Musiio provide AI-powered analysis, tagging and search tools to some of the world’s biggest music catalogues, counting Sony Music, Hipgnosis, Amanotes, Epidemic Sound and Blanco Y Negro among their customers.
A Rock-loving guitarist turned Co-Founder & CEO, Hazel Savage has spent 15 years in the music industry working for some of the world’s largest music brands – from stacking shelves at HMV, to running teams in businesses at the forefront of music listening and recommendations, Hazel understands the needs of the industry from musician through to large multinational.
You’ve been in the music industry for over 15 years, what makes you so passionate about music, and why did you want to get involved in the music industry?
My parents were pretty rock and roll. They were huge music fans, so I was always surrounded by music growing up. Then, for my 13th birthday, I got a guitar. I still play and have a passion for performing live. So when I was figuring out what I was going to do with my life, it made sense to focus on something that I’d dedicated nearly all my time to.
I ended up doing loads of things tangentially related. I played in a band. I managed bands. I ran club nights. I was handing out flyers for other people’s club nights, running guest lists, and before I knew it, it became a career, although definitely with a tech slant.
Could you share the genesis story behind Musiio?
My first job out of university was stacking shelves in HMV (the UK record store). So, you could say I’ve been aware of the problems with categorising music ever since then. Fast forward a few years (via Shazam, Pandora and Universal), and I was working for a UGC music platform with thousands of tracks uploaded a day. I worked with a playlister who had to manually collate the best music uploads into playlists. He’d listen to hundreds of tracks a day. Some days he had enough suitable content for a playlist. Some days he didn’t. I began to wonder whether there might be a way to automate finding the best tracks for a given scenario. That way, he could use his skills as a music expert for curation, rather than just acting as a filter for bad music.
Musiio was formed when I met my co-founder Aron Pettersson through the start-up incubator Entrepreneur First in Singapore in 2018. Aron is an AI genius. When we were talking about ways we could work together, we realised that we might be able to use Aron’s AI skills to solve the problem of music-based filtering, automatically tagging or searching music with genres, moods, BPM, etc or fingerprint based searches. Aron built a prototype of the algorithm in an afternoon, and we set it working on a free archive of music. We went out to lunch, leaving it to process the data. When we got back, we were amazed by the accuracy of the results. We couldn’t have hoped for a more successful proof of concept. From there, we’ve optimised the algorithm massively. We have a music team that helps teach the AI and conducts QA, and we’ve released products for tagging, audio reference search, playlisting and even song segment selection for platforms such as TikTok.
What are the different types of machine learning algorithms that are used?
We’ve built our own proprietary algorithms, and we consider this our secret sauce! My co-founder Aron has been at the forefront of machine learning for over a decade working across molecular biology, neuroscience, physics and even games development. He leads our AI team. We also take advantage of great available technology such as TensorFlow, Kubernetes and Google Cloud Services for scalability and to deliver our products on a massive scale, at our largest volume we were tagging 5,000,000 tracks per day! We have also spent significant time and effort streamlining our workflows in JIRA; it’s not just about which tools you use but how efficiently you can work with a team of devs and music experts. The combining of the two teams AI and Music is the second part to our secret sauce.
What are some of the challenges behind building a search engine for music?
Speed and accuracy are the big challenges with search. It has to be fast because people are using it in real-time. This is different to tagging because a user will often make several search queries, but tagging only happens once.
There are various things you can do to speed up search. You could just show tracks that share the same tags as your seed track, but you’d sacrifice accuracy. A pure audio-reference search across a catalogue of 200 million tracks, for example, can take a long time, so you are constantly balancing speed and accuracy and looking for solutions. It’s tricky and some of it is hard-won knowledge, but what I can share is that we convert audio files into spectrograms, highly detailed fingerprints of audio files and when we conduct an audio-reference search, the algorithm analyses up to 1,500 data points – far beyond what’s possible with word-tags alone. And it picks up hard-to-describe musical features such as vocal quality, ambience and vibe. We also allow users to define filters, so their searches can be faster and more focused.
