News came the other week that Senzari had announced the MusicGraph knowledge engine for music. The Semantic Web Blog had a chance to learn a little bit more about it what’s underway thanks to a chat with Senzari’s COO Demian Bellumio.
MusicGraph used to go by the geekier name of Adaptable Music Parallel Processing Platform, or AMP3 for short, for helping users control their Internet radio. “We wanted to put more knowledge into our graph. The idea was we have really cool and interesting data that is ontologically connected in ways never done before,” says Bellumio. “We wanted to put it out in the world and let the world leverage it, and MusicGraph is a production of that vision.”
Since its announcement earlier this month about launching the consumer version on the Firefox OS platform that lets users make complex queries about music and learn and then listen to results, Senzari has submitted its technology to be offered for the iOS, Android, and Windows Mobile platforms. “You can ask anything you can think of in the music realm. We connect about 1 billion different points to respond to these queries,” he says. Its data covers more than twenty million songs, connected to millions of individual albums and artists across all genres, with extracted information on everything from keys to concept extractions derived from lyrics.
By the end of the year it also expects to fully release its graph API that can help enhance apps with deep musical intelligence. Bellumio says that can be a help both to developers and scientists. The API will have three versions, the first of which can be used to create music-listening services integrating its play-listing algorithm, along the lines of Pandora. “But ours is graph-based so I would say it’s more flexible than some traditional play engines out there,” he notes.
The second API is tuned to driving graph searches combining different nodes whose results can be applied to e-commerce or news sites – for instance, directing users to articles similar to the one they’re reading about other artists who produced songs in the same vein. The third API is focused on providing access to the underlying data in the graph.
“We started to think about opening our technology to the world so other people could do cool stuff with it. With the underlying data, we’re excited to see what the scientific community does in terms of coming up with better algorithms for extracting higher-level features out of features we have,” he says.
Senzari actually collaborates with a variety of societies for music research, and they generally lack access to all this information, he says. “They’re working with very obscure data sets and very limited, but now they can access everything,” he explains.
The opportunity for music research to benefit is big, and the parties that could realize value include big universities like Stanford, Columbia and Berkeley that Bellumio says all have very important music departments very closed tied to their big data programs and computer science schools. “Music was one of the first areas where Big Data [made an impact] because there is so much data there,” he says.
For example, “we extract 200 megabytes per song of acoustic data – for 20 million songs that’s a lot of data. So what we hope to accomplish is in terms of helping scientific people doing analysis or recommendations – that is a very hot area. People also are doing lot of acoustical work to develop different algorithms for real-time transmission or surround sound acoustics. All these things you always need the underlying data.” He also discusses the area of instrument detection and the research around training algorithms to see if they can detect signals to identify instruments, where data about low-level features like pitch waves enters the picture.
“What we see our approach is to put it all together in an ontology that is accessible to graph algorithms. It is sort of like putting Wikipedia on Facebook’s Open Graph, where everything is structured, semantically connected, and you can start extracting knowledge, and its all structured and edges are waited so you can understand how the data relates to each other,” he says.
Senzari will provide both free access for personal and licensed versions for commercial use, but it’s using the utility cost model to make it affordable and easy to incorporate into business models. He estimates Senzari’s approach is going to cost users about ten times less than other alternatives. “We think we can open up to a lot of different innovations in the music space and hopefully the fact that we priced this so low will let smaller companies and others not have to build all this infrastructure and data and instead just build cool products on top of it,” he says.
It’s also keeping an eye out on the algorithm side as users tell it about more signals they’d like to see represented in the graph. Weather is one that it’s probably going to add in, leveraging data from sources that cover that area. Music, he says, is very tied to emotion and so is the weather, so adding that into the mix can lead people to develop some interesting solutions. “How should music change to take the weather into account?” he asks. “What can we learn from that? As more people start using Music Graph data and as we put more data into the Graph, it will be interesting to see how different pieces can work with each other.”