OpenAI voice and speech

The bundled openai plugin registers speech synthesis for the tts surface.

Available models: gpt-4o-mini-tts, gpt-4o-mini-tts-2025-12-15, tts-1, tts-1-hd. Available voices: alloy, ash, ballad, cedar, coral, echo, fable, juniper, marin, onyx, nova, sage, shimmer, verse.

extraBody is merged into /audio/speech request JSON after OpenClaw's generated fields, so use it for OpenAI-compatible endpoints that require additional keys such as lang. Prototype keys are ignored.

{
  tts: {
    providers: {
      openai: { model: "gpt-4o-mini-tts", speakerVoice: "coral" },
    },
  },
}

The bundled openai plugin registers batch speech-to-text through OpenClaw's media-understanding transcription surface.

Batch transcription can use the selected OpenAI API-key or ChatGPT OAuth profile on the standard transcription endpoint when the account permits it. Configured models, prompts, and language hints work through the same request path. Access and quota errors are reported without switching credential classes; OAuth support does not imply included or unlimited transcription. Custom endpoints and request overrides require an API-key profile. See Audio and voice notes for selecting a separate audio API-key profile when desired.

• Default model: gpt-4o-transcribe
• Endpoint: OpenAI REST /v1/audio/transcriptions
• Input path: multipart audio file upload
• Used wherever inbound audio transcription reads tools.media.audio, including Discord voice-channel segments and channel audio attachments

To force OpenAI for inbound audio transcription:

{
  tools: {
    media: {
      models: [
        {
          type: "provider",
          provider: "openai",
          model: "gpt-4o-transcribe",
          capabilities: ["audio"],
        },
      ],
      audio: {
        enabled: true,
      },
    },
  },
}

Language and prompt hints are forwarded to OpenAI when supplied by the shared audio media config or per-call transcription request.

The bundled openai plugin registers realtime transcription for the Voice Call plugin.

The bundled openai plugin registers realtime voice for the Voice Call plugin.

Available built-in Realtime voices for gpt-realtime-2.1: alloy, ash, ballad, coral, echo, sage, shimmer, verse, marin, cedar. OpenAI recommends marin and cedar for the best Realtime quality. This is a separate set from the Text-to-speech voices above; a TTS-only voice such as fable, nova, or onyx is not valid for Realtime sessions. Set the model explicitly to gpt-realtime-2.1-mini when you prefer the smaller, lower-cost Realtime 2.1 variant.

Custom Realtime endpoints

Set talk.realtime.providers.openai.baseUrl to the full WebSocket endpoint of an OpenAI Realtime-compatible server, and select gateway-relay. Keep provider: "openai": a custom URL does not register a new provider.

{
  talk: {
    realtime: {
      provider: "openai",
      transport: "gateway-relay",
      model: "your-realtime-model",
      providers: {
        openai: {
          apiKey: "${REALTIME_API_KEY}",
          baseUrl: "wss://voice.example.com/custom/realtime?realtime=true",
        },
      },
    },
  },
}

The default is wss://api.openai.com/v1/realtime. The URL's path and query parameters are preserved; OpenClaw sets the model query parameter from the selected model. Standard URL parsing applies; use a properly URL-encoded query for signed endpoints. Include the complete endpoint path, not only an API root such as /v1. https: and http: URLs are converted to wss: and ws: respectively. Prefer TLS outside trusted local networks. URLs with embedded credentials or fragments are rejected.

The same baseUrl option is available in the OpenAI realtime provider block for Voice Call and other Gateway-side voice bridges. It does not change chat, transcription, or TTS endpoints. When a custom endpoint has no explicit model, the GA Realtime default remains in use rather than GPT-Live.

This is a server-side Realtime WebSocket override, not a general protocol adapter: the service must support OpenClaw's Realtime session events, audio formats, voices, and tools. Provider-specific compatibility, including DashScope, is not implied. The API key stays on the Gateway and authenticates the socket directly; no browser client-secret endpoint is required. Browser WebRTC, GPT-Live, and Azure endpoint/deployment settings cannot be combined with this override. Those combinations fail rather than sending credentials or audio to the default OpenAI endpoint.

