vercel/ai @ai-sdk/elevenlabs@3.0.68 on GitHub

9874c89: feat(elevenlabs): text-to-dialogue streaming
doStream() supports providerOptions.elevenlabs.dialogue (Text-to-Dialogue WebSocket for eleven_v3 / eleven_v4). Each text part's voice (default: the call's voice, else the first dialogue voice) selects its speaker and must be listed in dialogue.voices; a voice change starts a new turn, and providerOptions.elevenlabs.newTurn on a part forces one. Options Text-to-Dialogue does not support produce warnings.
109a4fe: feat(elevenlabs): streaming speech over WebSocket; type v4/v3-conversational model ids
ElevenLabsSpeechModel implements the experimental Experimental_StreamingSpeechModelV4 (doStream()) over the text-to-speech stream-input WebSocket, forwarding text deltas as they arrive and reporting usage.inputCharacters (Unicode code points). Adds eleven_v4, eleven_v4_turbo, and eleven_v3_conversational to the typed speech model IDs. The API key is sent in the WebSocket's first message (native WebSocket works), the portable outputFormat maps to an ElevenLabs output_format (or set streaming.outputFormat natively), text parts with a different voice are rejected, partial words are held back to whitespace, and keepalives are sent while the text stream is idle unless inactivityTimeout is set. generateSpeech ignores streaming options with a warning.
Updated dependencies [8202f5b]
Updated dependencies [85d84a1]
- @ai-sdk/provider-utils@5.0.67
- @ai-sdk/provider@4.0.33