Another challenge is how you manage relevance. Most people won’t go past the first page of results, so we’ve spent a lot of time on that.
What problems does Musiio solve for b2b clients?
We serve anyone with a music catalogue. We’ve built the tech to scale, whether you’re a musician who doesn’t have time to tag music and wants to focus on creating, or a streaming service with hundreds of millions of tracks.
We help record labels organise their data for better catalogue navigation, we help sync companies (who put music to video/TV and film) uncover hidden gems and we help streaming services build better playlists. The problem all these companies face is that processing audio manually by listening to every track is labour intensive and hard to do accurately for a sustained period. I tagged 1000 tracks as an experiment. It took two weeks and was not fun at all. Our AI can tag millions of tracks a day with 90-99% accuracy.
With our Musiio Search product, we allow our B2B customers to offer audio reference search as a feature. If a video producer is looking for a music placement, they would start by understanding their client’s expectations of genre, mood, BPM and then search on their chosen site.
Musiio shortcuts this process with our partners who install our search by allowing the same video producer to use a ‘reference track’ to search the entire database within seconds. Our AI will scan the reference track and return the closest audio matches.
Musiio recently launched a NFT Song Slicer product, could you describe what this is?
NFT Song Slicer is a prototype designed to help artists get more value from their music. It uses an AI-driven process to find desirable hooks in a track – up to three per song – and give timecodes so an artist can mint these song sections as NFTs. It can also do this automatically for entire catalogues, making it easier for labels and artists with large back catalogues to quickly create new digital collectable assets.
What are some potential use cases for this type of Song Slicer product?
For catalogue owners or artists with an extensive back catalogue, NFT Song Slicer can select the most valuable sections in millions of songs a day. Record labels, for example, can then turn these song slices into NFTs, and sell them as limited-edition digital merchandise.
With the streaming revolution, it’s become difficult for fans to get a dollar in the pocket of the artists they love. We look at NFT Song Slicer as a way for fans to support their favourite artists, and for fans to own digital collectables. Each slice can also be priced differently by a rights holder. For example, a chorus might cost more than a verse.
And, because NFT Song Slicer identifies the most valuable sections of a track, we see this technology offering value predictions for NFTs and even entire music catalogues.
What’s your vision for the future of Musiio?
I say that Musiio is one-third of a billion-dollar company. To build that company, you need three parts. The first is legal access to large volumes of data, or a “pipeline”. The second part is the tech. This is us, and we’re very good at what we do. The third and final part is a label: a way to monetise what you find, search or discover. Musiio is always working towards this long-term goal.
Do you feel that AI will be able to write and generate music in the near future?
I’m pretty outspoken about not being a huge fan of AI for creativity. It’s a fun academic experiment, and there are systems that do it, but I just don’t see the need for it. Musiio works so well because nobody wants to tag thousands of songs a day. It’s not fun, and you don’t need a person to do it effectively or quickly. But music-making? I’m not so sure. There’s no shortage of people who want to make music.
Even so, I think we’re at least five to 10 years away from AI-generated music sounding good. I heard some AI generated piano music the other day, and it’s hard to tell whether it’s written by AI or just someone who’s not very accomplished. I’m not convinced an AI performance will ever be indistinguishable from an accomplished human player.
And why would you want it to be? So much of what makes music interesting is the lore around an artist, their persona, style and message. It’s not just about the music.
Is there anything else that you would like to share about Musiio?
I’m very excited that Musiio was just awarded the number 4 spot in Fast Company’s 10 Most Innovative Music Companies of 2022. Our team and tech has grown from the seed of an idea to getting international recognition alongside huge industry names such as Hipgnosis and SoundCloud. It’s a tribute to the blood, sweat and tears our team have put into our industry-leading products. We’re so excited to be at the cutting edge of the intersection between music and technology. And knowing there are use cases that we haven’t even thought of yet makes me very excited about the future.
Thank you for the great interview, readers who wish to learn more should visit Musiio.
Credit: Source link
