Amazon Transcribe

With Nuance dropping the ball with Dragon Dictate for OSX, I decided to finally figure out how to try Amazon’s Transcribe service. I already have an amazon aws account so this wasn’t as difficult task as I thought. I created a mono audio file on my Mac using Audacity. I read about two minutes worth of an insurance industry textbook. About 200 words. I then created a S3 bucket (amazon’s online storage), and uploaded the audio file. Amazon provides a web page to create a transcription job. It took about 2 minutes for the job to run and a transcription was created. You get multiple outputs. The first being the transcription in paragraph form. The second being something called a json file where it gives the confidence of each word and other information.

You also have the ability to upload custom vocabulary to help with the transcription. There is a free period of 12 months with a hour of transcription each month. If you go over or longer the cost is $1.44/hour. I think they do partial billing in one second increments, minimum of 15 seconds.

The results: about 90-95% accuracy. I get about the same with using dictation in Drafts on iOS or using the built in dictation on OSX. At least for me, at least initially not worth the additional technical steps of using Amazon. I may try again in the future with a custom vocabulary list. Of course, if you are very tech savvy you can use python or another language to automate the uploading of the audio file and requesting a transcription.

1 Like

Thank you for providing this information. Are the hoops that you have to jump through worth the price?

I’m looking for a way to caption video for my side-gig. At my main gig we use 3PlayMedia and it works wonderfully, but it’s not a cheap option ($3/min). I can use Youtube to create the captions, but Youtube doesn’t work well without a transcript…

I guess my question really is, do you think the work required to use Transcribe worth the ~$1.50/minute savings?