Case Study

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

Tech giant develops AI that recognizes songs based on users humming, whistling or singing the melody
Global tech company trains virtual assistant to recognize millions of songs without words

One of the world’s biggest tech companies wanted to train their virtual assistant to recognize tunes without being given the name, the artist or a single word of the song. 

Thousands of audio recordings were needed to train the AI, covering renditions of tunes by people of all skill levels and from different population groups. 

After consulting with the company to identify their requirements, RWS curated a diverse team of global workers who created the data by humming, singing and whistling tunes based on a list of 10,000 songs. The data was then categorized into recordings of people humming, singing or whistling, and by basic demographics.

RWS
After consulting with the company to identify their requirements, RWS curated a diverse team of global workers who created the data by humming, singing and whistling tunes based on a list of 10,000 songs.

Challenges

Train AI to identify songs: 

  • With no lyrics, using the melody only 
  • Based on everyday users humming, singing or whistling the tune 
  • Regardless of the user’s musical ability or demographics

Results

  • Time savings and reduced human error achieved with bespoke automations 
  • 160,000+ recordings – over 5× the requested amount – delivered within budget 
  • AI trained to recognize millions of songs 
  • New feature successfully launched and widely used around the world 
  • Ongoing partnership established for other AI needs