Executive brief
SoftVC VITS Singing Voice Conversion is an open-source tool used for AI-based voice synthesis. A security flaw in its song-processing server allows an unauthorized person to remotely access and steal private audio or media files stored on the server. Additionally, an attacker can force the server to create or overwrite files in unintended locations, potentially disrupting operations or leading to data loss.
Technical details
A path traversal vulnerability exists in the `/wav2wav` Flask route of the `so-vits-svc` full-song inference server. The `audio_path` parameter in unauthenticated POST requests is passed directly to `librosa.load`, `torchaudio.load`, and `soundfile.write` without validation or sanitization. An attacker can provide arbitrary filesystem paths to exfiltrate decoded audio data via the HTTP response body. Furthermore, the application automatically writes a converted `.wav` file to the same directory as the input path, allowing for arbitrary file writes and potential file clobbering. The vulnerability is exacerbated by the server's default binding to `0.0.0.0` and a wildcard CORS policy. The repository was archived in November 2023, and no official patch is available; users should avoid exposing the inference server to untrusted networks.
Affected products
- svc-develop-team so-vits-svc (SoftVC VITS Singing Voice Conversion) through commit 730930d337d171479eadf305f96cbed4bb393e77
Timeline
- 2023-11-12: other: Project repository archived by owner
- 2026-07-23: advisory: Vulnerability disclosed by VulnCheck