Get to Know Your Voice from the Inside Out
You know how your singing sounds. But what’s actually happening inside your voice when you hold a long note or reach for a high one? With Inside Voice, you can finally see it. Pick one of your existing Smule performances or record a new clip, and watch a 3D singer perform it, with the jaw, tongue, lips, and throat moving just the way yours did.
Vocal tract videos have been popping up on social lately, and they got us excited. So we’re sharing an early look at our research into the future of vocal coaching.
Yongyi, Smule’s Director of Research, shows you how it works, from recording a quick clip to seeing your results.
How it Works
It takes less than a minute.
- Choose your song. Open Smule’s Inside Voice web app and record a new 5 to 20 second clip, or sign in with your Smule account to pick one of your past performances.
- Watch your voice in 3D. A 3D singer performs your song in sync, showing how your jaw, tongue, lips, and throat moved as you sang.
- Hear it side by side. Use the slider to compare your real voice with a physically synthesized version of it.
- Get your insights. Discover fun facts about how you sing, like how far your tongue traveled in a single phrase or whether your larynx rises on high notes.
How it’s Built
Inside Voice uses a physical model of the human vocal tract, built on research from the Smule Labs team. The technique, called inverse articulatory synthesis, has been studied for decades. We use modern GPUs to make it fast enough to run in seconds.
Here’s the simple version. A virtual singer listens to your performance and tries to sing it back. It compares its attempt to your recording and keeps adjusting until it matches. Then we look at how its jaw, tongue, and throat were shaped. That’s our best estimate of what you were doing as you sang.
Before the analysis, we use standard AI audio tools to remove background noise and estimate pitch, much like the tools found in music production software. Our team also used AI coding tools while building Inside Voice. The 3D singer itself is a physical model built from techniques commonly used in computer graphics, not generative AI, and Inside Voice doesn’t use your voice to train AI. For the full technical story, read the Smule Labs deep dive.
Private by Design
Inside Voice doesn’t save your recording or your results. Once your video is ready, the analysis is deleted, and Inside Voice never uses your voice to train AI. What you share is up to you.
Share Your Inside Voice
Export your result as a vertical video or image, ready for TikTok, Instagram, Facebook, or X. Each one includes QR codes so your friends can try Inside Voice or get the Smule app. Share it and challenge them to try it with their own voice. Here’s an example of a finished video:
Inside Voice is available in English for two weeks only, from November 13th to November 27th. You must be 18 or older to use Inside Voice. We believe technology is a tool to help you understand and improve your voice, not replace singers or human creativity. Inside Voice is an early step in that direction, and your feedback will shape what we build next. Try it now and show us how you really sing, then tell us what you think in our quick feedback form.
FAQ
Is Inside Voice AI?
Inside Voice isn’t generative AI. It’s built on articulatory synthesis, a physical model of the human voice. We use standard AI audio tools only to remove background noise and estimate pitch before the analysis.
Is my voice saved or used to train AI?
No. We don’t keep your audio or your results. Inside Voice doesn’t use your voice to train AI.
Will it change my existing performance?
No. Inside Voice only analyzes it. Your performance stays in your Smule account exactly as it was.
How long is Inside Voice available?
Inside Voice is available in English for two weeks only, from November 13th to November 27th. You must be 18 or older to use Inside Voice.
Are the insights accurate?
They’re estimates, based on how closely a physical model can match your recording. Inside Voice is built on top of a category of technology called analysis-by-synthesis, and also borrows its common drawback: while we can track how accurately a virtual singer is able to match your singing, we cannot guarantee its muscle movements are accurate and match exactly with your true movements. They’re meant to be fun and to help you learn about your voice, and we encourage you to submit feedback to this experience through our feedback form.
