A system that processes 8 GB of audio per second – behind the scenes of a project for the music industry
MicroBe - Music Recognition
Project goal_
The goal of the project for MicroBe was to develop a dedicated system from scratch that would meet the challenge of identifying music played by selected radio and television stations, operating continuously around the clock (24/7).
A key task was to design a web portal that would display detailed statistics and charts related to the recognized tracks. The project aimed to meet the needs of the Audio and Video Producers Association (AVPA) and its members, enabling them to analyze the broadcast music tracks.
Additionally, the system was required to generate periodic AirPlay reports and utilize an enhanced Echoprint algorithm for audio pattern recognition, with track fingerprint storage in the LMDB database.
Solutions Delivered_
theGoodPeople team developed a comprehensive backend system that monitors selected radio and television stations, along with a web portal that presents various statistics and analyses regarding the recognized tracks.
The system is based on SQL for storing recognitions and metadata of tracks, offering real-time updates and enabling quick access to data on broadcast music.
Furthermore, tools for operators and monitoring were implemented to support daily operations, including automatic insertion based on stream metadata and dedicated equipment for monitoring TV channels.
Project Products_
- Complete backend system for recognizing music tracks.
- Web portal for the Audio and Video Producers Association (AVPA) and record labels, displaying statistics, charts, and AirPlay reports.
Technologies used_
Web portal: Angular.js, TypeScript, ASP.NET
Backend: SQL Server, LMDB, C#, .NET 8, ASP.NET Core 8, RabbitMQ, FFmpeg
Monitoring: OpenTelemetry + Grafana/Loki/Jaeger for infrastructure monitoring. Custom web portal for real-time monitoring of system operations.
Benefits of the solution_
- Modular design
- Service Oriented Architecture
- High-scalability, fully redundant architecture
- 650 000+ recognitions / month
- 300 000+ songs in database
- 4x high-end servers 20-core (40-core hyperthreading),
- 128 GB of RAM, working non-stop with ~50-80% load
- 24/7 support
- Stream metadata analysis
- Dedicated hardware to listen TV channels
- 99 radio and TV stations monitored 24/7
- Reports generated every week/month/year
- Ultra-fast LMDB database (800k random reads per second, processing of average 8 GB of data per second)
- Custom system monitoring and automatic alerts
Project team_
- 2x frontend developer,
- 2x backend developer,
- 1 data developer,
- 1 QA,
- 0,5 PM
