Hi everyone, I got sick of hearing the same ads over and over again in podcasts, so I made this ad blocker for podcasts.<p>How it detects ads?<p>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.<p>Why not use ML to identify ads?<p>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.<p>If anyone has any other ideas on how to tackle this we are all ears.