I recently had several projects that required transcripts of audio and video recordings, including a great conversation with Kim Dority about finding her unknown unknowns. I identified four services that were regularly mentioned and ran tests to compare their suitability. (Note that this is distinct from dictation software such as Dragon NaturallySpeaking; my focus was solely on tools that reliably transcribe recorded speech and generate a reasonably well-formatted and punctuated document.)
I tested these tools by uploading fifteen minutes of a podcast and comparing both overall quality and specifically the number of errors in a test paragraph. I found that they differed primarily in how many filler words (“um” and “uh”) they removed, whether they could distinguish between speakers, and how well they identified the start and end of sentences and punctuated accordingly. None of the transcripts were perfect but the two 3-star tools offered what I consider good-enough transcription. You will still need to review the transcripts and polish them if you want to publish or distribute them, but this is much less time-consuming than transcribing manually. Note that all these services offer a free trial, so test them with a representative sample of the audio or video you are working with, to see how well each one meets your needs.
AUDIO AUTO-TRANSCRIPTION TOOLS
Temi.com
Cost: $6/audio hour
Time stamp? YES
Multiple speaker ID? YES although not always accurate
# of errors in test paragraph: 6
Comments: Included filler words. Didn’t capitalize most proper names.
Overall grade: ♦ ♦ ♦
Trint.com
Cost: $15/audio hour ($12/hour with subscription plan)
Time stamp? YES
Multiple speaker ID? NO
# of errors in test paragraph: 3
Comments: Eliminated most filler words. Had one UK spelling in test paragraph.
Overall grade: ♦ ♦ ♦
HappyScribe.co
Cost: $6/audio hour
Time stamp? YES
Multiple speaker ID? NO
# of errors in test paragraph: 6
Comments: Lots of run-on sentences. Eliminated most filler words.
Overall grade: ♦ ♦
Spext.co
Cost: $15/audio hour ($10/hour with subscription plan)
Time stamp? Only when viewed on web, not embedded in transcript
Multiple speaker ID? NO
# of errors in test paragraph: 12
Comments: Far more errors than other tools. Eliminated most filler words.
Overall grade: ♦