Nodes/MOSS-TTS/MOSS-TTS Voice Clone
ComfyUI Node

MOSS-TTS Voice Clone

Zero-shot voice cloning from a ComfyUI AUDIO reference.

By Saganaki22·Created 2 months ago·Updated 2 months ago· 31
MOSS-TTS Voice Clone
  • mosstts_model
  • reference_audio
  • audio
textThis line will be spoken in the reference voice.
languageauto
instruction
duration_tokens0
quality
sound_event
ambient_sound
max_new_tokens4096
do_sampletrue
text_temperature1.00
text_top_p1.00
text_top_k50
audio_temperature1.70
audio_top_p0.80
audio_top_k25
audio_repetition_penalty1.00
seed0
CategoryMOSS-TTS

Inputs (19)

NameTypeDefaultDescription
mosstts_modelMOSSTTS_MODELLoaded MOSS-TTS model bundle.
reference_audioAUDIOReference speaker audio. The node encodes it with MOSS-Audio-Tokenizer-v2 before generation.
textSTRINGThis line will be spoken in the reference voice.Text to synthesize. Inline [pause 0.5s], Pinyin, and IPA are passed through to MOSS-TTS.
languageCOMBOautoLanguage hint. v1.5 performs best when the language is specified.
instructionSTRINGOptional free-form style or delivery instruction.
duration_tokensINT00–45000Optional duration target in audio tokens. 0 omits the field. MOSS runs around 12.5 tokens per second.
qualitySTRINGOptional quality hint field exposed by the MOSS user-message schema.
sound_eventSTRINGOptional sound-event hint field exposed by the MOSS user-message schema.
ambient_soundSTRINGOptional ambient-sound hint field exposed by the MOSS user-message schema.
max_new_tokensINT40961–45000Generation budget in audio frames. At 12.5 frames per second, 4096 is roughly 5.5 minutes.
do_sampleBOOLEANtrueUse stochastic sampling. Disable for deterministic greedy decoding.
text_temperatureFLOAT1.000–2Sampling temperature for assistant text/audio-control tokens.
text_top_pFLOAT1.000–1Nucleus sampling for assistant text/audio-control tokens.
text_top_kINT500–4096Top-K sampling for assistant text/audio-control tokens.
audio_temperatureFLOAT1.700–3Recommended v1.5 audio sampling temperature.
audio_top_pFLOAT0.800–1Recommended v1.5 nucleus sampling cutoff for audio codebooks.
audio_top_kINT250–1024Recommended v1.5 Top-K cutoff for audio codebooks.
audio_repetition_penaltyFLOAT1.001–2Penalty for repeated acoustic code patterns.
seedINT00–92233720368547760000 leaves sampling unseeded. Positive values make identical settings repeatable.

Outputs (1)

NameTypeDescription
audioAUDIO