Can we use machine learning to expand global access to music education?
Podcast with Sam Walder the author of the Trala App.
Have you ever walked into a grocery store for a gallon of milk and only been allowed to buy cereal? Well, Sam Walder makes the argument that music education has been like that for a good long time — and it’s ripe for technology (his app, Trala) to swing in and change that. Along the way, he and his cofounder have learned a lot about what magnetizes a user base without relying on dopamine tricks to do so. So: if music is the food of love, play on…
—
Brett: All right, everybody. Welcome back to the Clearview podcast. Today I am speaking with Sam Walder from Trala. Trala is a company that delivers online violin lessons through an app. They are soon going to be expanding to all instruments. They use AI to listen to what you play and tell you what you’re doing wrong. Sam, how are you doing today?
Sam: Doing great. Thank you so much for having me on.
Brett: Of course. Tell me a little bit more about Trala. What are you guys doing over there?
Sam: Our mission at Trala is to make world-class music education accessible to everyone on Earth. Historically, the music industry has been very good at servicing suburban children, but they are a small, small segment of the world population. A bunch of other people want to learn music, so those are the people who we are trying to help. We have started with violin, which is the world’s hardest instrument to learn, and we are the industry leader for remote learning the violin.
Everything we do is done through an app. You can get private lessons. You can watch video tutorials featuring diverse violinists through many genres. You have a technology where you can play music, and the app gives you instant feedback through signal processing regarding the things you’re doing right and the things you’re doing wrong. We’ve got about 10,000 active students right now, in 130 different countries, so we are very geographically diverse.
Brett: What brought you to do this? What made this interesting for you?
Sam: I’m a violinist. I started learning when I was a little kid and I’ve been playing violin my whole life. I also have a background in computer engineering, so I studied both in college.
I was lucky enough to study under a professor — my undergrad thesis advisor. His background was in audio signal processing for music, so he was also a musician and an engineer. He taught me a bunch about how to build algorithms that basically take in audio and give you really interesting data from it.
I was able through that to start working on the initial codebase of Trala, which involved listening to you play an instrument and telling you whether or not you are on-pitch or off-pitch — and whether or not you were hitting the right rhythm.
That got me really interested in seeing what we can do with technology to improve the music education industry.
Brett: I’m curious about the AI tech behind it. For example, if you’re processing somebody’s music and giving them feedback, you want to be giving them feedback on things that are specific: like you just said, pitch and rhythm. Are those separate algorithms? Do you have a mixture of algorithms that are each tuned to some particular aspect or is there one algorithm that’s able to produce an output that is directing the user towards what they need to improve?
Sam: That’s a great question. We’re running several algorithms concurrently that are able collectively to give you some really, really in-depth and helpful information on what you’re playing.
Brett: Can you tell me a little bit about what’s under the hood from a [Machine Learning] perspective?
Sam: We are taking in audio data live through the microphone and processing it. Most of this is deterministic, by the way. It is not like stochastic. It is not machine learning. It is deterministic. Then there’s AI stuff that we’re doing, which gives more pedagogical feedback, but that’s not something that is user-facing right now. All of that is internal.
What we do right now is take in that live audio data and do a bunch of fancy math on it. Then we can know things based on what you’re trying to play at that point, and we will say, “wait, you’re playing something else.” That’s where it gets interesting, because even deterministically we can do things: such as saying ‘your finger is in the wrong spot, and that’s almost certainly because your wrist is collapsed and it should be straight.’ We can give pretty detailed feedback.
These are things that we can get pretty sophisticated with. You and I are talking just through audio. I can’t see you right now. If you were to pick up a violin and start playing, I would be able to just through the audio tell you all sorts of things about whether or not you were standing incorrectly, whether or not your bow had enough rosin on it. Obviously, you would need to know what that even means, but there are a lot of little technicalities which all you need is audio for.
Brett: It is amazing that you can do that level of subtlety deterministically. I can imagine you can just connect an algorithm to some digitized sheet music and have a range of expected sounds, but for it to be able to infer from one kind of off-pitchness versus another kind of off-pitchness being related to finger positioning, wrist positioning or something about the violin itself — or the way you are standing — that’s really fascinating to me.
Sam: All of that comes from having a base of knowledge around how to teach the instrument. [Trala] is led by musicians, which is what really makes the difference. We can shortcut all of the training that would have to go into AI and all the terabytes of audio by just having a little bit of knowledge about how to be a good teacher.
