OpenaIA Realtime with Eleven Labs TTS?
# general-englishd
de.hungriest
01/07/2025, 11:57 AMfor example is there any way to use realtime api from openai, but only speech-to-text and use the tts from 11labs because their voices are a lot better? i need to know what the best available options are right now for cold calling
a
AiTitus :)
01/07/2025, 2:43 PMwatching
v
voxelholic
01/07/2025, 3:06 PMI belive you can just set the session modality to ["text"] as that specifies what the server responds with. That doesn't stop. you from sending client audio frames. Have you tried it?
d
de.hungriest
01/07/2025, 3:14 PMi did this with python scripts without vapi, but i wonder if its possible with vapi also because they state that when you use realtime api on vapi, then most of the other functions are not usable anymore
v
voxelholic
01/07/2025, 3:15 PMoh I see
d
de.hungriest
01/07/2025, 3:15 PMbut still i dont need a "perfect" solution right now. it would also suffice to at least know within the current way of "STT->reasoning model->TTS" what the fastest combination is
de.hungriest
01/07/2025, 3:16 PMfor example nova2->groq->11labs or whatever. somebody know the best api latencies? because i remember with openai to have at LEAST 1 second for every api call
a
AiTitus :)
01/07/2025, 3:17 PMeleven labs has decreased latency a lot with their eleven_flash_v2_5 model recently
v
voxelholic
01/07/2025, 3:17 PMI noticed that too.
a
AiTitus :)
01/07/2025, 3:18 PMand for STT english only my fav is assemly ai
AiTitus :)
01/07/2025, 3:18 PMcant wait until they support more languages in realtime
AiTitus :)
01/07/2025, 3:18 PMwill def switch from Deepgram at that point
d
de.hungriest
01/07/2025, 3:18 PMah okay interesting!
de.hungriest
01/07/2025, 3:19 PMyou guys also using for cold calling or more something like a receptionist?
a
AiTitus :)
01/07/2025, 3:19 PMmore for customer support, ticketing, etc.
d
de.hungriest
01/07/2025, 3:24 PMcool, I'll have a glance through it
de.hungriest
01/07/2025, 3:24 PMokay then for your use case at least its not the most important to be as fast as possible
a
AiTitus :)
01/07/2025, 3:25 PMwell people expect a quick response when talking to customer service so honestly it has to be as quick as possible 😛
a
Alejo
01/13/2025, 5:24 PMIf you're looking for speed, definitely:
Deepgram -> GPT4o-mini -> ElevenLabs Flash. It's about 700ms.
Other than that, Realtime API is the winner for me.
5 Views