Skip to content

Speak

The Speak node is a TTS-only step: it voices whatever text its input selector resolves to, segments it into sentences, dispatches each sentence to a dedicated TTS strand, and emits PCM audio frames — without running any language-model inference. Use it to voice upstream-produced text, for example CannedResponse → Speak so a scripted line is spoken without paying for a generation pass.

The text it speaks is selected by input, exactly like every other input-bearing node: leave it empty to voice the user_message slot, or set it to the name of any upstream node's output slot (e.g. a preceding Generate or CannedResponse) to voice that instead. If the resolved slot is missing or empty, the node synthesizes nothing and exits via default.

Speak only sees text that an upstream node has already finished producing, so it cannot overlap synthesis with generation. Chained after Generate, audio begins only once generation has fully completed (Speak then pipelines TTS across the sentences of that completed text). When the goal is simply to speak a generated answer with the lowest possible delay, use Generate and Speak, which streams synthesis while the LLM is still generating. Reach for a standalone Speak when the text it voices did not come from a generation pass (e.g. CannedResponse → Speak), or when you need to branch on / transform the text before voicing it.

The voice is chosen by tts_model_name. Leave it empty to use the first TTS voice available in the server's model catalog, or set it to pick a specific voice.

Audio format is announced once (via OnTtsAudioFormat) before the first audio chunk; end-of-stream is signalled by turn completion. Unlike Generate and Speak, Speak does not emit answer-text deltas — the upstream node that produced the text already did.

NodeType: Speak.

Parameters

Param Type Default Range Structural Description
tts_model_name Optional[str] inherit model default — ✓ TTS model catalog name. Optional — when empty the server uses the first available TTS voice in the model catalog. Set it to pick a specific voice.
speaker_id int 0 ≥ 0.0 — TTS speaker / voice index (default 0). Valid range is 0 to the model's speaker count minus 1 — the shipped Supertonic 3 bundle has 10 voice styles (0–9); Pocket TTS is single-speaker, so it must stay 0 there (pick a Pocket voice with tts_voice instead). Agent creation fails with an out-of-range error for values the model cannot satisfy.
speed float 1.0 0.1 – 3.0 — TTS speech-rate multiplier. 1.0 = native speed.
tts_lang Optional[str] session locale -> model default — — TTS language hint for multilingual models that select the language per call (e.g. Supertonic 3: "en", "de", "fr", …). Resolution order: this param, then the session locale (CreateSessionRequest.locale), then the model's catalog default. Set it only to override the session — one character who keeps speaking French in a German playthrough. For a whole-game language switch, set the session locale instead and leave this empty: it reaches every agent and every node with no per-node mutation. Ignored by families whose language is fixed by the model itself (e.g. Pocket TTS per-language models).
tts_voice Optional[str] inherit model default — — Reference-voice WAV for zero-shot voice-cloning families (Pocket TTS): a path relative to the session storage root pointing at a 10–20 s clean mono recording of the target voice. Empty = the model's default voice. Ignored by fixed-voice families (use speaker_id there).
input Optional[str] inherit model default — ✓ Slot name this node voices. Empty = "user_message" (same default as every other input-bearing node). Structural: immutable after creation — rebinding would re-wire the slot dataflow that is validated once at agent creation.
min_sentence_chars int 12 ≥ 1.0 — Minimum characters before a sentence boundary fires. Lower values reduce first-utterance latency at the cost of more, smaller TTS calls.
tts_voice_description Optional[str] (multiline) inherit model default — — Natural-language stable voice identity: age, accent, pitch, timbre and baseline pacing. Empty = use speaker_id / catalog default. Appended after default_exit so existing FlatBuffers field IDs do not move.
tts_delivery_instruction Optional[str] (multiline) inherit model default — — Natural-language sustained delivery direction for this synthesis call. Empty = neutral / model default.

Exits

Each exit is a structural string field on the node's params; its value names the target node (empty = END).

Exit Param field Description
default default_exit Default exit target — routes here after synthesis completes. Empty string = END.

Exit routes

Route Fires when
default Always, after synthesis completes (including when the resolved text was empty).

Side effects

  • Emits OnTtsAudioFormat once, then OnTtsAudio PCM frames as sentences are synthesized.
  • Does not write a slot or emit OnAnswerText (Speak only voices an existing slot; it produces no text of its own).

Diagnostics

When enable_diagnostics = true, this node contributes to debug_info.nodes[].diagnostics. See Turn Diagnostics JSON for the envelope.

Speak runs no language model, so it emits only a parameters section — no input, output or engine. Unlike Generate and Speak, the voice settings live in parameters rather than in a separate tts section.

Key Meaning Absent when
parameters.input The slot whose text was voiced. Using the user_message default.
parameters.speaker_id Speaker index within the voice model. —
parameters.speed Playback rate multiplier. —
parameters.tts_lang Requested synthesis language. —
parameters.tts_voice Reference voice used. —
parameters.tts_voice_description The voice description prompt. Not set.
parameters.tts_delivery_instruction The delivery instruction. Not set.
parameters.sample_rate The TTS model's native sample rate, in Hz. —
parameters.min_sentence_chars Sentence-boundary threshold before a chunk is synthesized. —
parameters.time_to_first_audio_ms Delay before the first audio chunk was produced. No audio was produced.

Because the node produces no text, an unexpectedly silent turn is diagnosed upstream: check the output.text of whichever node fed input, and the interaction slot snapshot.