AI-music analysis startup Musiio has had a good 2019: it recently raised $1M in seed funding, is already working with a slew of clients, and is looking to raise money again in 2020.
Musiio’s technology is a mixture of the advanced (AI that can listen to music) and the pragmatic (efficiently labelling catalogue). Right now, its focus is on two B2B products: the automated tagging of tracks, and smart searching of catalogue, all using their proprietary AI. But the future holds possibilities that may be more end-user-friendly, like smarter playlists, sensing “good” songs, and more human sub-genre search.
Music Ally spoke to Musiio co-founder Hazel Savage, who has years of music-tech experience, having worked for Shazam, UMG, Pandora and Bandlab, before meeting her co-founder and CTO Aron Pettersson whilst part of the Entrepreneur First incubator in Singapore.
One of the practical problems Musiio aims to solve is particularly pertinent in a time when companies are aggressively buying and consolidating catalogues – which have often been sorted and given arbitrary metadata in completely different ways.
Savage sees this as Musiio’s opportunity to bring parity in how songs are labelled: “There is a huge explosion in content: Spotify now has 40 thousand songs uploaded each day, catalogue is being acquired so quickly, and there’s more music being made and licensed than ever. The companies [who suddenly have so much more content] don’t really have any way to scale. So we help with that problem.”
Tagging isn’t all that Musiio does but it is a good way to get an overview of what Musiio is capable of. You can try it yourself and receive results in the form of 15 or so varied descriptive tags, such as genre, audio quality, key, vocal type, tempo, and mood.
In our quick test, Musiio was on the money with, ahem, Chris de Burgh’s “Lady in Red” – identifying a “Chill, Easy Listening” song possessing a “Romantic” mood, and a vocal that sounds both “male” and “female”.
It’s a useful demonstration of the company’s tech, and Savage says most of its clients arrive after trying this web interface. Results are currently very industry-centric: those hungry for an automated, touchy-feely update to the glory days of Last.fm-style song-tagging may not have their itch scratched, but anyone with a giant and unwieldy database of music might find a solution to their problem.
Tagging is uncool but essential. Tagging wasn’t even Musiio’s original aim, she explains, it was its “strongest product”, search. But tagging is still important for the stability and historical value of catalogues.
Tagging is also a boring job that, counter-intuitively, has been de-prioritised and executed in a way almost precisely designed to be highly fallible: “If you’ve ever tried to manually tag a series of tracks, it’s one of the most tedious things you can do. It always gets shoved to the bottom of the list or sent to an intern – and it’s massively affected by how that human being is feeling on any given day.”
But it’s one of the more important tasks in a business that seeks increasing revenue from putting the right songs in the right places – in the recorded music industry alone, 16% of revenue is from Placement and Sync.
It’s clear that there’s a long way to go in terms of digging deep into sub-genres – if you search for “French post-rock from 2005” the AI doesn’t know it yet – but their AI can be retrained to tackle this, Savage says. “Maybe you’re a jazz label and have sub-genres like soul-jazz, blues-jazz, and so on. One company requested a hundred custom tags from us for their catalogue – so we retrained our AI and created custom taxonomy, training for genres we don’t have.”
Those who deal with a catalogue and its plethora of merged spreadsheets, cobbled-together documents, and ad-hoc filing systems may be interested in how another client uses Musiio: “One company has half a million tracks in their catalogue, through acquisition. Their large catalogue is very mismatched in its data and also a bit unapproachable, because a lot of data is missing or partially complete. So we tag that entire catalogue – up to a million tracks a day at a cost of a few cents per track.”
Musiio’s search tool is what excites Hazel for the future of the company: she mentions how Audio Network uses Musiio’s search function to quickly pull tracks with specific parameters from its catalogue of 170,000 tracks.
The one tagging category missing at the moment from all music-AI services is also the most important – “Song Quality.” Savage is confident that in the future, Musiio’s AI will be able to sift for this too. “AI can already tell you what the production quality is like on a track – you may only want to pull high-quality content for a playlist, for instance. But of those 40 thousand songs uploaded every day, the one thing we don’t know is which are really good.”
Savage thinks that the AI that can pick out good songs and send them to the right people will be the most successful. “The differentiator will be the quality and the personalisation of the product. I think we’ll see those shifts in the next few years.”
Musiio is also interested in working with streaming services that don’t have the kind of in-house AI technology that Spotify and Tencent have, who can use Musiio’s service off the shelf. Musiio offer full-service options, building dashboards for clients who don’t have developers in-house.
For now, any AI analysis is a prompt for humans, and Savage is aware that tagging is an ongoing challenge that still requires some fundamental decision-making – “when I say ‘Rock’ you might be thinking of AC/DC, but I might be thinking Coldplay.”
Musiio is one of a number of startups, including the Berlin-based Cyanite, that use AI to slice and label music with tags and moods. (Hazel says that Cynaite’s mood perception “is a great feature that people want”.) These AIs sound highly appealing: they replicate a job humans don’t want to do, with consistent results, at a speed humans can never achieve.
Competition may come from the elephant in the room: Spotify (via its The Echo nest acquisition in 2014) has a huge tranche of song-analysis data, right down to data points like ‘danceability’ and ‘speechiness’. It’s not impossible to foresee Spotify offering similar B2B tools based on its own analysis.
Until then, Musiio is happy to provide this service – and while automated tagging isn’t the glamorous or headline-grabbing kind of AI, reliability means money – and a business whose most valuable data is notoriously un-standardised may find Musiio’s tech highly persuasive.
Category: AI music analysis and tagging
Founders: Hazel Savage and Aron Pettersson
Funding so far: $1M seed funding, Jan 2019; aims to raise second round in 2020
Musiio is seeking:
Musiio can offer:
Contact details: firstname.lastname@example.org