A few months ago, I started screwing with AI, I asked ChatGPT to make my weather website look more modern. It did an amazing job at it!
Then Broadcastify too away our embedded scanner players that many of use used on our own sites. This forced me to find a new scanner player for my website, after trying a bunch of different players I went back to ChatGPT and asked f it could make me a player for me. It did so but I still was not 100% happy with it as it did not display meta data, and the audio was really delayed. So I explained to ChatGPT that I wanted a low latency player that could also display meta data. I told it at home I was using ProScan software to send my audio to broadcastify. To which ChatGPT told me how to setup Proscan so we could stream directly from Proscan and have low latency and metadata. I was happy. It turned out to be an amazing player it is so low latency that the audio coming from the player is almost in sync with what I am hearing sitting in front of the scanner.
Then one day I noticed on the scanner app on my phone that the app was offering live transcriptions of my scanner feed. Whatever they were using to transcribe the audio was doing a horrible job. So I logged into ChatGPT to see if I could have it transcribe my scanner feed for me. It had me setup my own local GPT service on my computer using something called Whisper. Well its transcription wasn't much better than the Scanner app on my phone. I was frustrated and let ChatGPT know and it then asked me to send it some recordings of my scanner feed, and it listened and told me the audio quality was garbage.
Next thing it did was give me tips on how to reconfigure my SDS-200 and Proscan software to improve the audio quality. And wow what a difference! We tried transcription on my local machine again and it was better but was still making a lot of mistakes, so ChatGPT asked if we could have them try listening and transcribing the audio, so we set up an API with OpenAI and let it listen and transcribe. We went from a 50% quality of the transcription on the local GPT setup to about a 93% quality with OpenAI listening to my feed. It was much better but not perfect.
It was also expensive for it to translate everything, and I was about to give up and then had and idea, and asked ChatGPT if there was a way we could have it report only the real calls for service from the Newington Police Fire and EMS service And it replied back that it has never done anything like that before but it should be possible. It then went to work creating a local listener on my pc, that records the calls with ffmpeg and then passes along the audio that is longer then a few seconds to the OpenAI api for it to listen and decide if the call should be reported or not.
I run the What's Happening in Newington Facebook group and one of the most common questions we get every day is "What are all the sirens?" or "Why are all the police at..." and so I designed this Scanner Summary service to answer these questions... automatically 24/7...
It worked! But it was making a lot of mistakes. Over a few weeks time I had to tell it how to decide our towns 10 codes into plain language and for which 10 codes to report and which to ignore. II then had to teach it all the street names in town so it would report and sell them correctly. And I had to tell it common business names so those were reported and spelled correctly as well.
I then found a few friends to help me beta test this, and the feedback I got was they didn't like the summaries it was reporting the exact address of a call and also including peoples names in the summaries as well. While I felt this is all public information I also understood their fears and worked with Chat to make the summaries more private while still letting user know what was going on. So a call of "Mary Smith of 143 Main Street is reporting a 10-55 regarding a 10-25 who keeps coming in her yard digging holes." became "Police were called to the 100 block of Main Street for a neighbor dispute over a dog that keeps coming into the callers yard and digging holes."
It was working great, and then I asked if the AI could update calls when there were updates to provide for the call. And we then worked for almost two weeks on this to get it reliable, so it would go something like this...
"The Fire Department were dispatched to the 400 block of Robbins Avenue for the report of a smell of Natural Gas."
And as updates would come in they would be added to the summary. So then with an update it would become "The Fire Department were dispatched to the 400 block of Robbins Avenue for the report of a smell of Natural Gas. Upon arrival the fire department used their meters and detected a slight gas leak. Fire Department requested CNG to the scene."
Once CNG was on scene and had an update The summary would change to "The Fire Department were dispatched to the 400 block of Robbins Avenue for the report of a smell of Natural Gas. Upon arrival the fire department used their meters and detected a slight gas leak. CNG was called to the scene where they foud a small leak at the residents furnace. They shut down the gas service to the home and advised the resident to have the furnace serviced. No further fire department action was necessary and the fire department were cleared from the scene."
When I first launched this to the public this past sunday the summaries were only kept for 3 hours. I did this because I found keeping the summaries for longer slowed the system down, as the summaries were auto reloading every 5 seconds. I asked ChatGPT if here was a way to keep the summaries for 24 hours without bogging my server down. It thought about it and added a button to the page which would show the older summaries which were not reloaded eaver 5 seconds. So now features a 24 hour call summary archive all well.
