How it detects ads?
The app detect ads by either a user identifying an ad (by clicking the Podblocker button) or by a large minority of users skipping forward at the same point in a podcast. Once an ad is identified, we have a team that locates the exact starting and ending position of the ad.
Why not use ML to identify ads?
We would love to, but we can't work out how to make it economically feasible. A bag-of-words model we think would work reasonably well to detect an ad. However, at this stage, the cost of speech to text recognition is around $0.02 a minute, so if you are identifying ads in 10,000 podcasts that release once a week with an average duration of 45 minutes you are looking at spending around $468,000 a year just to get speech to text recognition.
If anyone has any other ideas on how to tackle this we are all ears.
We allow users to subscribe to up to 6 podcasts without paying the subscription fee. I think this is a pretty good deal. Unfortunately, unlike a desktop ad blocker - we need a lot of human involvement to block the ads.
