Voices

The voice is config.tts.voice: an id from the chosen provider’s roster. Browse rosters with GET /v1/catalog/voices (catalog_voices over MCP) before writing an id; a wrong id fails at synthesis, on a live call. A realtime agent picks its voice on config.realtime from the model’s own roster.

The `tts` object

Field	Type	Meaning
`provider`, `model`	string	A `tts` catalog entry for the agent’s channel. Default `sarvam/bulbul:v3`.
`voice`	string \| null	Voice id as the voices endpoint returns it. `null` saves the entry’s `default_voice`.
`voice_name`	string \| null	Dashboard label; never sent to the provider.
`speed`	number, default `1.0`	0.5–2.0, then the entry’s `speed_min`–`speed_max`. Only `1.0` when `supports_speed` is false.
`stability`, `similarity_boost`	number \| null	0.0–1.0. Only on entries with `supports_voice_settings` (ElevenLabs); refused elsewhere.
`expressive`	boolean, default `false`	Let the agent write delivery tags. See expressive delivery.
`fallback`	object \| null	Second voice with its own settings. See model fallbacks.

There is no language on tts; the agent’s one language applies.

Browsing voices

curl -G https://api.talqing.com/v1/catalog/voices \
  -H "Authorization: Bearer $TALQING_API_KEY" \
  --data-urlencode "provider=elevenlabs" \
  --data-urlencode "language=hi" \
  --data-urlencode "gender=female"

Parameter	Meaning
`provider`	Required.
`model`	Scope to one model. Pass it for `raya`, `soniox` and `deepgram`, whose voices work on one model only; required for `tts_cache_router`.
`kind`	`tts` (default) or `realtime`.
`search`	Free text over name, id and description. Every provider.
`language`, `accent`, `gender`	Facet filters, where the provider has them. `accent` needs a `language`.
`page`	ElevenLabs only; read `has_more`.

Each voice has id and, where the provider publishes them, name, gender, language, accent, country, locale, description, sample_url, image_url, tier and owner_id. Only id goes on the agent. The response also returns the facets the provider supports (languages, accents, genders); an empty list means no such filter. With a provider key saved, the listing runs on your key and workspace_key is true; for ElevenLabs your own account’s voices come first.

Provider	Roster	Filters
`elevenlabs`	Your account, then the shared library	language, accent, gender
`deepgram`	Live; the voice id is the model string and encodes the language (`aura-2-thalia-en`)	language, accent, gender
`xai`	Live; shared by TTS and Grok Voice	language, gender
`raya`, `soniox`	Live, per model	`raya`: language. `soniox`: gender
`openai`, `sarvam`, `gemini`	Fixed list in the catalog entry	gender

A tts_cache_router model (model=elevenlabs/eleven_flash_v2_5) lists the underlying provider’s voices, through your router account when a router key is saved.

ElevenLabs voices need their settings

An ElevenLabs voice set without stability and similarity_boost speaks with ElevenLabs’ global defaults (0.5 / 0.75), not its own tuning. Talqing always sends a speed, and ElevenLabs then resets every omitted setting to the global default.

Find the voice

GET /v1/catalog/voices?provider=elevenlabs. A voice already in your ElevenLabs account has no owner_id; skip to step 3.

Save a shared-library voice

POST /v1/catalog/voices/elevenlabs/add with voice_id, owner_id and name. It is added to your workspace’s ElevenLabs account; use the returned voice_id from here on.

Read its settings

GET /v1/catalog/voices/elevenlabs/{voice_id}/settings returns stability and similarity_boost. An unsaved library id returns 404.

Write all three onto tts and publish

voice, stability and similarity_boost together. null settings mean the voice has none of its own; leave them unset.

VOICE_ID=$(curl -s -X POST https://api.talqing.com/v1/catalog/voices/elevenlabs/add \
  -H "Authorization: Bearer $TALQING_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"voice_id":"hpp4J3VqNfWAUOO0d1Us","owner_id":"a1b8c0d2e3f4a5b6c7d8e9f0a1b2c3d4","name":"Hope"}' \
  | python3 -c 'import json,sys; print(json.load(sys.stdin)["voice_id"])')

curl -s "https://api.talqing.com/v1/catalog/voices/elevenlabs/$VOICE_ID/settings" \
  -H "Authorization: Bearer $TALQING_API_KEY"
# {"stability":0.45,"similarity_boost":0.8}

curl -X PATCH https://api.talqing.com/v1/agents/6f1d0a2c-6a0e-4f77-9b1e-6c1a2d3e4f50 \
  -H "Authorization: Bearer $TALQING_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"config\": {\"tts\": {\"provider\": \"elevenlabs\", \"model\": \"eleven_flash_v2_5\",
       \"voice\": \"$VOICE_ID\", \"stability\": 0.45, \"similarity_boost\": 0.8}}}"

const agent_id = "6f1d0a2c-6a0e-4f77-9b1e-6c1a2d3e4f50";

const { voices } = await talqing.catalog.voices.list({ provider: "elevenlabs", language: "en" });
const picked = voices[0];

let voiceId = picked.id;
if (picked.owner_id) {
  const added = await talqing.catalog.voices.elevenlabs.add({
    addVoiceRequest: { voice_id: picked.id, owner_id: picked.owner_id, name: picked.name ?? picked.id },
  });
  voiceId = added.voice_id;
}

const settings = await talqing.catalog.voices.elevenlabs.settings({ voice_id: voiceId });

await talqing.agents.update({
  agent_id,
  updateAgentRequest: {
    config: {
      tts: {
        provider: "elevenlabs",
        model: "eleven_flash_v2_5",
        voice: voiceId,
        voice_name: picked.name,
        stability: settings.stability,
        similarity_boost: settings.similarity_boost,
      },
    },
  },
});
await talqing.agents.publish({ agent_id });

agent_id = "6f1d0a2c-6a0e-4f77-9b1e-6c1a2d3e4f50"

picked = talqing.catalog.voices.list(provider="elevenlabs", language="en")["voices"][0]

voice_id = picked["id"]
if picked.get("owner_id"):
    voice_id = talqing.catalog.voices.elevenlabs.add(
        voice_id=picked["id"], owner_id=picked["owner_id"], name=picked.get("name") or picked["id"]
    )["voice_id"]

settings = talqing.catalog.voices.elevenlabs.settings(voice_id)

talqing.agents.update(
    agent_id,
    config={
        "tts": {
            "provider": "elevenlabs",
            "model": "eleven_flash_v2_5",
            "voice": voice_id,
            "voice_name": picked.get("name"),
            "stability": settings["stability"],
            "similarity_boost": settings["similarity_boost"],
        }
    },
)
talqing.agents.publish(agent_id)

The dashboard’s voice picker does all four steps. Over MCP they are catalog_voices, add_elevenlabs_voice and elevenlabs_voice_settings; all three take provider=tts_cache_router for an ElevenLabs voice behind the router. The settings belong to one voice: re-read them when you change the voice, and clear them when you move to a model without supports_voice_settings, or the save is refused (text-to-speech model '…' has no stability setting - clear it, or pick one of: …).

Expressive delivery

Laughs, whispers and pauses on the models that speak tags.

Language

The one language every speech model shares.

Get started

Build

Guides

The `tts` object

Browsing voices

ElevenLabs voices need their settings

Next

Expressive delivery

Language

​The tts object

​Browsing voices

​ElevenLabs voices need their settings

​Next

Expressive delivery

Language

The `tts` object

Browsing voices

ElevenLabs voices need their settings

Next