We then also added the low latency scanner player we made for the homepage and added it to the Summary Service. We then made it into a PWA so that people could bookmark the page to their phones home screen turning it into a pseudo phone app. I then had an issue where the scanner audio feed would stop playing when a users phone went into lock mode. So I explained to ChatGPT wanted to keep the audio playing all the time even if the users device went its sleep or locked mode. A few seconds later it shot out a new file for me to use and now you can use it as your own free personal Newington Police, Fire and EMS scanner.
And for the most part it has been really accurate but there were a few times something was spelled wrong which I caught. So I asked it to give me the ability to edit the calls when I see a mistake. I also asked it that in the same editor that it show me the exact transcript that it heard so I can see the unredacted transcript which has already came in handy as I have a friend who has been having health issues and there was a call to that persons street the other day and I missed the street address over the scanner audio and wanted to make sure my friend was ok so went and and saw the address and was relived to see it was not them.
I also had it add the ability to manually add missed calls. As well as add Community Notes and Town Meeting notifications. It now lets me do this from a special secure interface. I also added the ability to have other users be able to login to it as well and make corrections as needed. Those folks get their own login and all changes and additions are logged so I can see who made those changes.
So far with the first of running this my API costs for this with OpenAI have been around $11 which is very reasonable. Of course I also pay $20 a month for my ChatGPT plus account that writes all the code for me. Each morning it has me upload the transcript logs and it looks for things to improve and tighten up and then gives me some new files to update with its fixes. This thing now is almost flawless. And I am very happy with it!
So now when someone tried asking "Why are all the police at...?" or "What were all the sirens?" I can no give them a URL that will tall then what was going on and in many cases the outcome of those calls as well. All while protecting peoples privacy as well.
I am offering this service to anyone who wants to use it for free, with no ads, no sign ups, no logins, no apps to download at NewingtonWeather.COM - Home or directly at Newington Scanner Summaries
If you in Newington and hear sirens and don't have your scanner on now you can know what they were for, plus have a full Newington scanner feed at your fingertips.
Give it a try and let me know what you think. As far as I know I am the first person to do this anywhere. And it's something I am proud of as it works so well. NewingtonWeather.COM - Home
Then Broadcastify too away our embedded scanner players that many of use used on our own sites. This forced me to find a new scanner player for my website, after trying a bunch of different players I went back to ChatGPT and asked f it could make me a player for me. It did so but I still was not 100% happy with it as it did not display meta data, and the audio was really delayed. So I explained to ChatGPT that I wanted a low latency player that could also display meta data. I told it at home I was using ProScan software to send my audio to broadcastify. To which ChatGPT told me how to setup Proscan so we could stream directly from Proscan and have low latency and metadata. I was happy. It turned out to be an amazing player it is so low latency that the audio coming from the player is almost in sync with what I am hearing sitting in front of the scanner.
Then one day I noticed on the scanner app on my phone that the app was offering live transcriptions of my scanner feed. Whatever they were using to transcribe the audio was doing a horrible job. So I logged into ChatGPT to see if I could have it transcribe my scanner feed for me. It had me setup my own local GPT service on my computer using something called Whisper. Well its transcription wasn't much better than the Scanner app on my phone. I was frustrated and let ChatGPT know and it then asked me to send it some recordings of my scanner feed, and it listened and told me the audio quality was garbage.
Next thing it did was give me tips on how to reconfigure my SDS-200 and Proscan software to improve the audio quality. And wow what a difference! We tried transcription on my local machine again and it was better but was still making a lot of mistakes, so ChatGPT asked if we could have them try listening and transcribing the audio, so we set up an API with OpenAI and let it listen and transcribe. We went from a 50% quality of the transcription on the local GPT setup to about a 93% quality with OpenAI listening to my feed. It was much better but not perfect.
It was also expensive for it to translate everything, and I was about to give up and then had and idea, and asked ChatGPT if there was a way we could have it report only the real calls for service from the Newington Police Fire and EMS service And it replied back that it has never done anything like that before but it should be possible. It then went to work creating a local listener on my pc, that records the calls with ffmpeg and then passes along the audio that is longer then a few seconds to the OpenAI api for it to listen and decide if the call should be reported or not.
I run the What's Happening in Newington Facebook group and one of the most common questions we get every day is "What are all the sirens?" or "Why are all the police at..." and so I designed this Scanner Summary service to answer these questions... automatically 24/7...
