Combining Voice, Personalization & Search to discover your next favorite artist

Massimo Quadrana is a scientist in Pandora and researcher at Politecnico di Milano. On October 30th in Microsoft house, he had a speech about what is he doing for a music app, Pandora. I found this one on MeetUp app and the event held by ODSC Milano Data Science.

First of all, this is my face when the speech started, the face of someone who’s Italian is not good enough to follow a topic about Artificial Intelligence and User Experience in Italian. But like before as the slides were in English, I stayed and tried to learn what I was there for.


Massimo Quadrana is a scientist in Pandora and researcher at Politecnico di Milano. On October 30th in Microsoft house, he had a speech about what is he doing for a music app, Pandora. I found this one on MeetUp app and the event held by ODSC Milano Data Science.
first of all, this is my face when the speech started, the face of someone who’s Italian is not good enough to follow a topic about Artificial Intelligence and User Experience in Italian. But like before as the slides were in English, I stayed and tried to learn what I was there for.
Anyway, he was talking about what are the consequences of developing a machine learning system to understand human verbal interaction. The first category of queries he talked about was called: known-item queries. This is when a user asks for a specific song or playlist. but there will be some bugs that the system might face. ASR (Automated Speech Recognition) errors, Partial/Imperfect Queries, and ambiguities. There might be the same song names but with completely different genres and musical tastes, there might be a song’s name which is not easy for the system to recognize the name. Thematic Queries is the next set of queries which is when you ask for example for road trip songs, the system should be able to play a playlist containing songs you might be interested in for the road trip, not a song that is called road trip or songs that are boring for you.
After that, he focused on word features and how sentences can be processed. You can follow what he was talking about on the images below.

The next queries are Broad queries. Like when you simply say play something new.

Anyhow, I really enjoy learning about AI and human-machine interaction. This talk helped to get more familiar with ASR and difficulties of understanding natural human language.

Leave a Reply

Your email address will not be published. Required fields are marked *