optimistic-goldâ˘3mo ago
The newly announced voice input settings is not even working
Can only use the deepgram model and if you try to use the other model like the cartesia, it will give you an error (in canvas) while in prod, it just doesn't work. Total disappointment after that announcement.
3 Replies
Hey there đ sorry for the confusion!
⢠Cartesia was added only for speech-to-speech output (the voice your agent speaks with).
⢠All speech-to-text input is still handled by Deepgram under the hood â there isnât a Cartesia recogniser today. If you select a Cartesia model in the Input dropdown inside the new Voice settings, the canvas will throw an error and the live agent wonât transcribe anything â exactly what youâre seeing. Fix 1. Open your project â Settings â Behaviour â Voice.
2. Under Input model keep the default âDeepgramâ (Speed / Balanced / Accuracy presets or Advanced).
3. Under Output voice you can pick any of the new Cartesia voices (Sonic 2 or Sonic Turbo) or another provider. 4. Save & redeploy. The team is working on adding more STT engines, but for now Deepgram is the only supported option for recognition, so sticking with it will restore full functionality. Quick tuning tip: if you need faster responses, move the Simplified slider toward âSpeedâ; for more accurate transcripts, choose âAccuracyâ or increase the Endpointing / Timeout values in Advanced mode. Hope that clears things up â let me know if anythingâs still not behaving đ (Docs for reference: https://docs.voiceflow.com â see âConfiguring Voice Settingsâ and âCartesia Voicesâ)
⢠All speech-to-text input is still handled by Deepgram under the hood â there isnât a Cartesia recogniser today. If you select a Cartesia model in the Input dropdown inside the new Voice settings, the canvas will throw an error and the live agent wonât transcribe anything â exactly what youâre seeing. Fix 1. Open your project â Settings â Behaviour â Voice.
2. Under Input model keep the default âDeepgramâ (Speed / Balanced / Accuracy presets or Advanced).
3. Under Output voice you can pick any of the new Cartesia voices (Sonic 2 or Sonic Turbo) or another provider. 4. Save & redeploy. The team is working on adding more STT engines, but for now Deepgram is the only supported option for recognition, so sticking with it will restore full functionality. Quick tuning tip: if you need faster responses, move the Simplified slider toward âSpeedâ; for more accurate transcripts, choose âAccuracyâ or increase the Endpointing / Timeout values in Advanced mode. Hope that clears things up â let me know if anythingâs still not behaving đ (Docs for reference: https://docs.voiceflow.com â see âConfiguring Voice Settingsâ and âCartesia Voicesâ)
optimistic-goldOPâ˘3mo ago
additionally, you cant even edit the advance settings on deepgram.
"If you select a Cartesia model in the Input dropdown inside the new Voice settings, the canvas will throw an error and the live agent wonât transcribe anything â exactly what youâre seeing." - Then why is it even on the voice input options, including the AssemblyAI?
Oof, this is a bug on our side then - we'll see what happened. Thanks for pointing out, we didn't do a major release announcement just in-app @Seiha
this is working now - sorry for the bug đ @Seiha