Differentiate between speaker voices

Answered

Comments

4 comments

  • Official comment
    Avatar
    Connor

    Wow, thank you everyone for expressing an interest in this!

    Our current platform is capable of speaker separation, but sometimes it can get confused by overtalk. This may lead to the assumption that two speakers are in fact one.

    Special thanks to Rob for the additional suggestions and links, I'll pass this feedback on to our development team. In the meantime, keep an eye on the Product Development page to keep updated on the changes we've been making. 

    https://support.trint.com/hc/en-us/community/posts/360000285925-Trint-s-Product-Development

    All the best,


    Connor

  • Avatar
    John Grimsmo

    I agree wholeheartedly! 

  • Avatar
    Jonathan Key

    I'm guessing this would be a pretty big technical challenge. But it would be great to not have different speakers run together on the same line of the transcription.

  • Avatar
    Rob Wilkinson

    In my use case which is oral history, the interviewer is female and most interviewees are male (don't ask!). M/F voice discrimination should be easy to do.

    Perhaps as a quick fix the Trint user could be asked to highlight a section representing each voice so that it can be fingerprinted. EG. Highlight and name each voice once using the current voice tool and the system could infer the rest and render it either by a change of font/style and/or insert a paragraph break.

    If you want a quick but slightly dated overview of this problem visit:

    https://stackoverflow.com/questions/5758485/open-source-code-for-voice-detection-and-discrimination

    or more up to date: http://www-lium.univ-lemans.fr/diarization/doku.php/quick_start

     

     

     

Please sign in to leave a comment.