Seed Voice Clone

Node name

BytePlus Seed Voice Clone

Node ID

BytePlusSeedVoiceClone

Find it

Double-click the canvas and search for “Seed Voice Clone”, or browse BytePlus ModelArk/Speech

Keys needed

Seed Speech API key

Runs when

Its output is used by another node, such as Save Image or Save Video. It is not an output node.

What it does

Trains a voice on a clip of someone speaking and outputs its speaker_id. You then give that ID to Seed Speech TTS with the model seed-icl-2.0, or use it as a reference source in Seed Audio 1.0.

The node needs your separate Seed Speech API key (Singapore region only) and a voice slot from the Seed Speech console.

On the canvas

Default settings. Connect the reference clip to audio and enter a voice slot ID.

Inputs

Input	Type	Default	What it does
speaker_id	text		Voice slot ID (S_...) bought in the Seed Speech console, or your own postpaid custom voice ID. Each slot can be trained 15 times; every run with changed inputs trains again.
language	choice	en	Language spoken in the reference clip. Cross-language cloning is not supported.Choices: en, zh, ja, es, id, pt, de, fr, ko, it, th, vi … (21 in total)
reference_text	text		Optional: the text read in the clip. Training fails if the audio differs too much.
demo_text	text		Optional: text for the demo clip (4-80 characters, same language).
audio	AUDIO		Reference voice clip, e.g. from Load Audio: 10-15 s of clear speech, up to 10 MB.

Advanced inputs (1)

Input	Type	Default	What it does
disable_volume_normalization	on/off	off	Keep the reference clip's loudness instead of normalizing it (closer similarity).

Notes on inputs

• speaker_id: a voice slot ID that starts with S_, bought in the Seed Speech console, or your own postpaid custom voice ID. A custom ID has 8 to 256 letters, digits, - or _, starts with a letter, does not end with - or _, and avoids the reserved prefixes and suffixes of Seed Speech.
• Reference clip: connect 10 to 15 seconds of clear speech to audio, for example from Load Audio. The clip is sent as WAV, which must stay within 10 MB.
• language: the language spoken in the clip. Cloning across languages is not supported.
• reference_text: optional. If you give the text that is read in the clip, training can fail when the audio differs too much from it.
• demo_text: optional text of 4 to 80 characters for the demo clip, in the same language.

Outputs

Output	Type	What it holds
speaker_id	text
demo_audio	AUDIO
status_json	text

Notes on outputs

• speaker_id is the ID you entered for the voice. Pass it to custom_speaker_id of Seed Speech TTS.
• demo_audio is the demo clip from the service. If the service returns none, the node outputs half a second of silence and logs a message.
• status_json is the last status answer of the service as JSON, without its top-level demo_audio link.

Limits and tips

• Each voice slot can be trained 15 times. Every run with changed inputs trains again and uses one of them (see the tooltip of speaker_id).
• The node starts the training, then checks its status every 2 seconds. It stops with an error after 600 seconds if the voice is still training.
• The voice is ready when the status is 2 (success) or 4 (active).
• The first TTS call with a new voice activates the slot and starts its billing.
• Turn on disable_volume_normalization only if you want the loudness of the clip kept. According to its tooltip, this gives a closer match.
• Interrupting a run stops the node’s wait. The node sends no cancel request for training that has already started.

Common errors

Message	What to do
[BytePlus] speaker_id is empty: enter a voice slot ID (S_...) from the Seed Speech console, …	Enter a voice slot ID (S_...) from the Seed Speech console, or your own postpaid custom voice ID.
[BytePlus] '…' is not a valid custom voice ID: 8-256 letters, digits, - or _, …	Fix the ID: 8 to 256 letters, digits, - or _, starting with a letter, not ending with - or _, without reserved prefixes such as S_ or suffixes such as _bigtts.
… This voice slot has no training attempts left (15 per slot). Use another speaker ID.	The slot was trained 15 times. Use another speaker ID.
… The reference audio does not match reference_text.	Empty reference_text, or type exactly what is said in the clip.
… The reference audio is too noisy (SNR check failed).	Record or choose a cleaner clip.

Used in templates

• Seed Voice Clone