#@Blus any news on this by any chance?

1 messages · Page 1 of 1 (latest)

gleaming river
#

Hey can you check what codec telenyx is using ? Telnyx typically uses G.711 (PCMU/PCMA) at 8kHz for SIP trunking

wintry owl
#

Hey @gleaming river, thanks again for jumping in.
We’ve tested pretty much everything — G.711 U/A, G.722, different Telnyx regions, TLS + SRTP, 8 kHz and 16 kHz settings on the agent side (PCM 16000 Hz recommended) — and calls do go through fine, but the audio still sounds metallic/compressed.

Since our whole startup relies on this voice flow for the Italian market, it’s really important for us to get it working properly.
Do you happen to know if any SIP providers work better with ElevenLabs in Europe/Italy (since Twilio doesn’t have Italian numbers)?

Also — do you think switching to a WebSocket/Realtime setup could solve this faster instead of trying to fix it over SIP?

gleaming river
#

i think Telenyx offers Opus codec for SIP ensuring high quality, but if thats not the option

WebSocket/WebRTC is the right choice
Better quality (48kHz Opus)
Lower latency
More reliable

wintry owl
#

We’ve confirmed Telnyx SIP does support Opus, but ElevenLabs SIP trunking doesn’t seem to decode it — the call doesn’t go through or remains silent.
So it looks like WebSocket / WebRTC is the only realistic path for HD-quality, low-latency audio. Would you confirm or there are other alternatives with Elevenlabs?

gleaming river
#

Yeap WebSocket/WebRTC is the best option in this case cause it provides HD audio (48kHz) you can also use native SDKs

wintry owl
#

Thanks, @gleaming river — that makes total sense, and we’re now testing exactly that setup via a WebSocket bridge (Telnyx → Supabase Edge Function → ElevenLabs Conversational AI).

Everything connects properly, Telnyx opens the stream, ElevenLabs acknowledges the session update, but we’re still not getting any output audio from the agent.

We’re using μ-law 8kHz for telephony (both input/output) and the agent already has a “First message” defined, but nothing plays.

Do you happen to know if the agent needs an explicit trigger (like conversation_initiation or another event) before it starts generating audio on the WS connection?

Or maybe the WS endpoint expects raw PCM/Opus instead of ÎĽ-law?

Just want to make sure we’re framing the audio correctly before diving deeper. 🙏

gleaming river
#

Hey, no explicit conversation_initiation event is needed - the first message plays automatically after the WebSocket connection is established and the agent is initialized

#

Can you check your Agent Audio Format Configuration ?