Brett: I saw on your website a list of teachers. How does that work with the app? You’ve got the algorithms — are those algorithms being honed by these teachers over a dataset of user recordings?
Sam: Yes. Those teachers teach private video lessons. The algorithms are just one part of Trala, but the robot doesn’t teach you violin. People teach you violin. Violin is the hardest instrument out there, and no matter what you are trying to learn, it is best to do so with a person.
Also, when you’re learning an instrument, it shouldn’t just be about learning technical skills. You’re trying to be impacted as an individual. We’re trying to bring more music into the world, not just bring more technical skills into the world. It doesn’t matter how good your rhythm is if you don’t have a groove and if you don’t know how to play music — and how to bring that to other people.
We have got two things. We have highly-produced video content from some of the top multi-genre violinists in the world. Multi=genre is important here because almost all music ed focuses on classical music from the 18th and 19th centuries. We are modernizing that for the very first time. We have people who are focusing on modern Middle Eastern music, Celtic fiddling, the Blues, Jazz, and Pop and Hip Hop. It’s very diverse, and this is what our students are interested in. That’s one part of it.
The other part of it is our private teachers, with whom you can take lessons, who are trained on the Trala method and who can assign homework to you through the app so that you’re basically getting feedback all the time — not just during the 30 minutes you have with them, but throughout the rest of the week.
Brett: So you do these private lessons with real humans who are helping to deliver the full experience of learning to play music, including the emotional and artistic components. Then they give you homework, and that homework is then the stuff that’s graded by the algorithms so that you can scale these teachers’ time over more people without dropping it to the level of a music app that’s deterministically responding to what you’re playing, but without the soul?…
Sam: Exactly. Nobody has done this before, so this is what we call the ‘Trala school of music.’ This is a full music school in one integrated experience, where you’re able to get the benefits of technology and the benefits of the best teacher you could possibly have — who is also matched to you.
That’s one of the great things about scale. The fact is that most teachers are not right for most students. It is hard to find a match. Let’s say you have a local teacher. What’s the chance that you actually really vibe with them and they vibe with you and they understand your goals? It is actually pretty low.
With scale, with technology, we can say if you are 55 years old and interested in bluegrass and are just trying to be a hobbyist, we have a teacher who understands that and is there to help you with those goals. That’s why we call it the ‘Trala School of Music.’ It is holistic, and it is complete.
Brett: Tell me a little bit more about the journey of starting this.
What was the hardest thing for you going from recognizing this need, having this idea? You mentioned you had a connection to a professor who was early into signal processing. You had yourself both a music and a computer science background. Tell me how you went from zero to one here.
Sam: That’s a great question. My college roommate and I, Vish, started this together. We had been building products together for years. We did a ton of hackathons.
We basically figured out how to win any hackathon. We would go across the country and build products and pitch them and win some money. After a while, we thought, ‘hey, we can probably just do this as a job.’ We always knew we wanted to build something, and it was obviously going to be something to do with music and technology because we saw this huge, huge problem of accessibility in the space. It is the perfect thing that’s primed for technology.
Most people don’t have access to this kind of thing. The reason they don’t have access is because of geographic issues, because of class and race issues, because of age issues. Things where people don’t feel like they are welcome in a beginner music school.
So this is perfect for technology.
That’s why we started building this, and that’s why we started writing the signal processing code. We spent two years just writing signal processing code, and then we put it on the app store pretty early on and started getting organic downloads just like that. It turned out a lot of people wanted what we were building. You know how they say the the ‘if you build it, they will come’ thing is actually BS. That wasn’t really the case for us. We built it, and they did come.
Brett: When you guys launched on the app store, how many competitors were out there? How many other music apps? I am imagining that was something that hit the market pretty quickly.
Sam: There are hundreds of violin apps, but we quickly became the number-one violin app because we were delivering. I talked you through a couple pretty big things that we are doing. All of the signal processing tech, all of the video content, and all of the private lessons — all three of those taken separately could be a pretty successful product. The fact that we are combining all of them and doing all of them at a really high quality is basically unprecedented in the industry. People recognize that quality.
Brett: You launched on the app store. You described that you had these features and then some distribution of people picked your app over other apps based on whatever, and then they started growing it virally. It sounds like that’s kind of what happened. There wasn’t really any marketing beyond just the app store look-and-feel description. Is that right?
