What's the best speech to text software?
April 21, 2016 8:05 AM   Subscribe

I have lectures (around and hour each) that I'd like to transfer from recorded audio to text. I figure that my students are more likely to read than to listen. What is the best way to accomplish this?
posted by Fister Roboto to Technology (3 answers total) 1 user marked this as a favorite
Hire a transcriber in India. Text-to-speech software tends to be not-great for working from recordings.
posted by If only I had a penguin... at 8:07 AM on April 21, 2016 [1 favorite]

This has come up before (8 years ago), and unfortunately the answers haven't really changed: Hire a transcriber or buy the software and equipment to do it yourself.

Dictation software will give you an endless run-on sentence with no punctuation and at least some errors, which is nigh-impossible to read and more trouble to fix than having it properly transcribed in the first place. Dictation software is really an alternative input device, not a way for computers to understand speech beyond the level of individual words.
posted by jedicus at 11:39 AM on April 21, 2016

I use Dragon Naturally Speaking professionally, and it is phenomenally accurate, particularly the most recent version. The speech-to-text features are far, far better and more accurate than they were 8 years ago. In addition, it has become far better at transcribing recorded audio. The most recent version has Natural Punctuation that inserts periods and commas at logical points (based on your speaking style.) You will have to then go back and add the paragraph breaks and other punctuation, but you would want to proof-read the transcription anyhow. In addition, the latest Dragon also can transcribe two separate voices (e.g. an interview).

Here's the process:
1. You train Dragon to understand your voice by recording yourself reading one of their training scripts. The more training you do, the more accurate the output.

2. If there are terms you use in your lectures that are acronyms, proper names, abbreviations, or scientific terms, then quickly train Dragon to recognize them using the Vocabulary editor. This up-front preparation makes the much more accurate.

3. Then you point Dragon to the audio recording you want transcribed and have it output it as text. You should turn on "Natural Punctuation" option to at least capture the commas and periods.

4. After it has been transcribed, you'll go back and edit the text, adding in the paragraph breaks and other punctuation.

The quality of the output is extremely high. Speech recognition has come a long way in the past few years, even for speech-to-text transcription tasks.
posted by apennington at 10:46 AM on April 22, 2016 [1 favorite]

« Older Ok, so my entire music library just disappeared...   |   Should I Eat It? Newer »
This thread is closed to new comments.