News for hackers. Stuff that matters.

Ask HN: What to do with 500M call recordings?

We have around 500 million call recordings with average duration of 1 minute in English/hindi and other languages spoken in India. Just wondering, what can we do with this huge dataset? What type of models can we create?
( 9 comments | 2 points | on HN )
Re:Ask HN: What to do with 500M call recordings? by muzani on Tuesday February 04, 2020 @10:08PM (#22243655)
I'd probably run it through some kind of sentiment analysis, try to get it to model as happiness/satisfaction on y-axis, vs time on x-axis.

Then you can map that to conversations and see what words increases or decreases satisfication. Look for sudden changes in the happiness contour.

You can also try to map that to customer surveys at the end of the call - see if you can improve perceived quality, by say, greeting them cheerfully early on, or if different phrases will diffuse anger better. Maybe even see if you can spot weird patterns, like if certain accents trigger anger or contempt.

Re:Ask HN: What to do with 500M call recordings? by sharemywin on Tuesday February 04, 2020 @11:39AM (#22237611)
can they be transcribe via a free speech to text library?

Can you find a existing transcription service that you can up sell to clients?

use hat service to refine you dataaset

then you can build your own model. and cut out the service.

assuming you have it organized by client you could map it to industry. and then build an industry classifier.

run it through sentiment analysis, translation.

Re:Ask HN: What to do with 500M call recordings? by Spooky23 on Tuesday February 04, 2020 @03:01PM (#22240275)
Analyze them for quality purposes to improve whatever it is that you're supposed to be doing.
Re:Ask HN: What to do with 500M call recordings? by mtmail on Tuesday February 04, 2020 @10:39AM (#22236750)
Did the people consent to the recordings?
Re:Ask HN: What to do with 500M call recordings? by urupvog on Tuesday February 04, 2020 @10:42AM (#22236798)
Yes, all of these are business to consumer calls.
Re:Ask HN: What to do with 500M call recordings? by mtmail on Tuesday February 04, 2020 @10:54AM (#22236982)
I'd be interested what the consent looks like. Did users agree to recording "for quality purposes" or specifically for to-be-defined-in-the-future data modeling?
Re:Ask HN: What to do with 500M call recordings? by urupvog on Tuesday February 04, 2020 @11:06AM (#22237143)
Ofcourse, for quality purposes.
Re:Ask HN: What to do with 500M call recordings? by AnimalMuppet on Tuesday February 04, 2020 @12:28PM (#22238310)
Then you're kind of restricted by the consent they gave. They didn't give their consent for you to use them for any purpose that you could come up with; they gave their consent for very specific purposes.
Re:Ask HN: What to do with 500M call recordings? by mtmail on Tuesday February 04, 2020 @12:25PM (#22238273)
Delete all data. They didn't give permission to anything else than quality purposes.