Sam: Pretty much. We say that our students are our brand. The students are the ones who we care about, and we focus in on them. They are the ones who are going to bring the word of Trala out to the world. There is a Trala student who has a tattoo of the Trala logo. He got that pretty early on. People are pretty obsessed with the product.
Brett: Yeah, wow, a tattoo is a pretty solid endorsement. How about the economics here, the business model? Do people pay to download the app and then it is free? Do they pay for private lessons with teachers? Is that an in-app purchase? Are you paying money to the app store for those things? How does this work?
Sam: The self-serve version of Trala with no teacher is a subscription model. That gets you all the tech and all the video content. To take private lessons is priced on top of that. Most students pay a recurring subscription for lessons. All of that happens within the app.
Brett: A bullet point I have for this conversation is to ask you about targeting a truly international audience as a startup. I am curious about that. You said you have people in 130 different countries. Is this all in English or have you internationalized, localized into other languages?
Sam: Right now, it is all in English. It is super-international but not localized, which is kind of nuts. It is something we should probably get around to doing. It will be a really good growth lever once we do that. But yes, everybody who is using Trala right now is using the English version. I don’t know. I don’t how they found us, but they did. They really worked to find us.
Brett: What’s holding you back from localizing?
Sam: It is just a focus thing. We are trying to execute on so much right now. We are currently all this video content. We are going to have subtitles and closed captions for all of it. We are building stuff so quickly that we didn’t want that to slow us down. We are still at the point where we want to focus on the core value that we are providing before we go into mega-growth mode — but once we do that, then localization is really going to be staring us in the face.
Brett: You mentioned videos and closed captioning. What are some of the other things that are on your plate for finishing out the core offering?
Sam: Oh gosh, so much. Ask any founder what their product road map looks like. It never ends.
The majority of the things that we want to build are in order to facilitate a student to have a better experience with a teacher. That means flowing data back and forth between the teacher and the student.
The other thing we really want to execute on is the playing experience. We want to make sure that when you’re actually playing your instrument, you’re in a flow state.
Brett: Something you said earlier is that you kind of got your start into this by really just being into hackathons, and going from one hackathon to the next and developing a product idea, developing a product, winning the hackathon, getting some money, going on to the next one — and then recognizing you could maybe just build a product that was successful. When you started to do that — when you started to build Trala — what was different about that that you didn’t feel you were prepared for that was not just the Hackathon environment?
Sam: Everything. We were so overconfident. First of all, what we wanted to do was create a shock to the system for music ed. We wanted them to kind of stare in the mirror and think gosh, aren’t we decades behind.
We focused really, really strongly on the technology from the beginning, and it turns out that building really good tech with a great user interface takes a ton of work. We couldn’t just hack stuff together, but the nice thing is we found it was really fun as well to keep iterating. We had constant energy.
The best part — the thing that really kept us going — was having actual people use our products. When you’re building something for like a hackathon, nobody actually uses it. You’re the only person who uses it. But when you build a product it is ultimately validating, because you get to change people’s experiences and sometimes even their lives through the product building. I think that’s the good and the bad. It takes a lot more work, but you get to make more impact.
Brett: I imagine the iteration process could be a little bit different if you are iterating in the course of a couple of days for a judge versus in connection, in conversation with users.
Sam: It is so much more fun and so much more difficult to build for actual people.
Brett: How do you balance or consolidate all of the different types of user feedback and requests to stay sort of on mission while meeting the most of the needs that are coming in?
Sam: That’s a really great question. I think we are way more user-focused than most companies. We used to send an interview request to every single person who downloaded Trala.
I have personally done over 500 phone calls with random students. There is also a nomenclature thing here. We don’t have ‘users.’ We have students, which implies a sense of service. We are here to serve.
Brett: To that end, products that are trying to get ‘users’ will use dark patterns or some design to get people hooked — their dopamine system. This happens with language-learning apps and music apps trying to give people a little bit of positive reinforcement.
So how do you keep people engaged in Trala? What I am trying to ask is: What do you do if you don’t want people simply to become addicted to the app just to get the amount of use that makes your investors feel good, but instead you’re trying to have people get the most out of it, and have that be the reason why they come back. How are you doing that in the app?
