ComfyUI Node
Character Speech (GPT-SoVITS)
Reads a script with GPT-SoVITS character models. {happy} switches the emotion, {main} switches back; [name] switches the speaker; [pause:1s] inserts a pause.
Character Speech (GPT-SoVITS)
- audio
- info
◄character▾►
◄text►
◄seed0►
◄speed1.00►
◄languageauto►
◄reference_audio►
◄reference_text►
◄gpt_weights▾►
◄sovits_weights▾►
◄pause_seconds0.30►
◄top_k15►
◄top_p1.00►
◄temperature1.00►
◄repetition_penalty1.35►
◄batch_size8►
◄cross_lingualadjust►
◄volumenormalize►
CategoryAnomalous/TTS
Inputs (17)
| Name | Type | Default | Description |
|---|---|---|---|
| character | COMBO | A character in the gpt_sovits models folder | |
| text | STRING | The script. {happy} switches the emotion, {main} switches back; [name] switches the speaker; [pause:1s] inserts a pause. | |
| seedopt | INT | 00–18446744073709550000 | — |
| speedopt | FLOAT | 1.000.5–2 | Speaking speed |
| languageopt | COMBO | auto | auto: decided for each sentence from its text |
| reference_audioopt | STRING | Main reference clip: a path inside the character's folder, e.g. refs/xxx.wav. Empty: the one in the character's settings, or a 3–10 s clip with a transcript, picked automatically | |
| reference_textopt | STRING | What the main reference clip says. Empty: looked up automatically | |
| gpt_weightsopt | COMBO | auto: the one in the character's settings, else the most trained | |
| sovits_weightsopt | COMBO | auto: the one in the character's settings, else the most trained | |
| pause_secondsopt | FLOAT | 0.300–5 | Pause between sentences |
| top_kopt | INT | 151–100 | — |
| top_popt | FLOAT | 1.000.05–1 | — |
| temperatureopt | FLOAT | 1.000.05–2 | — |
| repetition_penaltyopt | FLOAT | 1.351–2 | — |
| batch_sizeopt | INT | 81–64 | Sentences generated at once. Lower it when VRAM runs out |
| cross_lingualopt | COMBO | adjust | adjust: a sentence in another language than the reference clip (a Japanese character speaking Chinese, say) uses top_k at most 10 and temperature at most 0.8, which is steadier; lower values you set stay |
| volumeopt | COMBO | normalize | normalize: each sentence's voice is brought to about -20 dBFS (peaks under -1 dBFS), so a new seed or sentence does not jump in loudness. off: the loudness the model gives |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| audio | AUDIO | — |
| info | STRING | — |