About a year ago I reconnected with an old friend from high school by the name of Mike Grabowski. Mike was Solon High School's freakishly smart valedictorian in the Class of 2000. Despite his being at Stanford, we had fallen out of touch for no good reason. When we met up last year, we were both bubbling with enthusiasm about our current projects. I was excited to be at JotSpot, and Mike was even more thrilled that his startup had just raised funding. We met up for lunch, and it was that afternoon that I tried a music search engine that late last week launched under the name of Midomi.
Mike was insistent that he and his friends had developed technology that could recognize music based on the sound itself. I'm a huge music lover, so I was naturally enthralled with the idea. That day I sang The Beatles' "Yesterday" into Mike's computer, which was connected via the internet to a small database of songs. Low and behold, the system was able to determine what I was singing.
It took a year for Mike's team to bring their new music search technolgy to market, but I think they're going for something pretty cool with Midomi. Midomi is most importantly a search engine for discovering music based on what you sing or hum. It's also a social network that could be a lot of fun for singers who love to share their recordings and listen to others'. BusinessWeek already called Midomi "MySpace meets iTunes meets Google".
I played with Midomi this weekend and found it to work pretty well. I sang "Wonderwall" and "Don't Look Back in Anger" and the system returned the song titles at the top of the results. I hummed the tune of "Fields of Gold" since I can't remember the lyrics and the system nailed it. And then I sang "Anna Begins" and Midomi had no clue.
So as of today, Midomi's technology is far from perfect. But the good thing is that it will naturally get better as people use it. The search engine uses recordings as tags, and when a song isn't well tagged (or tagged at all), the system doesn't work. I believe the system only has around 10,000 unique songs tagged, meaning that it's not going to find much of what you sing into it. But as I understand things, this is largely a restriction due to the fact that getting more recordings indexed would be a financial and logistical challenge for a startup. Thus, Midomi's greatest technical challenge is going to be building a comprehensive index of music that has enough tags that it can recognize when guys like Michael Arrington sing into it.
Probably even more importantly, Midomi will face challenges in terms of business model. Google and Yahoo have made billions of dollars through web search, but that's because those search engines have 3 core attributes:
- technology that helps people find what they're looking for
- a way of consistently monetizing a search
- millions of repeat users
When you consider those attributes, you can approximate just how much money any search engine makes. Midomi has an algorithm that might eventually lead to a solid answer for #1. But the company is trying to sell PlaysForSure DRM'd songs and is linking out to Amazon in its search results. Thus, the company's answer to #2 is quite weak. And as for #3, the company just launched and it's not clear how it can acquire the millions of users it needs to make money and improve its core search technology. And it's also questionable whether people need to search for music by singing. I guarantee that "if you build it, they will come" is not going to work for Midomi.
So, Midomi has some very promising technology, but as a business, they've still got a ways to go. But if I were Apple/Microsoft/Google/Yahoo/Viacom/Verizon/Cingular/Sprint, I'd take a close look at what these guys are doing. Existing digital music stores already have the sales and marketing down, and Midomi search would be a great way to differentiate their products and sell more music.
