The fingerprinting algorithm described in the paper handles noise and quality issues - it’s not just for lab environments. That said there are many parameters that can be tuned from real-world data, additional filters that can be applied, and no doubt Shazam has improved things over the years.

With regards to engineering, I believe the lion’s share of Shazam’s engineering goes to things other than the core algorithm: app development, song ingestion pipelines, enterprise analytics for the record industry, etc.