Seed Speech TTS

Node name: BytePlus Seed Speech TTS
Node ID: BytePlusSeedTTS
Find it: Double-click the canvas and search for “Seed Speech TTS”, 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

Reads your text aloud with a Seed Speech voice and returns it as ComfyUI audio. Pick a voice from the voice list with seed-tts-2.0, or type a speaker ID in custom_speaker_id to use a TTS 1.0 speaker or a voice you cloned.

The node needs your separate Seed Speech API key (Singapore region only), and every run sends one billed request. It does not save anything itself: connect Save Audio or Preview Audio to keep or hear the result.

On the canvas

The Seed Speech TTS node with model seed-tts-2.0, a short English sentence and the voice en_female_stokie_uranus_bigtts — Default settings with the TTS 2.0 voice list.

Inputs

Input	Type	Default	What it does
`model`	choice	`seed-tts-2.0`	seed-tts-2.0 for the voice list; seed-icl-* for cloned voices (custom_speaker_id).Choices: seed-tts-2.0, seed-tts-1.0, seed-icl-2.0, seed-icl-1.0
`text`	text
`voice`	choice	`en_female_stokie_uranus_bigtts`	TTS 2.0 voice. Each voice supports specific languages; see the Seed Speech voice list.Choices: en_female_stokie_uranus_bigtts, en_female_dacey_uranus_bigtts, en_male_tim_uranus_bigtts, en_female_nadia_uranus_bigtts, en_male_kevin_uranus_bigtts, en_male_jamie_uranus_bigtts, en_female_hayley_uranus_bigtts, en_female_skye_uranus_bigtts, en_male_bruce_uranus_bigtts, en_male_alex_uranus_bigtts, en_male_yangguangjieshuonan_uranus_bigtts, en_male_ronald_uranus_bigtts … (177 in total)
`custom_speaker_id`	text		Overrides voice: a cloned voice (connect Seed Voice Clone's speaker_id, model seed-icl-2.0) or a TTS 1.0 speaker ID.
`context_text`	text		TTS 2.0 only: style instruction or preceding dialogue, e.g. 'Speak slowly, sounding heartbroken.' Not billed.
`emotion`	text		Emotion for voices that support it, e.g. happy, sad, angry. Empty = neutral.
`emotion_scale`	integer	`4`	Range: 1–5
`speech_rate`	integer	`0`	100 = 2x speed, -50 = 0.5x.Range: -50–100
`loudness_rate`	integer	`0`	100 = 2x volume, -50 = 0.5x.Range: -50–100
`pitch`	integer	`0`	Pitch shift in semitones.Range: -12–12
`sample_rate`	choice	`24000`	Hz. seed-tts-2.0 supports 24000, 16000 and 8000.Choices: 24000, 16000, 8000, 22050, 32000, 44100, 48000
`explicit_language`	choice	`auto`	Read only text in this language. auto handles mixed Chinese and English.Choices: auto, en, zh-cn, ja, ko, id, th, vi, ms, fil, es, es-mx … (22 in total)
`silence_duration`	integer	`0`	Silence added after the last sentence, in ms.Range: 0–30000
`filter_markdown`	on/off	`off`	Strip Markdown syntax so bold is read as 'bold'.
`enable_subtitle`	on/off	`off`	Return timestamps (Chinese and English): subtitles for 2.0 voices, word timestamps for 1.0 voices.
`seed`	integer	`0`	Not sent to the API; change it to synthesize again.Range: 0–18446744073709552000

Advanced inputs (8)

Input	Type	Default	What it does
`detect_language`	on/off	`off`	Detect the text language automatically.
`context_language`	choice	`default`	Reference language for Western European text: default English, id Indonesian, es Mexican Spanish, pt Brazilian Portuguese.Choices: default, id, es, pt
`read_emoji`	on/off	`off`	Keep emoji in the text instead of filtering them out.
`read_latex`	on/off	`off`	Read LaTeX formulas aloud (also enables filter_markdown).
`read_parentheses`	on/off	`off`	Read text inside parentheses (filtered out by default).
`unsupported_char_ratio`	number	`0.3`	Fail when more than this share of the text is in an unsupported language.Range: 0–1
`use_cache`	on/off	`off`	Reuse audio synthesized for identical text in the last hour (no timestamps from cache).
`tone_fidelity`	on/off	`off`	seed-icl-2.0 only: stay as close as possible to the training audio's voice, emotion and accent (same language only).

Notes on inputs

Models: seed-tts-2.0 uses the voice list. seed-tts-1.0, seed-icl-2.0 and seed-icl-1.0 have no entry in that list: they need a speaker ID in custom_speaker_id. Use seed-icl-2.0 with the speaker_id from Seed Voice Clone.
custom_speaker_id wins: when it is filled, voice is ignored. The voice list always has a value, so you can leave it as it is.
Style: context_text (TTS 2.0 only) takes an instruction such as “Speak slowly, sounding heartbroken.” emotion and emotion_scale work for voices that support emotions. An empty emotion means neutral.
Subtitles: turn on enable_subtitle to fill subtitles_json and srt. The 2.0 models (seed-tts-2.0, seed-icl-2.0) return subtitles, the 1.0 models word timestamps.
Language: explicit_language limits reading to one language. auto copes with mixed Chinese and English.
seed: it is not sent to the API. Change it to synthesize the same text again.

Outputs

Output	Type	What it holds
`audio`	AUDIO
`subtitles_json`	text
`srt`	text

Notes on outputs

audio is decoded from the raw PCM the API streams back, at the sample_rate you chose.
srt stays empty and subtitles_json has no segments unless enable_subtitle is on and the service returns timestamps.

Limits and tips

seed-tts-2.0 accepts only the sample rates 24000, 16000 and 8000. The node stops with an error for the others.
context_text works only with seed-tts-2.0, and tone_fidelity only with seed-icl-2.0 (same language as the training audio).
Timestamps are available for Chinese and English. Cached audio (use_cache, identical text within the last hour) comes without timestamps.
A voice supports specific languages. If the service says the voice is not available, check that the model matches the voice: TTS 2.0 voices need seed-tts-2.0, cloned voices need seed-icl-*.
filter_markdown and the advanced read_emoji, read_latex and read_parentheses control how Markdown, emoji, LaTeX and text in parentheses are read. unsupported_char_ratio makes the request fail when too much of the text is in a language the voice cannot read.
Interrupting a run stops the node’s wait. The node sends no cancel request, so a request that was already sent may still be billed.

Common errors

Message	What to do
`[BytePlus] No Seed Speech API key is set. Add it in Settings > BytePlus …`	Add the Seed Speech key in Settings → BytePlus. See API keys.
`[BytePlus] … needs custom_speaker_id: the voice list only holds TTS 2.0 voices.`	You picked `seed-tts-1.0` or a `seed-icl-*` model. Fill `custom_speaker_id`, or switch to `seed-tts-2.0`.
`… The voice is not available for this key or model. Check the speaker ID …`	Check the speaker ID and that the model fits it. TTS 2.0 voices need `seed-tts-2.0`, cloned voices `seed-icl-*`.
`[BytePlus] seed-tts-2.0 supports the sample rates 24000, 16000 and 8000 Hz.`	Set `sample_rate` to 24000, 16000 or 8000.
`… The text is longer than the model accepts.`	Split the text into shorter parts and run the node once per part.