Nodes/ComfyUI-Raon-OpenTTS/Raon OpenTTS Generate (Voice Clone)
ComfyUI Node

Raon OpenTTS Generate (Voice Clone)

Zero-shot voice cloning with Raon-OpenTTS (F5-TTS-style CFM/DiT + HiFi-GAN).

By Saganaki22·Created 6 days ago·Updated 6 days ago· 11
Raon OpenTTS Generate (Voice Clone)
  • raon_model
  • ref_audio
  • audio
textHello! This is Raon OpenTTS running natively inside ComfyUI.
ref_text
steps32
cfg_strength2.00
sway_sampling_coef-1.00
speed1.00
seed42
fix_duration_seconds0.0
target_rms0.10
use_vad_durationtrue
cross_fade_ms150
do_splittrue
max_chars0
CategoryRaonOpenTTS

Inputs (15)

NameTypeDefaultDescription
raon_modelRAON_OPENTTS_MODEL
textSTRINGHello! This is Raon OpenTTS running natively inside ComfyUI.Text to synthesize. Long text is split into chunks automatically and cross-faded together.
ref_audioAUDIOReference voice clip for zero-shot cloning. Clean speech with little noise works best.
ref_textSTRINGExact transcript of the reference clip. Strongly improves cloning quality. Use the Whisper Transcribe node to generate it.
stepsINT321–64NFE steps for the flow-matching ODE (euler + EPSS schedule). 32 is the official default.
cfg_strengthFLOAT2.000–10Classifier-free guidance strength. 2.0 is the official default; 0 disables CFG.
sway_sampling_coefFLOAT-1.00-2–2Sway sampling coefficient for the time schedule. -1.0 is the official default.
speedFLOAT1.000.5–2Speech-rate multiplier for duration estimation (>1 = faster).
seedINT420–21474836470 uses the current random state. A positive value is repeatable (incremented per text chunk).
fix_duration_secondsFLOAT0.00–60Force the generated segment length in seconds. 0 estimates it from the reference speech rate (official behaviour).
target_rmsFLOAT0.100–0.5Loudness normalization target for the reference clip (official: 0.1).
use_vad_durationBOOLEANtrueEstimate the generation length from the VAD-trimmed reference length (official default, robust for quiet speakers) while conditioning on the untrimmed audio.
cross_fade_msFLOAT1500–500Cross-fade between generated text chunks in milliseconds (official: 150).
do_splitBOOLEANtrueSplit long text into chunks and generate them one by one (cross-faded together). Off = always one chunk, whatever the length.
max_charsINT00–2000Chunk size budget in UTF-8 bytes. 0 = auto, estimated from the reference speech rate (official behaviour: ref_bytes/ref_seconds x (22 - ref_seconds)). Any positive value forces that budget, so the split becomes deterministic across speakers.

Outputs (1)

NameTypeDescription
audioAUDIO