Welcome to MTurk Crowd

Work on Amazon Mechanical Turk, learn from the best, and have fun doing it. Join the crowd today!

Sign Up

voice to text

Discussion in 'MTurk Help & Resources' started by Jeremy, May 19, 2017.

  1. Jeremy

    Jeremy
    Expand Collapse
    New Member

    Joined:
    May 17, 2017
    Messages:
    11
    Likes:
    3
    Has anyone tried using a voice to text program for the transcribing hits? I'm curious to know how well it would work
     
  2. electrolyte

    electrolyte
    Expand Collapse
    L'aînée de la communauté

    • Administrator
    Joined:
    Jan 10, 2016
    Messages:
    12,922
    Likes:
    29,168
    I've tried the built in voice to text option on http://transcribe.wreally.com. It recognizes about 50% of my words but then I have to go back and correct the mistakes and put in all the punctuation. It takes longer than just typing it out.

    I've also tried https://speechnotes.co and found the same thing. That one seems to drop or leave out a lot of words I say so I have to go back to edit it which again takes longer than just typing it to start with.

    Both of those are what I've tried for me verbalizing as I'm listening to audio. I've never tried anything to directly transcribe the speaker's voice from the audio.

    I think @Blue@Blue uses Dragon NaturallySpeaking and may have some insight.
     
    Collapse Signature Expand Signature
  3. Blue

    Blue
    Expand Collapse
    Unlikeable Ass

    • Contributor
    Joined:
    Jan 11, 2016
    Messages:
    3,246
    Likes:
    10,092
    DNS allows you to train the voice recognition to a fairly fine point, much better than the built in windows voice recognition, and this gives it a few miles lead (functionality and reliability wise) on both of the services you listed. But the quality of your microphone strongly dictates the reliability of any of these programs. I use a Blue Yeti mic on my desktop that I picked up a few years to start doing a series of audio lectures. It greatly increased the reliability of voice programs... particularly the VoiceBot program I use for voice activated macros.

    DNS can very likely keep up if you already enunciate decently and train it well so that it has lots of examples of how you speak to build its library... might be less accurate if you don't enunciate well without active effort... a decent mic would certainly help as well.

    Edit: I forgot to add that DNS does punctuation too. But obviously, if the character isn't in the autopunctuation or you find the autopunctuation to be too inaccurate to use, having to say "new line" and "comma" "period" will slow your dictation down some in relation to the audio you are trying to transcribe.
     
    Collapse Signature Expand Signature
    #3 Blue, May 19, 2017
    Last edited: May 19, 2017
  4. Jeremy

    Jeremy
    Expand Collapse
    New Member

    Joined:
    May 17, 2017
    Messages:
    11
    Likes:
    3
    gotcha, before finding mkturk i tried doing transcribing on a few websites and it seemed like an hour long audio would take me several hours to finish because i was having to start and stop the thing over and over again just to keep up. my idea was to run the audio on either my phone or computer and use the other device to transcribe to text, then go back and fix whatever mistakes it might have made. might just skip these hits since it doesnt really seem worth the trouble.