Command injection in Retrieval-based-Voice-Conversion-WebUI
GitHub Security Lab identified multiple command injection vulnerabilities in Retrieval-based-Voice-Conversion-WebUI version 2.2.231006 where user input is passed unsanitized into shell commands executed on the server. The vulnerabilities affect the preprocess_dataset and extract_f0_feature functions and could permit arbitrary command execution.
What happened
Retrieval-based-Voice-Conversion-WebUI version 2.2.231006 contains command injection vulnerabilities in multiple functions that take user input and concatenate it into shell commands without sanitization.
The preprocess_dataset function in infer-web.py passes user-supplied exp_dir, n_p, trainset_dir and sr variables directly into a command executed on the server. Similarly, the extract_f0_feature function concatenates user input for exp_dir, n_p and f0method into multiple executed commands without proper neutralization of special elements.
Reported impact
These vulnerabilities may lead to arbitrary command execution on the server.
Governance controls
GitHub Security Lab initiated coordinated disclosure on 2025-01-20 by creating an issue requesting a vulnerability contact. On 2025-01-25, a contributor responded that the repository is not fully active. A second disclosure attempt was made on 2025-02-26 requesting activation of the private vulnerability reporting feature. On 2025-04-23, GitHub Security Lab assigned CVEs under its 90-day disclosure policy.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.