Suno AI: New "Chirp" App Turns Lyrics Into Songs With Vocals
Suno AI's new Chirp app is the free text-to-song service that everyone's been waiting for. Google and Meta may have rolled out some great instrumental text-to-music apps this year, but neither of those apps supported the option to add lyrics or generate vocals. That missing piece is precisely where Suno AI excels and it's one of the reasons Chirp has exploded in popularity in such a short time.
You may already be familiar with their flagship product, Bark. It's a text-to-speech app that generates realistic, multilingual speech with audio backing tracks. The Bark model was celebrated for its ability to produce realistic nonverbal audio like laughing, sighing and crying. But Chirp took things to a whole new level.
Less than a year ago, a small group of musicians started using text-to-speech converters like Uberduck to craft vocal melodies in the style of celebrity musicians. They would input lyrics and export a low quality audio file to chop up over a beat, using autotune to make the voices sing. It was a lot of manual work.
With Chirp all you need is a free Discord account and a few minutes to spare. You get around 20 seconds of music almost instantly, complete with vocals and instrumental backing tracks.
I reached out to Suno's cofounder, Keenan Freyberg, to learn more about their vision for AI music generation. In this article I'll share that exclusive interview, along with a full rundown on how Chirp works, with techniques for getting the best results and inspiring new music.
Table of contents
What is Chirp by Suno AI?
You know those vending machines with the plastic capsules that take your quarter and give you a mystery item in exchange?
Chirp doesn't take your money, but it will take your lyrics and give you two 30-second songs with a funny little image based on your text. You never know what kind of vocals or musical backing track will come out the other side.
When you do get something you like, Chirp lets you extend that audio beyond thirty seconds, keeping the same vocal and instrumental style while adding new lyrics of your choice. This means you can create an unlimited number of song ideas and piece them together like a collage. It's like having an AI bandmate that never runs out of ideas.
The audio is generated within a Discord community, so you can see what everyone else is typing in and listen to their creations. This might give you ideas for how to structure your own lyrical prompts. If you hear a style of music you like, you can extend it with your own lyrics as well.
An interview with Suno co-founder Keenan Freyberg
I spoke with one of Suno's co-founders, Keenan, to learn more about their origin story and what they have planned for the near future. Here's what he had to say.
Ezra: What is the origin story of Suno?
Keenan: Suno was founded by four friends and fellow audio enthusiasts: Mikey Shulman, Georg Kucsko, Martin Camacho and myself (Keenan Freyberg).
We all worked together at Kensho, where we cut our teeth doing Audio AI powering earnings call transcription for S&P Global. We were struck by how archaic the Audio AI toolchain was — especially compared to the early signs of magic we were seeing from image generation models and LLMs.
We started Suno in 2022, and we've been developing foundation models capable of generating hyper-realistic speech, music and sound effects. In terms of workflows, we aim to accelerate the process of idea to audio for existing creatives and to reduce barriers to entry for aspiring creatives.
Ezra: What does Suno have planned for future versions of Chirp and Bark?
Keenan: We’re focused on improving quality and controllability. We’re training a newer version of the model, which is extremely promising, and actively working on implementing heavily-requested features like genre prompting. We’ll also roll out a series of quality-of-life improvements that will make it easier to use the models over the coming days/weeks — including the ability to add Chirp to other servers and work in your DMs.
The model hasn’t finished training, but we’re already seeing a noticeable improvement in quality. You can prompt with just lyrics or lyrics and a specific genre. You can get pretty specific and see differences between something like [upbeat country] and [death metal] but also between [death metal] and [death metal, dark]. You can even prompt the model without lyrics with stuff like [beat], [pop][intro], etc. to elicit instrumental music with mixed results. Multilingual generation is also looking promising, which is really exciting.
How to sign up for Chirp AI and start making music
To get started with the Suno AI Chirp music app, use this Discord invite link. If you don't already have Discord installed, download it for free here. You can also run the Discord directly within your web browser.
Once you're in the Chirp Discord server, they'll ask to agree to their terms of service and then you can start creating music. On the left navigation menu, you'll see a Generate category with a number of channels labeled chirp-alpha.
There are already tens of thousands of people using the app, so to reduce the clutter, they've created a few clones of the same chirp-alpha channel. Pick any of them to get started. Each one features a long list of user-generated lyrics, music and art that you can freely explore.
At the bottom of the channel is a text area, where you can type the command "/chirp" and summon a special modal. This is where you'll type in your lyrics, or have ChatGPT make lyrics about a topic of your choice. Here's what that process looks like:
Once you've submitted your lyrics, be sure to check the time stamp. Chirp needs a few minutes to render the vocals and instrumental music. When it's finished, you'll have to scroll up to that time code to find it. This workflow is a bit annoying, but I'm sure they'll improve on it over time.
Midjourney took a similar approach with Discord, granting new users access to a free public channel, while paid users get a private bot to create with. The private conversation keeps your prompts and images invisible to other users, while also preserving them in a closed container so that they don't get buried by other user's content and lost over time.
The meaning of each icon on Chirp's control panel
When a new track has been generated in the Chirp-alpha channel, you'll find a set of controls directly below the lyrics. Here's what each of them does:
Recycle symbol: Generate two new songs using the same lyrics. The vocals and background music will change completely.
Heart symbol: Let the channel know that you liked something. A full-width feature of the song will appear to help make it more visible for everyone else. People can add emojis to celebrate it together.
Fast-forward (extend) symbol: Use the same vocal tone and instrumental arrangement to come up with new melodies and chord progressions, based on new lyrics that you provide.
Tweet: Share a Chirp song on Twitter so your audience can check it out
How to use "Continue" to expand on your Chirp songs
Of all these controls, the option to expand on your music is by far the most powerful option. Most AI music generators give you a single piece of content without the option to continue in that style. Suno's developers understood that people might want to hear more music in the same style, or build a complete song from multiple rounds of 30-second generations.
Submit the new lyrics and you'll receive a new vocal melody. I found this feature particularly exciting for Chirp songs that had real musical content. Sometimes it will rap your lyrics over a sparse drum beat, so there's not much in the way of melody or chord progressions. But when it does generate full arrangements, you can get some amazing song ideas.
To download your best ideas, hover over the song and click the download icon located in the top-right corner of the image as shown here:
If you're a musician, you can import the .wav files into your DAW and assemble them into a song. Once you've created that rough sketch, try re-recording the instruments yourself and use it as a source of creative inspiration.
I've personally transcribed a few Chirp songs to guitar and vocals already. The composing style is very free and as a result, I discovered new melodic ideas and chord progressions that I wouldn't normally use. This takes some ear training, but you can also use an audio-to-MIDI app like Samplab to get help figuring out what the chords and notes are.
Listen to examples of AI music from Suno Chirp
Don't have the time to sign into Discord and use the app? These YouTube videos feature a broad spread of AI music from Chirp. You'll get a sense of what it's capable of, but I still recommend using the app yourself. There's something special about watching your own songs come to life, rather than hearing someone else's.
Features we hope to see from Chirp in the future
Chirp is currently in alpha, meaning it's not technically even a beta app yet. There are likely to be many more features and capabilities coming from Suno before the end of the year. Here are some things that we would love to see from the next versions:
Choose a genre of music (coming soon)
Choose a specific vocal model (available on Bark, not Chirp)
Choose duration of musical output
Chatbot for generating music privately
For now, we're lucky to have free access to such a novel program. Go sign up and enjoy yourselves while they're still offering unlimited use!
Suno doesn't seem to have any plans for rolling out text-to-MIDI, so for those purposes you'll still be served well by the AudioCipher VST.
