I’m trying a new experiment this evening with Dragon Naturally Speaking and the online Social Networking service Utterz.  I’m trying to take the .MP3 file generated by Utterz when I phoned into their service and left a message from my cell phone. 

The message could likely be considered to be of the quality of a digital voice mail message.

1 I then downloaded the Utterz mp3 file to my computer and ran it through my Dragon Naturally Speaking profile to transcribe the file into text.  Here are a couple things that I learned:

  1. It is possible to get the file and run it through transcription
  2. The transcription was less than successful.  Accuracy was low and the likely culprit is the quality of the audio (recorded via cell phone and possibly at less than optimal bit rate for dragon naturally speaking  The bit rate was actually 64kbs which should have been fine if anything maybe a little high.)
    1. Utterz probably uses a bit rate slightly on the low side to serve thousands of audio messages
    2. Dragon Naturally Speaking works best with a higher MP3 bit rate (32 – 48 kbs or better)
  3. When I left the message, I did not leave a several second pause at the beginning before I started to speak which can sometimes make a transcription from a file or device skip the first few words in Dragon Naturally Speaking 9.

Follow Ups

I’m going to next try to create an audio dictation on my pocket recorder.  Then upload that audio to Utterz.  I will then run the original file through transcription and separately download the mp3 file from the utterz site separately and run it through as well.

I’m going to do this just to see if the audio quality is adjusted after upload.  I doubt that it is, but I just want to check.

Tip

If you happen to create an audio file that you want transcribed and you forget to leave a few seconds of space at the beginning. You can use the free program audacity to insert a space of several seconds into your audio file and then generate a new mp3 file.

If a file transcribes poorly, you might try converting the file to a different KBS rate.  I mentioned that I seem to get the best results at about 48kbs.  I tried this tip with my own file and it did improve the accuracy (although not enough as I was still getting word substitution errors that were nonsensical like child insurance for ‘file in this instance’).

Comments

Responses...