VoiceStudio — GPT-SoVITS Engine¶

GPT-SoVITS (RVC-Boss) is one of the most popular open-source voice-cloning systems (57k+ GitHub stars, MIT-licensed). It does zero-shot and few-shot cloning with excellent naturalness in Chinese, English, Japanese, Cantonese, and Korean, and it is very fast (RTF ~0.014 on suitable hardware).

Unlike VoiceStudio's other engines, GPT-SoVITS does not run inside the app. It ships as a standalone API server, and VoiceStudio connects to it over HTTP.

When to pick it¶

• You already run (or want to run) a GPT-SoVITS server, e.g. with few-shot fine-tuned voices.
• You need fast, natural cloning in zh/en/ja/yue/ko.

Setup¶

1. Install and start the GPT-SoVITS API server (upstream project):

cd GPT-SoVITS
python api_v2.py -a 127.0.0.1 -p 9880 -c GPT_SoVITS/configs/tts_infer.yaml

VoiceStudio talks to the api_v2 protocol (POST /tts). The older v1 api.py server is not compatible. To use fine-tuned voices, edit the custom section of tts_infer.yaml so it points at your own GPT_weights* and SoVITS_weights* files before starting the server.

1. Select the engine via Model Catalogue (TTS tab → Use) or OMNIVOICE_TTS_BACKEND=gpt-sovits.

VoiceStudio marks the engine available only when the server responds (2-second reachability probe).

Configuration¶

Remote servers: by default VoiceStudio only talks to loopback addresses — part of the local-first guarantee. To point at a server on another machine (e.g. a GPU box on your LAN), add its network to OMNIVOICE_TRUSTED_NETWORKS; otherwise the connection is refused as an untrusted endpoint.

Prefer https:// (or a private tunnel such as Tailscale/WireGuard) for any non-loopback server: with plain http:// the text you synthesize and the audio that comes back cross the network unencrypted. VoiceStudio does not disable certificate verification, so a TLS endpoint needs a certificate the system trusts.

Behaviour notes¶

• Output is 32 kHz mono (server output is resampled if needed).
• Requests go out unsplit (cut0); VoiceStudio chunks long text itself. Letting the server split on punctuation produces short fragments that can end early.
• Every request needs a reference clip. A voice profile supplies one; plain TTS uses the two default-clip variables above and fails with an error naming them when they are missing.
• The reference path is opened by the GPT-SoVITS server process, not by VoiceStudio, so a remote server needs the clip on its own disk. A clip that ends in silence, or a transcript that does not match it, gives near-silent output.
• Speed control is forwarded as the server's speed_factor.
• The GPU is whatever the GPT-SoVITS server itself uses (CUDA preferred); VoiceStudio's side is just an HTTP client.

Known limits¶

• Five languages only; for broader coverage use OmniVoice (languages.md).
• No voice design; server availability is your responsibility — if the server stops, generations fail with a "server not reachable" error.

Troubleshooting¶

• "GPT-SoVITS server not reachable": start the server or correct OMNIVOICE_GPTSOVITS_URL. The availability check is a validator-only GET /tts?text=&text_lang=en&prompt_lang=en. Answers of 400, 200 or 405 count as reachable. A 404 means there is no /tts route, which points to a v1 server or a proxy mismatch. Any other status is reported with its number and marks the engine unavailable. Redirects are refused rather than followed and show up as unreachable, so use the server's direct origin.
• "endpoint is outside loopback or OMNIVOICE_TRUSTED_NETWORKS": see Configuration above.
• Other issues: install/troubleshooting.md.

See also: benchmarks.md, expressive-speech.md.

Reference transcripts from saved profiles and uploaded clips use api_v2's auto language detection, independently of the output language. An English reference can therefore produce Japanese speech without being interpreted as Japanese. The configured default clip also uses auto unless OMNIVOICE_GPTSOVITS_REF_LANG specifies its language; that setting never applies to a different, explicitly supplied clip.