@Nir Simionovich Really appreciate you taking the time to break this down. Your explanation about the aggregator chain and codec sampling makes total sense. I've been researching local Polish SIP providers based on your recommendations, and I sent you a LinkedIn invite as well. :) Quick clarification question: If I find a local Polish SIP provider that natively supports OPUS or G.722 (16kHz) and offers VoIP-to-VoIP routing, would I still need Cloudonix in the chain? Or does Cloudonix add value beyond just the codec upsampling? From what I understand, the critical requirements for a local provider would be: ✅ VoIP-to-VoIP routing (bypassing PSTN) ✅ Native 16kHz codec support (G.722, OPUS, or AMR-WB) ✅ WebSocket API or SIP trunk with media streaming ✅ Polish DID numbers ✅ Solid API documentation for integration Without points 1-3, building a phone-based AI agent is basically impossible, right? I'm definitely interested in opening a Cloudonix account and learning more about what you can help configure. Just want to make sure I understand the architecture before we dive in. Thanks again for sharing your expertise! I would be really keen to participate in one of the scheduled calls.

Hi @Nir Simionovich , Thank you for your continued support and expertise throughout this integration process—your guidance has been invaluable. I wanted to share some positive findings from my testing with Telnyx and Deepgram Nova-2 that might be helpful for others in the community facing similar challenges. After extensive testing with my engineer, we've successfully implemented a stable setup using Deepgram Nova-2 for STT with PCMA codec at 8 kHz. The results have been excellent: - Accuracy: Deepgram Nova-2 achieves a median WER of 8.4%, with a 30% reduction in word error rate compared to competitors​ - Telephony optimization: Nova-2 is specifically designed to handle telephony-grade audio and performs exceptionally well with 8 kHz sampling​ - Codec switch: We transitioned from PCMU to PCMA with no noticeable latency increase—interaction with the bot remains seamless - Real-world validation: Testing on Polish Telnyx numbers confirms the bot understands speech accurately without requiring 16 kHz For anyone implementing Telnyx integrations and encountering difficulties, I'd recommend trying: 1. Deepgram Nova-2 for STT 2. PCMA codec in both your TeXML BIN setup and code configuration It appears that modern STT technology like Nova-2 can transport voice with full clarity to the LLM at 8 kHz, which is remarkable. For our current implementation, switching to 16 kHz isn't necessary—though we'll certainly keep it as an option for future optimization. Thanks again for being such a trusted resource in this space!😎

@John George Hi ! I created a mind map based on your YT video some time ago :) Check it out

@John George that's exactly what I do as well! Btw since you are in the business for some time now...Have you managed to successfully connect telephony system (telnyx/twilio) to you AI bot (pipecat)? I opened a new thread for that in Pipecat tab. Would you be so kind to have a look?

Hi @Kwindla Kramer - I am impressed with the work your team deliver.

Hello Nour, we both sharing similar experience. It's nice to be part of this community!