It worked! But it was making a lot of mistakes. Over a few weeks time I had to tell it how to decide our towns 10 codes into plain language and for which 10 codes to report and which to ignore. II then had to teach it all the street names in town so it would report and sell them correctly. And I had to tell it common business names so those were reported and spelled correctly as well.
I then found a few friends to help me beta test this, and the feedback I got was they didn't like the summaries it was reporting the exact address of a call and also including peoples names in the summaries as well. While I felt this is all public information I also understood their fears and worked with Chat to make the summaries more private while still letting user know what was going on. So a call of "Mary Smith of 143 Main Street is reporting a 10-55 regarding a 10-25 who keeps coming in her yard digging holes." became "Police were called to the 100 block of Main Street for a neighbor dispute over a dog that keeps coming into the callers yard and digging holes."
It was working great, and then I asked if the AI could update calls when there were updates to provide for the call. And we then worked for almost two weeks on this to get it reliable, so it would go something like this...
"The Fire Department were dispatched to the 400 block of Robbins Avenue for the report of a smell of Natural Gas."
And as updates would come in they would be added to the summary. So then with an update it would become "The Fire Department were dispatched to the 400 block of Robbins Avenue for the report of a smell of Natural Gas. Upon arrival the fire department used their meters and detected a slight gas leak. Fire Department requested CNG to the scene."
Once CNG was on scene and had an update The summary would change to "The Fire Department were dispatched to the 400 block of Robbins Avenue for the report of a smell of Natural Gas. Upon arrival the fire department used their meters and detected a slight gas leak. CNG was called to the scene where they foud a small leak at the residents furnace. They shut down the gas service to the home and advised the resident to have the furnace serviced. No further fire department action was necessary and the fire department were cleared from the scene."
When I first launched this to the public this past sunday the summaries were only kept for 3 hours. I did this because I found keeping the summaries for longer slowed the system down, as the summaries were auto reloading every 5 seconds. I asked ChatGPT if here was a way to keep the summaries for 24 hours without bogging my server down. It thought about it and added a button to the page which would show the older summaries which were not reloaded eaver 5 seconds. So now features a 24 hour call summary archive all well.
We then also added the low latency scanner player we made for the homepage and added it to the Summary Service. We then made it into a PWA so that people could bookmark the page to their phones home screen turning it into a pseudo phone app. I then had an issue where the scanner audio feed would stop playing when a users phone went into lock mode. So I explained to ChatGPT wanted to keep the audio playing all the time even if the users device went its sleep or locked mode. A few seconds later it shot out a new file for me to use and now you can use it as your own free personal Newington Police, Fire and EMS scanner.
And for the most part it has been really accurate but there were a few times something was spelled wrong which I caught. So I asked it to give me the ability to edit the calls when I see a mistake. I also asked it that in the same editor that it show me the exact transcript that it heard so I can see the unredacted transcript which has already came in handy as I have a friend who has been having health issues and there was a call to that persons street the other day and I missed the street address over the scanner audio and wanted to make sure my friend was ok so went and and saw the address and was relived to see it was not them.
I also had it add the ability to manually add missed calls. As well as add Community Notes and Town Meeting notifications. It now lets me do this from a special secure interface. I also added the ability to have other users be able to login to it as well and make corrections as needed. Those folks get their own login and all changes and additions are logged so I can see who made those changes.
So far with the first of running this my API costs for this with OpenAI have been around $11 which is very reasonable. Of course I also pay $20 a month for my ChatGPT plus account that writes all the code for me. Each morning it has me upload the transcript logs and it looks for things to improve and tighten up and then gives me some new files to update with its fixes. This thing now is almost flawless. And I am very happy with it!
So now when someone tried asking "Why are all the police at...?" or "What were all the sirens?" I can no give them a URL that will tall then what was going on and in many cases the outcome of those calls as well. All while protecting peoples privacy as well.
I am offering this service to anyone who wants to use it for free, with no ads, no sign ups, no logins, no apps to download at NewingtonWeather.COM - Home or directly at Newington Scanner Summaries
If you in Newington and hear sirens and don't have your scanner on now you can know what they were for, plus have a full Newington scanner feed at your fingertips.
Give it a try and let me know what you think. As far as I know I am the first person to do this anywhere. And it's something I am proud of as it works so well. NewingtonWeather.COM - Home
