Splice is an innovative music platform that gives creators access to high-quality samples on demand. Their massive audio library is rich with metadata, making it easy for musicians, producers and DJs to find loops and one shots that match their existing project.
But that's not all that they do -- the Splice AI innovation team rolled out a new AI music tool in June 2023 called Create a Stack. It places complimentary samples together in a vertical layout and lets producers create new music loops on the fly.
In this article we'll share an overview of how Splice works, before diving into the Splice AI features, called Similar Sample and Create a Stack respectively. If you already know about the basics of their platform, you can use our table of contents below to skip forward to the coverage of the company's machine learning tools.
Table of Contents
How does Splice work?
The Splice platform has a user-friendly interface where artists can search, preview, and download samples and presets. A simplified desktop app is also available for managing and access your saved files. Both of these interfaces include filters for key, instruments and tempo, making it easy to find what you need quickly.
You don't need a paid account to explore their vast collection of samples, which includes both classic and contemporary sounds. You'll find everything from vintage drum loops to modern synths. The platform collaborates with esteemed sound designers and producers to offer exclusive content, ensuring that users have always have access to inspiring, high-quality material to work with.
How do you browse Splice sound samples?
To start exploring Splice Sounds, simply click "browse" on the website's left navigation menu. The dashboard lets you search for samples by name, or alternatively, you can begin with categories like instrument and genre.
The screenshot above highlights the search and category sections. I typically browse the instrument category because I know the kind of sound I'm looking for. From there, I'll use the filters to select a genre, BPM, and key signature that I want.
Here's an example where I searched for bass instruments and selected the Genre filter to see my options. Each genre has a number next to it, representing how many audio files are available within it. You can further refine the list by selecting more filters. If you stay too broad, there will be an almost inexhaustible list of options to choose from.
How do you use Splice presets?
Alongside the massive sample library, Splice also offers a marketplace for presets. These are not audio or MIDI files. They are a special file type that loads into DAW plugins, like Serum and Massive, that you purchase separately. The preset menu will give you access to new software instruments that you can compose with.
I personally use Serum with Logic Pro. So in Splice, I click the preset tab to browse and preview sounds that match my song. Then I download presets using a few credits and add it to my local file directory so that it's accessible within the Serum plugin. After that, I can open my DAW, assign Serum to a MIDI track, and load the preset.
Here's a quick visual on what that process looks like in a DAW like Logic Pro X:
How do you access the Splice desktop app?
If you prefer to use the desktop app instead of browsing their website, you'll need to have a registered account. Navigate to the top right corner of the page and click on the circle icon. Select install desktop app from the list to get started.
Once you've installed the app, sign into your account and you'll gain access to the same browse feature, along with a list of sounds that you've purchased for your own library. The interface contains all of the most essential parts of the site, without some of the noise that comes from their software recommendations, like the DAW and Plugin partnerships they promote.
That covers all the essentials of using Splice. Now we're ready to move on to the main course: their artificial intelligence features.
What are the two big Splice AI features in 2023?
The two big Splice AI features are Show Similar Sounds and Create mode. They represent two very different purposes within the app. Based on these names, I'm sure you can guess what they do, but let's have a closer look at some of the details.
How do you use the Similar Sounds feature in Splice?
In 2019, the Splice AI innovation team built and delivered a machine learning tool that finds similar sounds to the sample you’re previewing. Most sample libraries include audio metadata that you can filter through. But musical information like BPM and key signature are often not enough to find a truly similar sound.
So instead of using conventional tags like genre, style, and tempo, the Similar Sounds algorithm analyzes the sonic signature of your reference file. Then it looks for files in its database that have similar sonic properties. Those search results are laid out in a right-navigation menu.
Here's how you can access the Similar Sounds feature within your Splice interface. The layout is the same if you're using the desktop application:
I have personally used Similar Sounds extensively and am always impressed by how well it works. The Splice algorithm is rich with musical metadata about the BPM and key signature, which helps when you're first getting started. But if you want something new with the same flavor, that's when you should use their AI tool.
For those of you interested in learning about the technical details behind Similar Sounds, I recommend this presentation by Ale Koretzky. He is the head of AI and Audio Science at Splice. I've timestamped the part of the video where he talks about their classification technique:
Here's a direct quote from Ale, during a key moment in the video where he explains how they arrived at this technology and how it works:
"We started going down the unsupervised route and this led to our for our first user-facing feature which was Similar Sounds. In this case we started training this Auto encoder model that could come up with really high quality embeddings of the audio data in our catalog and dynamically capture page harmonic profile, temporal structure, the semantics at times.
In a nutshell what we did is design a system that given any sound from the Splice catalog, it could retrieve the most similar sounds at scale in milliseconds. Where the context or the concept of similarities relative and it changes based on the input, so something very difficult to approach, using a clear box approach approach or even a supervised learning approach, this ended up working extremely well. It brought a lot of value to our users it increased conversion and this feature is currently available on the website."
