MuseScore Studio 4.6.4 released with Cantai support

Muse Group has released MuseScore Studio 4.6.4, Most updates to MuseScore Studio version numbers that involve changing only the number after the second decimal point simply patch small issues and bugs, and while MuseScore 4.6.4 does include those usual items, it is unlike any of these other updates. It packs a major punch if you work regularly with vocal scores or generally a music tech enthusiast.

The arrival of MuseScore Studio 4.6.4 brings support for Cantai, a new machine-learning-based vocal and lyric synthesizer, within a major music notation software application for the first time. It also ushers in a completely new way to render audio within a music notation software application.

Using Cantai in MuseScore Studio

Once installed, interacting with a score using Cantai works essentially the same as any other score. You will notice some new activity on the screen, but there are plenty of options to control this. As you write or edit your music, MuseScore is constantly letting Cantai know about these changes so it can be rendered again using the new pitch, lyric, dynamic or anything else you changed.

As you would expect, you can change playback to Cantai in the mixer. You can find it in the MuseSounds submenu and select from different choirs and soloists.

By default, MuseScore Studio will show you what sections of the score it is rendering and what percentage has been completed in that work.

Here is an example of a score with SATB and strings to get a sense of the playback. You may encounter an occasional hiccup (like the fermata at the very end of this score), but the realistic vocal rendering that includes lyrics is very well done and quite impressive.

 

In MuseScore Studio > Preferences > Audio & MIDI > Online Sounds you can change settings to control this visualization so it is only shown when not finished upon playback or never at all (you can always see the status at the very bottom of the window too). You can also control whether you want the rendering to happen automatically or if you prefer to manually trigger the rendering.

A very brief history of sung-lyric audio in notation software

For those that remember, in the early days of music notation software, vocal playback was limited to symbolic vowel samples triggered by MIDI, incapable of rendering intelligible lyrics. EWQL Symphonic Choirs and its lyric-construction engine WordBuilder were an early breakthrough, proving that phoneme sequencing could approximate sung text, but only through a separate, externally routed VST pipeline, making the process technically heavy and brittle inside scoring applications, and this approach was available for many years to those willing to follow the steps required.

Cloud-powered voice synthesis in plug-ins like Emvoice One later improved timbral realism and lyric support but still relied on render jobs outside the score editor and struggled at scale: complex passages frequently exceeded practical real-time limits or required export to a DAW for iteration.

MuseScore 4.6.4’s support for Cantai marks the first native, machine-learning lyric synthesis loop embedded directly in a major notation environment. Instead of loading a lyric engine via static MIDI routing, the score continuously pushes notation-aware state updates (pitch, dynamics, lyrics, articulations) to a cloud model designed specifically for vocal and lyric inference. This represents not only a dramatic improvement in realism, but a fundamental shift: the score becomes the authoritative source for lyric-literate vocal audio, rather than a disconnected pre-render process.

Still, Cantai can’t play in real time without the offline (as in non-real time, not disconnected from the web) rendering because of the way lyrics are set in scores. Cantai’s rendering doesn’t happen in real time, because, often, we can’t know how to pronounce the first syllable of a word until we see the rest of the word, and in some kinds of music, syllables could be several seconds apart. It is a pretty clever thing to have Cantai’s output rendered separately and then synced to the music score at playback — something that it will be easy for users to take for granted, just like the 1-second look-ahead delay that NotePerformer does.

Although Cantai differs in important ways from NotePerformer, it does remind us of that product’s first release in 2013, in that it transforms more-than-serviceable playback from an ancillary “hack” into a platform-level expectation in music notation software. Indeed, Cantai’s founder, Richard deCosta, has said in the past that he aims for Cantai to be “the NotePerformer of vocal music”.

A bright future for Cantai

Cantai was able to make its direct support in music notation software possible first with Muse Group, but is currently working to bring this support to Dorico and Sibelius as well.  In discussing Cantai with Richard deCosta, he stressed that this very new product will be improved greatly in the coming months with even more vocal and choir options, and more accurate rendering across the all voices on top of that.

There are many plans to grow Cantai in a variety of ways.  In the more immediate future Cantai will be adding new vocalist options and strives to soon support to two vocalist options for soprano, alto, tenor and baritone and 4 different sized choir settings. There are also plans in the works to support a long list of new languages and new singing styles like broadway, jazz and pop to the choral and operatic styles currently offered.

While vocal rendering is the immediate focus, there are ambitions to take this approach into instrumental audio rendering, an interface allowing for more granular controls, and better automatic switching of vocal styles based on the inputted dynamics. The current iteration offered in MuseScore Studio allows you to select to different vocalist performances like regular, operatic, soft and powerful manually.

Cantai also seeks to apply its technology to other VSTs one day, allowing you to use a different third party vocal library in the same seamless way.

Impact to audio rendering in the future

Normally all audio rendering for playback in music notation software happens locally on your device, but the system in MuseScore Studio connecting to Cantai’s cloud-based rendering is revolutionary.  There are plenty of cloud-based audio rendering and processing services out there, but none that are baked natively into music notation software playback. This technology could allow for direct support for more cloud-based and machine-learning-based applications in the future, or potentially allow access to full versions of VSTs that currently require a large storage footprint locally to be used via cloud rendering perhaps.

While it may be assumed that Cantai’s design as a machine learning based rendering option inherently requires significant processing power, the technology is designed to be quite lean and will eventually be able to run locally using a typical computer’s processing power and without requiring a large storage footprint either. This means that the current setup in MuseScore Studio may not necessarily need to be a permanent one either.

Richard deCosta, Cantai’s founder, pointed out that the rendering system for Cantai is very much “apples and oranges” when compared to how audio is rendered using a traditional VST and pointed to how even Large Language Models (or LLMs) like Ollama can be run completely offline while others like ChatGPT need to be online to use.

Other updates in MuseScore 4.6.4

In addition to Cantai support, MuseScore Studio 4.6.4 “includes critical fixes to improve app stability, security, and performance. It is recommended for all users,” according to Muse Group, who also advises: “It also includes enhanced features to support online sounds, ensuring the best compatibility with Cantai for performing lyrics in playback.” A full list of changes in 4.6.4 is provided at this page.

Availability

A Cantai subscription is now available via MuseHub for $14.99 per month allowing you access to use within MuseScore Studio today. Cantai is now also taking pre-orders for lifetime licenses for use within Dorico and Sibelius here for $150. This is a discounted price that will go up to $299 in the future.

MuseScore Studio 4.6.4 is free and available now for Mac, Windows, and Linux. You can download it directly from MuseScore.org or through Muse Hub. Existing MuseScore Studio users will see a prompt when launching the software to update.

David MacDonald contributed to this article.

---

Learn more

For full coverage of MuseScore Studio 4, please read our comprehensive reviews of MuseScore 4, MuseScore 4.1, MuseScore 4.2, MuseScore Studio 4.3, MuseScore Studio 4.4, MuseScore Studio 4.5, and MuseScore Studio 4.6.

For the latest information about compatibility for Sibelius, Dorico, and MuseScore Studio, as well as links to the latest news and reviews about product releases, please see the Scoring Notes Product Guide.