README.md

---
title: Nepali Song Voice Swap
emoji: 🎤
colorFrom: indigo
colorTo: gray
sdk: gradio
sdk_version: 5.49.1
app_file: app.py
pinned: false
license: gpl-3.0
short_description: Sing any Nepali song in your own voice, on ZeroGPU
models:
  - Plachta/Seed-VC
  - funasr/campplus
  - lj1995/VoiceConversionWebUI
---

# Nepali Song Voice Swap

A ZeroGPU Gradio front end that takes a Nepali reference song and a clip of
your own voice, and re-sings the song in your voice.

Pipeline:

1. [Demucs](https://github.com/adefossez/demucs) separates the song into
   vocals and instrumental.
2. [Seed-VC](https://github.com/Plachtaa/seed-vc) (F0-conditioned singing
   model) converts the vocal melody and lyrics to your voice's timbre.
3. The converted vocals are remixed over the original instrumental.

## Usage

Upload the full song (up to 4 minutes) and a 5-15 second clean clip of your
own voice. Diffusion steps trade quality for GPU time — 25 is a good
default, 50 for best quality. Pitch shift helps if your voice's natural
range is far from the song's.

## GPU quota

GPU time is charged against the *visitor's* Hugging Face daily quota, not
the Space owner's. Free accounts get roughly 5 minutes/day, which is about
one to two songs depending on length and diffusion steps.

## Licence and use

App code is Apache-2.0. Seed-VC (vendored under `modules/` and
`hf_utils.py`) is GPL-3.0. Demucs is MIT. Do not clone anyone's voice, or
convert a song, without the rights to do so.