Real-Time Radio Music Recognition

I need a complete system architecture for a platform that listens to live radio streams and returns the name of any song playing within a few seconds. My focus is pure music recognition, so speech detection or general sound-event analytics are out of scope for now. The critical feature is real-time song identification; future add-ons such as metadata scraping or playlist building can be left as optional notes rather than core components.

Here is what I expect from the engagement:

• A high-level and component-level architecture diagram showing how audio is ingested, fingerprinted, matched against a reference database, and the result delivered with sub-second latency.
• Technology recommendations (e.g., Python, C++, TensorFlow/PyTorch models, audio fingerprint libraries like Chromaprint or ACRCloud SDKs, plus cloud services such as AWS Kinesis, Lambda, DynamoDB, or their GCP/Azure equivalents).
• Scaling and fault-tolerance strategy for thousands of concurrent radio channels, including container orchestration (Kubernetes/EKS/GKE) and message queues (Kafka or Pub/Sub).
• Latency, accuracy, and capacity benchmarks I can use to evaluate the design.
• Security and compliance considerations for streaming copyright material.
• A concise written explanation (4–6 pages) that I can hand directly to my engineering team.

Acceptance criteria
1. Diagram is clear enough to guide implementation without additional clarification.
2. All tech choices include a brief justification and at least one fallback option.
3. End-to-end recognition latency target is specified and realistically achievable.
4. Document is delivered in an editable format (draw.io, Lucidchart, or similar) plus PDF.

If anything in the brief seems ambiguous, flag it early so we can keep the scope tight.