Writing a topline is fast. Hearing it sung is not. You hum the idea into your phone, or you sing it yourself badly, or you book a session for a melody that might not survive the week. text2vox fills that gap: give it a melody, the words, and a short clip of the voice you want, and it renders the line sung, plus the MIDI it followed.
It is free, it runs in the browser, and it runs as a plugin. Here is how to use both.
What it needs from you
Three inputs, and the third is the one people get wrong.
A melody. Either a MIDI file or audio whose melody gets transcribed. Exactly one of the two per render, not both.
Lyrics. Typed text, in one of three alignment modes covered below.
A reference voice. A short clip, roughly five to ten seconds, of the voice whose timbre you want. This is a timbre reference, not a performance reference. It does not need to be singing the melody, or singing at all, and a clean solo recording beats a loud one every time. Pull it from an a cappella, a voice memo, or a dry vocal take in your own session. If you know what is said in the clip, you can type it into the reference text field, which gives the engine more to work with.
Only use voices you have the right to use. A reference clip is a likeness.
In the browser
The web editor at the text-to-vocal tool is the same experience as the plugin, with the advantage that there is nothing to install.
It opens on a piano roll. You can author a melody by hand, or drag a .mid file onto the roll and it imports the notes. Drag notes to move them, drag edges to change length, and use the grid selector in the toolbar to control snapping. There is gesture-level undo and redo, and the project persists in your browser, so a refresh does not lose the melody.
Two tabs sit above the roll: notes, and pitch bend. The pitch bend tab lets you draw a curve over the melody for slides and scoops, rather than leaving every note dead straight.
Drop the reference voice into its upload slot, type the lyrics, and hit Render. When the job finishes you get playback in the page plus WAV and MIDI downloads. The returned MIDI is worth keeping even when you like the vocal, because it is the melody the render actually followed.
In your DAW
The plugin ships for macOS as a VST3 and a standalone app. Download it from the downloads page, unzip, then:
- Copy
text2vox.vst3to~/Library/Audio/Plug-Ins/VST3 - Rescan plugins in your DAW
Current builds are not yet notarized, so on first launch right-click the plugin or app and choose Open rather than double-clicking.
The plugin calls Audial's hosted engine for the render, so it needs your Audial User ID and API key, which are on your account page. The plugin itself is free; the synthesis step needs an active subscription.
Ableton Live. Drop text2vox on a MIDI track. Author or paste your melody, set the reference and lyrics, render, then drag the rendered audio onto an audio track to edit, time-stretch or comp it like any other vocal.
FL Studio. The VST3 loads in the plugin picker the same way. Render, then drag the result into the playlist.
Logic Pro. Logic loads Audio Units, and text2vox currently ships as a VST3 and a standalone app, so use the standalone: render there, save the WAV, and drag it into your Logic project. The MIDI the render returns imports alongside it if you want the melody on an instrument track too.
Whatever the host, the practical workflow is the same. Render a line, import the WAV, keep the MIDI, and treat the result as a take you now mix rather than a finished vocal.
The three lyric modes
This is the control that decides whether the output sounds like your line or like soup. The toolbar switch offers:
1:1 per note. Every note carries its own lyric, typed into the note in the roll. One note, one syllable or word, no guessing. Use this when the phrasing is the point: a staccato hook, a rapid-fire line, anything where a syllable landing on the wrong note ruins it. It is the most work and the most predictable.
Auto-align. You type the lyrics as free text and the engine distributes them across the melody. Fastest way to hear an idea. It works well when the number of syllables is close to the number of notes and the phrasing is conventional. It drifts when you have long sustained notes, melismas, or far more syllables than notes.
Explicit tokens. You write out the per-note tokens yourself in the lyrics field, which gives you 1:1 control without clicking into every note in the roll. Convenient when you are pasting from a lyric sheet and already know where the breaks go.
A reliable habit: sketch in auto-align, and the moment a specific phrase matters, switch that section to 1:1 and place the syllables yourself.
The render options worth touching
The options panel exposes the same fields in the browser and in the plugin, with the same defaults, so both render alike from the same inputs.
- Strict pitch, on by default. Forces the vocal to the melody exactly. Turn it off if the result sounds mechanical and you want more natural drift.
- Pitch shift, in semitones. The quickest fix when a melody sits outside the comfortable range of the reference voice. Shift the melody rather than fighting the reference.
- Pitch bend smoothing, default 100 ms, and a switch to disable bends entirely. Shorter smoothing keeps fast ornaments; disabling bends gives a flatter, more quantised delivery.
- Synthesis steps, default 64. More steps means slower and cleaner. Leave it alone while you are iterating on the melody, raise it for the take you intend to keep.
- Guidance strength, default 6.0. How hard the engine is pushed toward the reference and the conditioning.
- Leading silence, default 0.35 s, so the first note is not clipped by the start of the file. Useful to know when you are lining the WAV up to a grid, since that offset is real audio.
- Seed, default 2025. Fix it and you can re-render the same inputs and get the same take. Change it to roll a different performance of the same line.
Tips that save re-renders
Keep reference clips short, dry and solo. Reverb and backing music in a reference clip make the result muddier, not more atmospheric.
Render short. One phrase at a time gives you faster iteration and lets you comp the good passes, which is how human vocals get tracked anyway.
Fix melody problems in MIDI, not in options. If a line sounds wrong, the note lengths are usually the reason, not the synthesis settings.
Keep the returned MIDI. When a render phrases something better than your original sketch, that MIDI is the record of what it did.
Mix it like a vocal. Compression, de-essing, a bit of pitch correction and the right reverb do as much for the result as any setting in the panel.
Where to go next
The browser editor is the fastest way to find out whether this fits your process, and the plugin is the same thing without the round trip through a download folder.
Try it at the text-to-vocal tool, grab the plugin from downloads, and if you want it in a script or over REST instead, the text2vox function is documented on the API reference.