Gateway-controlled Realtime call cleanup

Closing a Gateway-controlled GA Realtime WebRTC session retires its Gateway authority and closes the local sideband before asking OpenAI to hang up the provider call. These are separate events; control closure does not establish provider acknowledgment or recall already queued media.

If hangup fails, explicit cancellation or cleanup reports the failure. The broker retries automatically after 1 second, then 5 seconds, with the existing 30-second timeout for each attempt. After all three attempts fail, the log reports cleanup INCOMPLETE. The exact cleanup obligation and its capacity remain reserved, including across plugin replacement: eight sessions globally and two per Gateway client. Restore provider connectivity; a later OpenAI broker/plugin runtime cleanup can retry these retained calls. Repeating End or talk.client.close is not that retry boundary because the Gateway session may already be retired.

Cleanup obligations are in memory only. Gateway exit, crash, or restart can lose them; restarting is not proof that the provider call ended. The adapter's 30-minute active-session lease is not a remote-lifetime guarantee or a fallback after failed hangup.

GA Realtime browser authentication

Ordinary GA browser Talk tries Platform auth first in this order: the configured realtime key, an openai API-key profile, then OPENAI_API_KEY. When a Platform credential is available, the Gateway mints an ephemeral client secret and the browser performs the SDP exchange directly.

When no Platform credential source is configured, ordinary GA browser Talk falls back to the OpenClaw ChatGPT OAuth subscription profile. The single-use Gateway offer broker keeps OAuth server-side, exchanges the browser's SDP, and returns only the answer SDP. An explicitly configured but unavailable Platform credential fails instead of falling back to OAuth.

Gateway-controlled GA relay, iOS client-owned WebRTC, GA Voice Call, direct backend sockets, and GA Discord realtime voice require Platform auth.

GPT-Live API

Talk and Gateway relay sessions select GPT-Live when no model is configured. An OpenAI Platform key, API-key profile, or OPENAI_API_KEY selects gpt-live-1 with marin; a ChatGPT-only account selects gpt-live-1-codex with cove. A configured Platform credential takes precedence over ChatGPT sign-in, including when that credential needs repair.

GPT-Live is audio-only, so the Control UI does not offer camera capture with this default. Choose gpt-realtime-2.1 explicitly to use the camera. talk.catalog describes the configured Talk agent's default for its discovery surface; a session scoped to another agent resolves that agent's accounts when it starts.

Explicit model choices and voices supported by that model stay in effect; an explicit GPT-Live model is audio-only. Requests requiring video or forced agent-consult replies retain gpt-realtime-2.1 when no model is pinned. Direct tool bridges, Discord and Voice Call without an explicit model, and Azure deployments retain their existing defaults. An explicit GPT-Live model in Discord or Voice Call uses the shared Gateway relay bridge and provider-owned delegation. Installing an update does not rewrite saved configuration or switch an active session.

Set talk.realtime.model explicitly to gpt-live-1 for the public GPT-Live API. GPT-Live handles the spoken conversation while delegated tasks run through your configured OpenClaw agent. It can listen while speaking; interrupting speech does not itself cancel agent work.

This model requires an OpenAI Platform API key with model access, selected from the configured realtime key, an openai API-key profile, or OPENAI_API_KEY. ChatGPT subscription credentials are not a fallback for gpt-live-1.

{
  talk: {
    realtime: {
      provider: "openai",
      model: "gpt-live-1",
      speakerVoice: "marin",
      transport: "webrtc",
    },
  },
}

Browser Talk uses Gateway-brokered WebRTC; the Platform key stays on the Gateway. Set transport: "gateway-relay" for the direct server WebSocket path. Discord realtime voice and Voice Call with gpt-live-1 use the same Gateway-owned direct WebSocket bridge and native agent delegation.

iOS uses the same brokered WebRTC path and displays public Live captions. Physical-device voice validation remains pending.

