Architecture for Real-Time Call Transcription

[JonasWhispa]

Hi everyone,

I need to perform real-time audio transcription on live 3CX calls.

I've analyzed the Call Control API, and here is my current understanding:

• GET /stream Endpoint: Only works if the call is anchored at a Route Point (e.g., an AI answering bot). It seems useless for transcribing calls once they are routed to an actual human agent because the stream is lost.

This leaves me with two main architectural options to capture live agent calls:

1. Silent Auditor Pattern: Build a SIP Bot, register it as an extension, and use the Call Control API to BargeIn on active calls to capture the RTP stream.
2. WebRTC Emulation: Reverse-engineer the Web Client protocol to "listen" via the browser API (seems fragile).

Questions: Has anyone successfully implemented any of the patterns recently or found another way to capture the live audio stream? This is an urgent issue for us and we are happy to hire anyone who is able to help us with that

Thanks!

 

[N_G]

Why would you do that? 3CX Enterprise edition includes an AI agent which allows you to use real transcription of OpenAI and prompt it as you like.

 

[JonasWhispa]

I don't want to have an AI agent answer the call. We are building a software that analyzes a call in real-time giving the human operator guidance, sentiment analysis and relevant customer information in real time. In order to do so I need to access the audio stream between the caller and and the agent in real-time.

 

[N_G]

I see, makes sense, I believe we have this in our architecture I will revert......

 

[JonasWhispa]

Thanks Nick, really appreciate the quick response.

Happy to wait for your update.

 

[JonasWhispa]

hi @N_G,

just checking in to see if you’ve had a chance to look into whether this is technically possible.

 

[JonasWhispa]

Hi @N_G,

just wanted to follow up once more. Do you have any updates whether this is possible to acheive?

 

[KyriacosS_3CX]

Hi @N_G,

just wanted to follow up once more. Do you have any updates whether this is possible to acheive?

Hello,

All new features are announced in our Blogs and general communication channels. Please keep monitoring those.

 

[TODD CONN]

Any progress on this?

 

[N_G]

We are working on this...

 

[N_G]

Hi @N_G,

just wanted to follow up once more. Do you have any updates whether this is possible to acheive?

Give us a little more time we are working on it

 

[erikg]

We need simple "WebSocket Audio Streaming" to ws:// address

 

[Elil_TeleProTech]

Hai, Any one got any work around above solutions?
If customer deployed On-premises 3CX and On-premise GPU (NVIDIA DGX Spark Founder's Edition).
If able to live stream transcriped and give guidence to the agent live guidence to response. off course there will be AI seperatly anylze these transcription and seach own customer specifc knowedge base and give suggestion to customer.

Let me know any one work around it.

 

[JonasWhispa]

I've built a desktop app which is recording the system audio. I listen to the /callcontrol/ws endpoint and whenever a new call event comes in the desktop app of that agent starts recording the call.

It is a workaround I'm not too happy with but now it is working pretty smooth.

Issues that still exist are that there are no events for hold events / muting of the of the microphone.

Still waiting for a proper solution from 3CX though.

 

[Samfiller]

Hey all,

Was any solution for native streams found?

Thanks!!

 

[podtris]

Is there any update on this please @N_G ? Would appreciate advice - thank you!

 

[podtris]

@N_G sorry to chase, would really appreciate an update on this one when you get chance though please! Many thanks

 

[N_G]

This is something we are working on. It will take some time though, for sure several months possibly more. We can not give an exact time.....

 

[podtris]

This is something we are working on. It will take some time though, for sure several months possibly more. We can not give an exact time.....

OK, thanks for coming back to me, I will see if we can build a workaround in the meantime. Please do let us know on this chat when fixed if you can please - thank you!