His explanation will be a bit technical for some of our readers. I personally find visuals to be helpful for understanding complex machine learning concepts. We previously covered the Spotify AI enhance algorithm and how its audio classification system works in a nutshell. It's different from the Splice technique but it might give you something to chew on.
Spotifuy analyzes each song across six attributes (energy, valence, acousticness, danceability, speechiness and liveness) measured on a scale of 0-1. Overlapping songs can be visualized to determine which tracks are the most closely related:
As I mentioned, the Splice AI and Spotify AI algorithms are different, but hopefully this gives you something to think about. Instead of using music theory concepts like chord structure or melodic contour, the ML model studies and maps audio properties like the ones above.
The CoSo mobile app and how it works
In 2022, Splice released an AI-powered music creation app called CoSo for iOS and Android mobile devices. Short for complementary sounds, CoSo’s AI flips and stacks samples from different styles, keys and tempos to create original music with the push of a button. You can watch a demo of the app below.
My first guess was that Splice might be using their sample metadata to match tracks together. But it turns out that CoSo does in fact use a machine learning algorithm to build its sample stacks.
During my research for this article, I found another interview with Ale Koretzky from 2022, explaining how the CoSo algorithm works. Here's one section that I found particularly interesting:
While it’s possible to build a rule-based system to match up sounds using annotated metadata like key, chords, or BPM, these approaches have many limitations and they don’t scale or generalize well. CoSo works at a much lower level and goes beyond metadata, matching sounds in real-time and always listening, adapting, and shifting its results based on the sonic qualities of the Stack as a whole.
CoSo has no prior knowledge of any of the sounds and relies only on the audio data itself, only leveraging basic metadata about instruments to give users more creative choices across styles and genres. Given the millions of sounds in the Splice catalog and the up-to-eight Layers within a Stack in CoSo, anyone using the app can generate billions of combinations as CoSo reacts to creative decisions in milliseconds.
One year later, near the end of June 2023, Splice re-released the CoSo experience as a browser feature called Create a Stack. This ai-powered music app loads directly on the website, so that users don't have to limit themselves to a mobile device.
How does the Splice AI "Create a Stack" feature work?
Create a Stack is accessible on the Splice website, located under the browse tab in the left navigation menu. Users pick a style of music to generate a stack of samples that loop seamlessly.
The app is currently marked as a beta version, but it works quite well and I didn't encounter any bugs. If you'd like to get a quick overview of the features, check out the company's video announcement below:
I experimented with the Create a Stack tool for close to an hour, to better understand how it performs. As a user I was able to modify the BPM without many time stretching artifacts. The volume for each sample could be adjusted, muted, or played solo. Any individual sample could be swapped out for another using the same instrument. Samples could be added or removed from the stack as needed.
It's easy to save the whole loop or download individual samples, but you'll have to spend some of your Splice credits. There's a new stack button to quickly start over if you get burned out on an idea or already downloaded what you need.
Overall, I found the Create app to be simple and fun to experiment with. The amount of time it takes to create a beat is far less than searching manually for individual samples and layering them within a DAW. The ability to time stretch all of the samples at once is also a benefit.
For non-musicians, I think the user experience would be superior to other AI music generators like Boomy and Soundful. But as a more experienced musician, the app might feel a bit limiting and contrary to your existing workflows.
What are the downsides of using Splice?
The biggest complaints I read about Splice are regarding copyright issues when producers use their samples in a hit song. Splice does offer licensing guidelines that you should read if you're concerned about copyright issues related to sample-based production. The video below explores this topic further.
People have also pointed out the rising monthly cost and their sneaky user retention tactic. I'm a customer myself and have been hooked on it for years, sometimes begrudgingly.
Splice has a frustrating but brilliant subscription model based on sound credits. You spend around 1-3 credits to get a sample and get to keep the sample forever. If you don't use all of your credits in a month, they roll over and accumulate to the next month. But if you cancel, you lose all of your credits.
The fear of losing all the credits has kept me subscribed for months at a time, even when I didn't really want to be. Every year or so I gorge on dozens of sample packs, and the cycle eventually repeats itself.
Along with this tricky customer retention technique, they have gradually raised the cost over time. When I first started using their software, it was only $8/month. The cost has grown to $13/month, which is not enough to make me quit, but it's a nearly 35% price hike.
Fortunately, Splice curates a high quality collection of sounds, so I always end up getting my money's worth in the end. When it's time to get back in the studio, I have a massive bank of credits to draw from.
I also appreciate that Splice has an innovation department rolling out state-of-the-art AI features, with a commitment to serve musicians specifically.
The company issued a public statement on June 26 2023 as part of their go-to-market. In this article they emphasize their commitment to human-first tech. That's going to become increasingly important in the next few years. I look forward to seeing how their tools continue to mature with time.
If you enjoyed this review, be sure to check out our article from last month on how to use AI-powered sample managers. Splice is just one of several options!