The default voice is marin. Supported built-in voices are alloy, ash, ballad, beacon, bossa, cedar, cinder, coral, delta, echo, gleam, marin, meridian, quartz, ripple, sage, shimmer, stone, tempo, verse, vesper, and willow. Unsupported configured voices fall back to marin. Choose the voice before starting a session. cove belongs to gpt-live-1-codex; selecting it with the public gpt-live-1 model falls back to marin.

The public API uses /v1/live/sessions for WebRTC creation and primary WebSockets. Direct sockets start with session.start; browser sessions start during the SDP exchange. A Gateway sideband handles delegated work while browser media stays on WebRTC. Public Live instructions describe session.thinking.append as quiet context and session.commentary.append as spoken updates. Custom instructions should preserve that distinction; the Codex route uses its separate commentary/speakable channel contract.

Both routes instruct the voice model to wait for delegated results instead of repeating the same backend request. New user follow-ups, corrections, and explicit retries remain allowed. These instructions guide model behavior; they do not guarantee that each request executes only once.

Public transcript events are fragments, not completed turns. The Gateway owns persistence of bounded received-text snapshots; clients display captions without saving another copy. Both live captions and saved snapshots can span several exchanges by the same speaker; they do not reconstruct chronological turns or cross-speaker timing. Overlapping fragments do not identify a completed utterance. Saving text does not establish that speech or playback has finished. See OpenAI's session guide for the public session and voice contract.

Released GPT-Live browser and Gateway relay authentication

The separate Codex GPT-Live model, gpt-live-1-codex, retains its ChatGPT subscription route. Its browser and Gateway-relay WebRTC try the OpenClaw ChatGPT OAuth subscription profile first. When OAuth is unavailable, the Gateway falls back to Platform auth in this order: the configured realtime key, an openai API-key profile, then OPENAI_API_KEY. Create the OAuth profile with openclaw models auth login --provider openai.

Discord uses this same Gateway-owned WebRTC bridge when channels.discord.voice.realtime.model is gpt-live-1-codex. Set channels.discord.voice.realtime.speakerVoice to cove for the Codex default voice. The other supported voices are arbor, breeze, ember, juniper, maple, sol, spruce, and vale. A voice selection is specific to its model route, regardless of whether the client is Talk, Discord, or Voice Call.

Voice Call uses this same OAuth-capable bridge when plugins.entries.voice-call.config.realtime.providers.openai.model is gpt-live-1-codex; set voice: "cove" in that provider block. Its audio adapter converts carrier G.711 mu-law at 8 kHz to and from the model's 24 kHz PCM stream for both WebRTC and direct WebSocket routes. Initial greetings and voicecall.speak requests use native session context.

Both GPT-Live routes produce continuous audio and own interruption. Gateway WebSocket and WebRTC use the same sample clock to send microphone input at its recorded rate and supply silence between captures. Discord does not add speaker-start cancellation or wait for a response-done event to play short replies. Explicit Discord requireWakeName: true or consultPolicy: "always" is rejected because GPT-Live cannot enforce those host policies; its default uses automatic provider delegation. Each speaker's delegated work retains their Discord identity and permissions. The shared OpenClaw agent conversation supplies room context, while each speaker's voice-model connection has separate acoustic conversation history. See GPT-Live in Discord.

Voice Call also leaves microphone input open during Live playback and does not add host speech-start cancellation. Native delegations use the existing call-owned agent consult, including its tool policy and cancellation lifetime. Explicit realtime.consultPolicy: "always" is rejected for GPT-Live. openclaw_end_call and custom realtime.tools require native function-tool support and remain unavailable on GPT-Live; delegation does not expose them. See GPT-Live in Voice Call.

