Thread Rating:
  • 0 Vote(s) - 0 Average
  • 1
  • 2
  • 3
  • 4
  • 5
Experimental support for speech recognition now available
#1
Dear Researchers,

We are *very* excited to tell you about a new cool grammar feature: SPEECH RECOGNITION in audio responses. With speech recognition, FindingFive will automatically transcribe participants' audio recordings for you! For the time being, it only works with English, but we are exploring the possibility of enabling other languages as well.

Highly experimental (Must Read)
Speech recognition on FindingFive is implemented by using a closed-source and yet openly accessible speech-to-text technology provided by Google for everyone to use. Because this technology is provided to everyone for free, Google offers no guarantee whatsoever on its accuracy or reliability. As a result, FindingFive cannot offer any guarantee either. 

In addition, this feature is "rate limited", meaning that Google may suddenly deny further requests to use this feature once the number of requests from a single user reaches a certain, undisclosed threshold on a daily basis. That means:
  • While you are creating and tweaking your study, the speech recognition feature may suddenly stop working because you have reached the limit. When that happens, take a break and it'll resolve itself the next day.
  • While a participant is participating your study, the speech recognition feature may stop working in the middle of the study, because the participant has hit their daily limit. The next participant, assuming they are not using the same computer as the first one, will not be affected. This limit is applied per Chrome browser installation. As a result, as long as you are not transcribing tons of audio recording in a single study, this limit is less of an issue in production studies because of the distributed nature of online testing (each participant is only restricted by their own limit)!
In summary, our advice is that please design your study in a way so that automatically transcribed speech text is the icing on the cake, rather than the cake itself! Big Grin You will always get the audio recording, even when speech recognition stops working.

Google Chrome Only
Also because the underlying speech-to-text technology is offered by Google, only participants with a Chrome browser can use it. FindingFive has implemented a robust check mechanism to prevent participants with Firefox from ever starting a study that uses speech recognition. Such participants would be asked to switch browsers before being allowed to start the study.

How to use
We have updated our Study Grammar documentation with the details, but it's very straightforward. Just add the following bit to either the audio response or the background audio response:


Code:
"speech_to_text": true

That's it! In the downloadable CSV containing participant data, you'll get both the original audio recording (saved separately as an audio file) and a transcribed text in the CSV.

Please feel free to leave your questions in this thread! We look forward to hearing your suggestions and feedback.
Reply


Forum Jump:


Users browsing this thread: 1 Guest(s)