Computer software helps artists take music to wider audiences

Startup supported by FAPESP develops novel acoustic fingerprinting platform that counts number of times songs are played by radio stations, TV channels and Internet users.

By Elton Alisson  |  Agência FAPESP – Music industry professionals can now use information technology to obtain accurate data on the music market as a basis on which to promote newcomers or sell new songs by well-established artists.

The computer platform, developed by Playax, counts the number of times songs are played by radio stations, TV channels and Internet users. A startup founded by a computer scientist with a degree from the University of São Paulo (USP) and a music producer, Playax has received funding from FAPESP’s Innovative Research in Small Business (PIPE) program.

The detailed reports generated by the platform, identifying radio stations that play a song and their locations, can be used to develop marketing strategies, schedule gigs, and understand and grow an artist’s audience.

“It’s Big Data for music,” said Daniel Cukier, co-owner of the startup and one of the authors of the project.

“The data generated by the platform can be broken down for publicity, marketing and copyright management purposes,” Cukier told Agência FAPESP. “Our analytics include things like most or least popular genres by region.”

Cukier earned a BSc and Master’s degree in computer science from the University of São Paulo’s Mathematics & Statistics Institute (IME-USP) and is currently studying there for a PhD in digital entrepreneurship.

According to Cukier, the software platform was originally developed as a copyright management tool for composers and musicians, but contact with prospective customers suggested to Cukier and the firm’s other partner, producer Juliano de Moraes Polimeno, that the data generated could also be used in other applications, including the identification of trends in the music industry.

“As we talked to people in the industry, we realized the technology we’d developed had far greater potential than just counting the number of times a song is played by radio stations, TV channels and Internet users,” Cukier said.

“Our system does that, too, but there’s much more value in the analytics that come out of its media monitoring data.”

Innovative technology

The core of the system is an audio fingerprinting algorithm similar to that used by music recognition apps like Shazam. An audio (or acoustic) fingerprint is a condensed digital summary generated from an audio signal that can be used to identify an audio sample or locate similar items in an audio database.

To enhance this technology, the team at Playax initially modified an open-source audio fingerprinting algorithm, but with time they decided to develop an entirely new algorithm of their own.

“We implemented our own version,” Cukier said. “Our algorithm performs far better than anything available previously.”

The other key component of the system is a database of songs supplied by artists who want to know where their music is being played.

The algorithm “listens” in real time to songs played over the Internet via audio and video streaming feeds from 5,500 radio stations throughout Brazil, as well as more than 80 TV channels, web radios and platforms such as SoundCloud and Palco MP3, among others. It then compares bits of them with the contents of the database. When the acoustic fingerprint of any item matches a song stored in the database, it registers the identification.

“We’ve succeeded in creating a system that’s both robust and computationally cheap and can identify millions of songs, even if there’s white noise or the volume is very low,” Cukier said.

In 2015, the platform detected more than 30,000 plays by TV channels, 125 billion on the Internet, and 137 million by radio-streaming services, involving 91,609 artists. The duration of all the plays detected totaled over 2.6 billion minutes, which would occupy a human listener for 5,065 years.

“We processed more than 8,000 terabytes of audio in 2015,” Cukier said. “That’s equivalent to a hard disk storing 1 billion songs.”

When Cukier’s team analyzed the data collected by the system, they found pop to be the preferred music genre for Brazilian Internet users, while sertanejo (Brazilian country music) is the genre most frequently played by radio stations in all five regions (South, Southeast, Center-West, Northeast and North).

Different applications

The system currently has 6,000 registered users. A monthly subscription costs between R$30 and R$300 (now about US$9-$91).

According to Cukier, the registered users are mostly artists who want to know the regions in which their songs are most frequently played in order to sell more tickets to concerts in selected locations.

Other applications include checking radio stations’ compliance with commercial agreements to promote artists and finding out whether songs are played more often in a particular region or city after a local gig.

“We present the data to artists via an online interface where they can search for information using a range of filters, such as the cities where their songs have been played most often. We also point to ways in which they can make their work better known,” Cukier said.

“For example, if a song isn’t being played any more by a particular radio station, the user can contact the station to offer a new song.”

The firm has recently developed a new application that recommends publicity or promotional campaigns for registered artists. Machine learning enables the system to select radio stations anywhere in Brazil as the most suited to promoting songs in a given style or genre, for example.

When artists select radio stations in this way, the system sends their songs for a fee in addition to the monthly subscription charge. Later, it generates a report with the results of the campaign, including the number of radio stations that played the songs during the period.

“We plan to develop several other applications to make the platform even more useful,” Cukier said. “New filters will be offered to let artists compare their performance with those of others or receive real-time alerts when a radio station starts to play one of their songs, for example.”

Today, the system identifies songs played by 80% of Brazil’s radio stations, the proportion that offer online-streaming services.

Playax is now in Phase 3 of FAPESP’s PIPE program and plans to expand by marketing its product in other South American countries, as well as in Europe and the United States.

“As far as we know, there are no systems like ours in other countries capable of detecting when songs are played by more than 5,000 radio stations,” Cukier said.