Sam: What we think is ultimately satisfying to people is discovering a voice inside themselves that they are able to share out with the world. That’s what music is. We don’t need to trick people into liking music. We don’t need to have dark patterns to allow them to understand that what they are doing is meaningful and fulfilling.
What we are trying to do is simply service the creation and development of something which is already inside of each student. That’s what makes working at Trala — and working in the education industry, I think — specifically very different from working on other products. Because if you just focus on educational outcomes, then people will surprise you.
I think we trust our students a lot to take what we are learning and bring that into their own personal lives and out into their communities. We love hearing stories of people saying ‘I played for my family. I played in my church. I just played for myself, and nobody listened. That was great. It was the only time that I did something for myself today.’ We love that.
Music is all about action. We like to say that we are “99% praxis.” There are people who are very theory-focused. We are very praxis-focused. Let’s make an impact, and let’s trust the students to do that.
Now, at the same time, it is our responsibility as an educator to tell students what they should be doing. That’s why teachers give homework. The student doesn’t necessarily know what they’re supposed to do. The student needs to trust that we are telling them the thing that is in their best interest to adjust. That’s what a good educator does” say, ‘I need you to practice for 30 minutes a day if you are going to hit your goals.’ The student says, ‘Okay, if that’s how I hit my goals, I’ll do it.’ They’ll trust you.
If you’re telling somebody they need to do something they don’t actually need to do, that trust will eventually break down. You’re going to lose them forever. It doesn’t matter how hooked they are in the moment. If you or I feel like somebody is trying to get something out of us, then we’re going to be turned off to that. Right?
Brett: Yeah, so something you just mentioned about this is something you really enjoy about working in education, but it also sounds like a way that you are specifically doing education because a lot of us had the experience of education where it was somebody giving us a syllabus and telling us what to learn and doing it from the outside in with this I am going to install knowledge in you. You’re doing it from this you already have it in you. We are going to help you develop it and bring it out to the world.
Sam: Yeah, that’s such a good point. I don’t know if you ever had this point in school — honestly, most people don’t — where somebody came to you and asked ‘what do you want to learn? Why are you here?’ Our experience with the education system is it is compulsory. You have to go to school.
Brett: Which of these classes do you want to take to get your credits.
Sam: Yeah, exactly.
It starts to get better if you are lucky enough to go to college. You get to choose your major, for instance, but people switch their majors so often. They’re in compulsory education for all these years, and then they have to go off and make their own decisions about how they want to be educated. They flounder with that.
People have so much trouble deciding what to do, and it’s no wonder. It’s because we didn’t allow people to choose. That’s something that we are trying to change at Trala.
Another issue with the music education space is that a lot of people take classical lessons. They take Suzuki lessons, and never once does the teacher ask them, ‘what do you want to play? What music are you interested in?’ That’s now how it goes. It always goes — okay, you just played this Haydn concerto, and now you are going to play this other concerto by Beethoven. Not once — and it sounds obvious, and it sounds crazy — but not once are people asked what they want to do.
Brett: Maybe they’re asked what they want to play from a certain classical selection, but what they really wanted to do — the reason they’re taking Suzuki — is because they wanted to get into Bluegrass, like you said earlier.
Sam: What if you were to go into a grocery store and they asked you what type of sushi you would like, when you’re actually there for some juice, but you’re not allowed to buy juice. You’re only allowed to buy sushi.
Brett: This has been amazing. I am really glad that you exist and that Trala exists. I mean I am really curious. I picked up a sort of a Native American flute. I ordered one on the Internet a month ago. It seems pretty simple. There isn’t much to it. I haven’t played a musical instrument since I was a teenager, and I am really feeling kind of called to it right now. When you guys expand from violin, maybe I will dive into Trala.
Sam: Maybe we will expand to. What’s the name of the flute?
Brett: I don’t even know. It was made by somebody who had a really hilariously authentic looking website. I was like okay, this person really cares about the flutes they are making, and that’s what matters to me.
Sam: I love that. Once we expand to that instrument, then I will send you a message.
Brett: Sounds great. Thank you, Sam. This has been a great conversation.
Sam: Yeah, it’s been a pleasure. Thank you.
Brett: Take care.
—
Clearview loves founders! We provide contract software engineering services to scalers, connecting them with the tech talent and strategy they need to grow quickly and robustly in the direction of their dreams.
