Deterministic Cover Pipeline

Your voice.
Any song.
Mastered.

Upload a track. Pick a voice model. ROADSTAR splits, converts, mixes and masters — deterministically.

Open StudioView Pipeline
RVC Conversion
MelBandRoformer Split
LUFS-Matched Mix
-9 LUFS Master

The Deterministic Pipeline

Every run follows the same path. Same steps. Same output.

01InputFile Parser
→ raw_audio.wav
02SplitMelBandRoformer
→ vocals.wav + inst.wav
03ConvertRVC v2
→ converted_vocals.wav
04Mixffmpeg + pyloudnorm
→ mixed.wav
05Mastermatchering
→ master.wav

Song or Stems

Upload what you have. The pipeline adapts.

Song Mode

Upload one full track. ROADSTAR auto-splits vocals and instrumental, then runs the full pipeline.

Drop a file or click to browse
WAVMP3FLACauto-splitfast

Stem Mode

Upload vocals, instrumental, and backing stems separately for precise, controlled mixing.

vocalsWAV / FLAC
instrumentalWAV / FLAC
backingsWAV / FLAC
vocalsinstrumentalbackingscontrolled

Real-Time Processing

Watch the pipeline run live.

run_a7f3c2
RUNNING· 12.4s
Split
4.2s
Convert
in progress
Mix
queued
Master
queued
LOG TAIL
[split] loaded MelBandRoformer weights
[split] processed 3:42 track in 4.2s
[convert] loading kay_v3 model...
[convert] running inference on vocals.wav
[convert] f0 median pitch correction active
▋
2 / 4
worker·kaggle·T4

Voice Model System

Loaded from GET /voices

Featured Model

kay_v3

RVC v2 model trained on studio-quality recordings. Optimized for pitch-accurate conversion with f0 median correction.

RVCkay_v3activeproduction
Pitch Modef0 median
ArchitectureRVC v2
Sample Rate40k Hz
Signal Fingerprint
kay_v3 · rvc_v2 · 40k
Voice model list loaded from GET /voices — add new models by dropping them into the models/ directory

Master Output

Ready to stream or download.

Run Output
cover_master_run_a7f3c2.wav
COMPLETE
0:423:42
Loudness-9 LUFS
True Peak-1 dBTP
FormatWAV 24-bit
Sample Rate44.1 kHz

Remote GPU Worker

Kaggle T4 accelerator. Start on demand, stop when idle.

kaggle-T4-gpu
RUNNING
GPU
Tesla T4
VRAM
16 GB
Provider
Kaggle
GET/kaggle/status
POST/kaggle/start
POST/kaggle/stop
Demo state:

API Surface

Base URL: https://roadstar.kaywan.de/api

Runs
POST
/runsStart a new cover run
GET
/runs/{id}Get run status
GET
/runs/{id}/logsStream run logs
Outputs
GET
/outputs/{run_id}List output files
GET
/outputs/{run_id}/masterDownload master WAV
Voices
GET
/voicesList available voice models
GET
/voices/{name}Get model metadata
Worker
GET
/kaggle/statusWorker status
POST
/kaggle/startStart GPU worker
POST
/kaggle/stopStop GPU worker

Before & After

Toggle to compare original vs. kay_v3 master.

Open Studio
Blinding Lights
The Weeknd · 3:20
Shape of You
Ed Sheeran · 3:54
Levitating
Dua Lipa · 3:23
drivers license
Olivia Rodrigo · 4:02
Stay
Justin Bieber · 2:37
Peaches
Justin Bieber · 3:18

Start a cover run.

Upload a song, choose kay_v3, and let the pipeline finish the master.

Open StudioRead Docs