CVE-2026-65701 - SoftVC VITS Singing Voice Conversion Path Traversal Vulnerability
A path traversal vulnerability exists in the full-song inference server of SoftVC VITS Singing Voice Conversion, affecting versions through commit 730930d, allowing unauthenticated remote attackers to read and exfiltrate arbitrary files by manipulating the 'audio_path' field in an unauthenticated POST request to the '/wav2wav' route.
A critical path traversal vulnerability, tracked as CVE-2026-65701, has been identified in the full-song inference server of SoftVC VITS Singing Voice Conversion, specifically affecting versions up to and including commit 730930d. This flaw permits unauthenticated remote attackers to exploit the server by manipulating the audio_path field within an HTTP POST request directed at the /wav2wav route. By supplying attacker-controlled filesystem paths, threat actors can force the application to read and exfiltrate arbitrary files, returning their contents via the HTTP response body. Furthermore, the vulnerability enables the writing of attacker-specified .wav files to arbitrary locations on the server's filesystem, leveraging functions like librosa.load, torchaudio.load, and soundfile.write. This vulnerability presents a significant risk for data exfiltration and potential arbitrary file write leading to further compromise.
Attack Chain
- An unauthenticated remote attacker identifies a SoftVC VITS Singing Voice Conversion instance exposing the vulnerable full-song inference server.
- The attacker crafts a malicious HTTP POST request targeting the
/wav2wavroute of the application. - The request includes an
audio_pathparameter containing path traversal sequences (e.g.,../../../../etc/passwd) to specify an arbitrary file on the server's filesystem. - The vulnerable server receives the request and passes the attacker-controlled
audio_pathvalue directly to file loading functions such aslibrosa.loadortorchaudio.load. - The application reads the content of the arbitrary file (e.g.,
/etc/passwd) from the server's filesystem. - The content of the accessed file is then decoded and included in the HTTP response body, effectively exfiltrating the data to the attacker.
- Alternatively, the attacker can specify a writable arbitrary path in the
audio_pathparameter, causingsoundfile.writeto create a.wavfile at the specified location with attacker-controlled content.
Impact
Successful exploitation of CVE-2026-65701 can lead to severe consequences for affected organizations. Attackers can read sensitive configuration files, user data, source code, or other proprietary information directly from the server's filesystem, leading to unauthorized disclosure and data exfiltration. The ability to write arbitrary .wav files to any location on the filesystem can also be abused for denial-of-service, planting malicious executables, or achieving persistence by writing web shells in publicly accessible directories. The high CVSS score of 9.1 reflects the critical nature of this vulnerability, indicating potential for complete compromise of confidentiality, integrity, and availability.
Recommendation
- Patch CVE-2026-65701: Immediately update SoftVC VITS Singing Voice Conversion to a version beyond commit 730930d to remediate CVE-2026-65701.
- Deploy the Sigma rule to your SIEM to detect attempts at exploiting this vulnerability.
- Enable web server access logging for the
webservercategory to ensure HTTP POST requests and their parameters to/wav2wavare recorded for detection and forensics.
Detection coverage 1
Detects CVE-2026-65701 Exploitation - SoftVC VITS Path Traversal
criticalDetects exploitation attempts for CVE-2026-65701 via HTTP POST requests to /wav2wav with path traversal sequences in the 'audio_path' parameter, indicating arbitrary file read or write attempts.
Detection queries are available on the platform. Get full rules →