Both credential types stay in the Gateway. The single-use offer broker exchanges the browser's SDP and returns only the answer SDP; it does not send an OAuth token, Platform key, or ephemeral client secret to the browser.

Gateway-relay WebRTC paces microphone audio against a continuous sample clock, so timer delays do not accumulate during longer calls. After a scheduler pause, queued audio resumes at its normal pace while timestamps account for the elapsed gap. Calls conceal malformed incoming audio packets and continue playing later audio. One rejected audio packet send does not end an otherwise connected call. Unusable codec state, unexpected stream changes, and terminal connection states still end the call. Packet-drop diagnostics omit raw error details.

The enabled OpenAI plugin starts the broker automatically, including when you sign in after the Gateway has started. The broker opens a provider session only when you start a voice session; signing in does not open the microphone or join Discord voice. Returning to the browser after sign-in refreshes the chat microphone's readiness.

Unlisted and private realtime transport paths

Unlisted or private browser Talk uses Platform-key client WebRTC with Gateway-owned control. Gateway relay and other direct backend consumers use the Platform-key bidirectional transport. Credentials and provider control remain on the Gateway.

Use the account-issued realtime model value. Unlisted model values are accepted as free-form Talk config but are not published through catalogs or diagnostics. Opt in explicitly with talk.realtime.model; the released model remains the default.

Current Platform-key sessions accept marin and cedar. OpenClaw defaults to marin and maps unsupported configured voices back to it.

Unlisted or private browser WebRTC prerequisites, in order:

1. A Platform API key configured through talk.realtime.providers.openai.apiKey, an openai API-key profile, or OPENAI_API_KEY.
2. talk.realtime.model set to the account-issued value — via Settings → Talk in the Control UI or the config below.
3. The bundled openai plugin registered in full mode. A restrictive plugins.allow list fails with "OpenAI realtime browser session broker is unavailable".

{
  talk: {
    realtime: {
      provider: "openai",
      model: "<account-issued-realtime-model>",
      transport: "webrtc",
    },
  },
}

Gateway relay uses the direct bidirectional transport:

{
  talk: {
    realtime: {
      provider: "openai",
      model: "<account-issued-realtime-model>",
      transport: "gateway-relay",
    },
  },
}

Browser Talk uses transport: "webrtc".

These rows describe implemented transports, not account entitlement or complete model capability parity. See the Discord voice policy limits and Voice Call tool limits before selecting an unlisted or private route for those consumers.

A 403 Voice session access denied response is overloaded and does not by itself prove an account entitlement problem: an invalid voice produces the same response. First verify the model and voice against the accepted lists above, then verify the Platform key and configured model against the same project.

The Codex GPT-Live Gateway-owned WebRTC route uses OAuth first with Platform fallback, routes sideband delegations through the configured OpenClaw agent, and keeps credentials away from relay clients. Unlisted or private browser WebRTC and the direct backend socket remain Platform-only. The direct socket enables Discord voice and Voice Call/telephony; OpenClaw converts G.711 u-law telephony audio to and from the provider's 24 kHz PCM stream. Android's client-side gate stays closed until the Gateway relay path has live proof from an Android device.

The WebRTC path creates a provider call and joins its sideband. The direct unlisted/private backend path opens one bidirectional session, sends a Frameless session.update, then carries PCM audio, transcripts, delegations, and delegation results over that socket.

Maintainers can exercise the Platform direct path and the separate GA browser OAuth path with the opt-in live tests. The account-issued realtime model is read from talk.realtime.model; missing credentials or model config produce sanitized skips, and the tests never print either value:

OPENCLAW_LIVE_TEST=1 OPENCLAW_LIVE_GPT_LIVE=1 node --import tsx scripts/test-live.mts -- extensions/openai/realtime-quicksilver.live.test.ts
OPENCLAW_LIVE_TEST=1 OPENCLAW_LIVE_GPT_LIVE=1 node --import tsx scripts/test-live.mts -- extensions/openai/realtime-quicksilver-gateway-bridge.live.test.ts