Voice Conversion
Change a Vocal Into a Violin — or Any Other Voice
KROMA Persona extracts a melody line from your track — the vocal, or a detected instrument — converts its timbre with a neural voice model, then mixes it back into the original backing. The melody and phrasing stay; the voice becomes something else.
How it works
- 1
Drop in a track
WAV or MP3. Works best on clear melodic lines — a lead vocal, a solo instrument.
- 2
Analyze to pick the source
Run the sound analysis so KROMA knows what's in the track — then choose which detected sound to convert.
- 3
Pick a persona
Violin, erhu, saxophone, guitar, piano, bass, Cyber Pop Idol, Peking or Cantonese Opera — each is a dedicated voice model.
- 4
Download
The converted part is mixed back over the original backing — you get a finished track, not a dry stem.
Conversion that keeps the melody
Persona is built on retrieval-based voice conversion (RVC): the source's pitch is tracked with RMVPE — an instrument-grade pitch tracker that stays accurate on vibrato and high frequencies where vocal-oriented trackers fail — then the timbre is re-synthesized by the target voice model.
The result: the melody, phrasing, and dynamics of the original performance, in a different voice. A sung line becomes a violin melody; an erhu passage becomes a saxophone.
Instruments count as sources too
Persona isn't limited to vocals. After analysis, any detected melodic instrument can be the source — convert the violin line to erhu, the flute to saxophone, the guitar melody to piano. The backing stays untouched, so the conversion lands inside the original mix.
Cloud models, no GPU required
RVC normally demands a local NVIDIA GPU and a pile of inference settings. KROMA runs the whole pipeline on cloud GPUs — upload, pick, download. No installs, no VRAM, no parameters to tune unless you want them.
Questions
Can I turn a sung vocal into an instrument?
Yes — that's the core use. Pick the vocals as the source and a persona like violin, erhu, or saxophone as the target. The melody is preserved pitch-accurately; the timbre becomes the instrument's.
Can I convert one instrument into another?
Yes — analyze the track first, choose the detected instrument as the source, and convert it to any persona. Woodwind to erhu, violin to saxophone, and so on.
Does the backing track change?
No — only the chosen source is converted. The rest of the mix (including your gain adjustments, if you make any) is preserved and the converted line is mixed back in at the right level.
Why does my conversion sometimes sound odd?
RVC works best on a single clear melodic line. Dense chords, heavily layered vocals, or very fast runs can produce artifacts. For low-register personas like bass, a +12 pitch option keeps the melody audible in a natural range.
Free to try — no install, works in the browser.