After the notorious "Fake Drake" released its AI song mimicking Drake and the Weeknd, the AI cover game was changed for good. The anonymous TikToker paved the way to myriads of YouTube channels, Reddit threads, and SoundCloud playlists full of AI tracks with artificial voices made by artists who wanted to repeat the success of the anon creator. Some were more successful than others and even got their tracks with AI vocals on top of the charts. This demand creates supply, so dozens of AI-powered voice cloners and changers popped up since, and LALAL.AI Voice Changer is one of them.
LALAL.AI, which is one of the leading music source separation solutions out there, allegedly didn't want to miss the fun and came up with its own solution that could modify any voice to sound like a voice of one of the sixteen select singers. Lined up alongside a stem-splitting solution, noise cancellation, and de-echo, Voice Changer is LALAL.AI's latest development.
So if you, say, always wanted to know how a chart-topping artist's voice would sound in your mix, you can now do that with literally three clicks. Simply select a voice from the sixteen packs and upload a mixed song with vocals to LALAL.AI. The output track will have the voice of a singer you've chosen.
"Our AI meticulously modifies pitch, tone, and timbre, ensuring a natural and high-quality transformation of your singing voice," shared the LALAL.AI dev team in their press release. "Catering to a global audience, we’re working with a multitude of languages, making it a versatile tool for creators around the world. No matter the language of your track, Voice Changer can handle it."
When testing the Voice Changer, though, we've noticed that if a track has female vocals and you want to change them to a male voice, it won't work properly due to some inconsistency between the input and the desired pitch and tone. The LALAL.AI team says the fix is in the works. The product page provides some recommendations for achieving better results, saying that you should upload a high-quality file with "minimum artifacts" at a 44.1kHz or 48kHz sample rate and a higher bit depth, preferably 24 bits.
We tested the tool, and here's our before vs. after (the track we used is royalty-free, downloaded from Artlist.io):
In addition to its pre-packaged voice options, the tool also expects collaboration on custom voice packs, allowing musicians to tailor their mixes to their unique preferences and creative visions.
The service is so far free to use because it's available in the beta phase. Besides, Voice Changer can be accessed only through the LALAL.AI web solution, despite two other products, Stem Splitter and Voice Cleaner, are available on desktop and mobile. For those looking for real-time voice modifications, LALAL.AI's FAQ page says that "the service doesn’t support real-time voice-changing technology yet."
At the end of last year, LALAL.AI released a source separation model, Orion, that reportedly outperforms all existing solutions for vocal isolation. Other relatively new functionality of the tool, such as dereverberation and noise cancelling, indicate that LALAL.AI might be thinking of becoming a more versatile audio-editing solution than just a source separator.
