Noise Monitor is a desktop app for Windows that turns any laptop or PC into an autonomous audio monitoring & security station. It continuously analyzes sound, detects volume threshold breaches, classifies sound sources, & instantly sends evidence (video, audio, photos) directly to Telegram.
As a developer, I created Noise Monitor to solve a specific physical problem. I needed a reliable tool that doesn't just record sound 24/7, filling up the hard drive, but reacts to specific triggers and gives me instant insight into the situation at a remote site.
The app doesn't rely on cloud servers for sound analysis. All processing happens locally on your hardware. This is key for privacy & reaction speed. You get a ready-to-use solution that you can download & install from Microsoft Store.
Why I developed this system: the remote server room problem
I have a second home where my server is physically located. Leaving powerful hardware running in an empty room 24/7 is always a risk. Software monitoring (CPU temp, disk load, uptime) only covers part of the issues. It won't tell you if a fire alarm went off, if a fan started making death-rattle noises, or if someone broke a window.
Install a regular IP camera? It records a continuous stream. If something happens, you'll have to watch hours of video. Cameras with motion sensors are useless if the problem is sound-related.
Noise Monitor filled this gap. I placed an old laptop near the server rack. Set a baseline noise level (fan hum). If a sharp decibel spike occurs, the app:
- Captures video from the webcam (incl. 5 secs before event thanks to circular buffer).
- Records a clean audio fragment.
- Identifies sound type with a local neural network.
- Generates a report with peak volume & duration.
- Sends the entire package to me on Telegram.
I instantly get the video in Messenger, watch it, & understand: it's just the cat jumping on the rack & knocking over a box, or the UPS beeping due to power outages.
How sound detection & classification works
Noise Monitor is based on a high-precision software sound level meter. The app captures raw audio stream from the selected mic (built-in or external USB) & calculates sound pressure level in real time.
You set the trigger threshold yourself. E.g., 60 dB. Anything quieter is ignored. Once sound exceeds the threshold, recording logic starts.
The most interesting part is local classification. The app uses a built-in ML model to analyze sound wave patterns. The model is trained to recognize dozens of typical sounds: dog bark, baby cry, breaking glass, sirens, claps, human speech. Processing happens on your PC's CPU. No audio data is sent to third-party APIs for deciphering. This ensures full confidentiality — a critical requirement when using the program at home.
The system can also back up collected evidence to the cloud (e.g., OneDrive or Google Drive, if you set up log folder sync), saving data even if the laptop itself is damaged.
Unusual Use Cases
Though I built the system for a server room, the architecture proved versatile. In practice, Noise Monitor solves a range of home & work tasks.
Improvised Baby Monitor
Specialized baby monitors are expensive & often have range or signal security issues. You can simply leave a laptop with Noise Monitor running in the nursery.
As soon as the baby wakes & starts crying, the mic detects a noise threshold breach. The local model identifies the "Baby Cry" pattern. You get a Telegram notification with the exact noise duration. Even if classification isn't 100% accurate (e.g., baby just fidgets loudly), you'll still get an audio recording & can assess the situation yourself without entering the room & risking waking the baby.
Dealing with noisy neighbors
Proving regular noise violations is hard. Police need facts, not emotions. Noise Monitor acts as an automatic recorder.
You leave the program running overnight. In the morning, you have a structured log: at 02:15, noise of 85 dB was recorded, duration 40 mins, classification — loud music/bass. An audio file is attached to each record. This is solid evidence for management or local police.
Pet Monitoring
Neighbors complain your dog howls all day while you're at work? You can check. The program will record every bark episode, its duration & intensity. You'll know if your pet misses you for 5 mins after you leave, or truly barks for hours.
Telegram Integration: Instant Data Delivery
I'm deeply convinced that the best notifications are those that arrive where you already are. Writing a custom mobile app for pushes made no sense. Telegram provides the ideal infrastructure for media file delivery.
Noise Monitor has native Telegram Bot API support. You create your own bot via BotFather (takes a minute), insert the token into program settings, & specify your Chat ID. No intermediate servers. The program communicates directly with Telegram servers.
Each notification is structured. It contains:
- Timestamp of event start & end.
- Max recorded noise level in decibels.
- Sound classification result (if model could identify it).
- Attached video (mp4) or audio file (wav/mp3).
- Photo (frame at peak volume).
This approach to Telegram notification setup I use in many of my projects because it works quickly and flawlessly.
Technical Implementation & Performance Optimization
Developing desktop software that must run for weeks without rebooting requires strict memory control. If a program records video continuously, it will quickly exhaust RAM or SSD resources.
Noise Monitor implements a circular buffer. The camera & mic work constantly, but data is stored in RAM in small segments (e.g., last 10 seconds). As soon as a noise trigger occurs, this buffer is flushed to disk, & the event itself is appended. This helps to see what happened seconds before the loud sound in the video.
The interface is designed to consume minimal resources. When you minimize the program to the tray, graph rendering is disabled, reducing CPU load to fractions of a percent. The local ML model loads into memory once at startup & activates only when the noise threshold is exceeded, to avoid draining laptop battery analyzing silence.
This is a classic example of how proper architecture makes an app invisible to the system but maximally effective in a crisis. If you're interested in my other solutions in automation & internal tool development, you can explore my portfolio.
Automation of Control & Security
Noise Monitor is not just a sound meter. It's an autonomous security agent. It doesn't tire, get distracted, & records facts with mathematical precision. Whether protecting server equipment, monitoring a child's sleep, or gathering evidence in conflicts — the app gives you full control over the acoustic environment.
Developing such tools requires a deep understanding of software-hardware interaction, OS memory optimization, & third-party API integration. If your business needs custom desktop app creation, automatic monitoring systems, or complex workflow integration via Telegram & local neural networks, I'm ready to design & implement a solution for your tasks.
FAQ
Is internet needed for Noise Monitor to work?
For basic app operation, video/audio recording, & local sound classification, no internet is needed. The program works fully autonomously. Internet is only required for sending Telegram notifications or syncing the log folder with cloud storage.
Where are recordings & videos saved?
All files are saved locally on your computer's hard drive in the directory you specify. No data is sent to third-party servers, except what you choose to forward via your Telegram bot.
Can microphone sensitivity be adjusted?
Yes, you can set a precise trigger threshold in decibels (e.g., ignore all sounds quieter than 50 dB) & specify the minimum noise duration for recording to start, to exclude false positives from accidental short sounds.
How accurate is local sound recognition?
The local ML model handles clear patterns well: crying, barking, siren, glass breaking. But accuracy depends on mic quality & room acoustics. Classification is a supplementary tool — the main trigger is always a physical volume level exceedance.