Hacker News new | past | comments | ask | show | jobs | submit | typealias's comments login

Last year I built an automated call screener using Google’s cloud speech APIs.

Extremely effective, runs headless on a $5 VPS box by proxying calls through VoIP, but the added latency is a bit of a no-go.


Depends if you’re on iOS or Android. iOS really only supports blacklisting with the exception of Do Not Disturb contact/favorite whitelisting.


I had the same problem a few months ago before I set up an automated virtual assistant to pre-screen calls[1]. It works with my existing phone number by forwarding all inbound calls to a small Linode box which plays an audio prompt and listens for specific keywords. (Whitelisted numbers get patched through immediately.)

Since I made that post, the system has filtered 100% of the spammers and 0% of my actual contacts. I think the non-bot spammers started to get sick of hearing my audio prompt because the number of spam calls that come in has dropped dramatically.

[1] https://andrewchidden.com/defeating-spam-callers-with-speech...


Impressive integration experiment! Is your phone line busy during call forwarding? Would it possible to package all this in mobile app instead? There is some Speech API available and maybe call control/recording too, but I am not sure - not an IOS dev.


Thanks! I think Google Voice has two lines in, so it’d return a busy tone when I’m in an active call with someone (one line for each person to do the call merging).

>Would it possible to package all this in mobile app instead?

Unfortunately the limiting factor would be background processing here; iOS isn’t permissive enough to really replace the server. I think it might be possible by using a background service on Android though.


Are you able to measure which cold calls are filtered, that you would otherwise like to hear?

For example... A call from a garage to say your car is ready to be picked up, or perhaps some kind of emergency service? I imagine their calls would start with similar wording to sales calls. "Hello, can I speak to so and so?"


Good question! I set my prompt to verify that they at least know my first or last name, so it’s pretty good at reducing false positives. For calls that do get filtered, I keep the recorded audio snippets used for keyword extraction (originally used for debugging). If I had more time, I’d implement sending a muted/non-ringing push notification whenever a call gets filtered with an option to play back the recorded audio snippets from that session.


Consider applying for YC's W25 batch! Applications are open till Nov 12.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: