Case Study: Global Tech Company achieves recognition of millions of songs from humming, singing and whistling with RWS

Global tech company trains virtual assistant to recognize millions of songs without words

A global tech company wanted its virtual assistant to recognize songs from melody alone—users humming, whistling or singing—without lyrics, artist or title. The challenge required thousands of diverse recordings that reflected a wide range of musical ability and demographic groups to train a robust recognition model.

RWS Data Services recruited a global, diverse pool of contributors to record 10,000 songs, then annotated and labeled the data by type (humming, singing, whistling) and basic demographics. Delivering 160,000+ recordings—over five times the requested amount—within budget, and using bespoke automations to save time and reduce errors, RWS helped train an AI that recognizes millions of songs; the new feature launched worldwide and RWS secured an ongoing AI partnership.