Hey folks! Looking for your wisdom on the best overall Vapi setup (LLM + Transcriber + TTS/Voice combo) for smaller languages like Slovak, Czech, etc. My dilemma is the classic trade-off: My current "fast & accurate" setup (Deepgram Nova-2 + OpenAI TTS-1 w/ GPT 4o realtime mini Cluster, Shimmer voice) has great low latency and is reliable for specifics like reading numbers corectly. I've also experimented with both GPT-4o Realtiem Mini and the standard GPT-4o Realtime as the LLM in this setup, but honestly didn't perceive a major difference between them for my needs. But using ElevenLabs for that beautiful voice quality adds noticeable latency. What combos are you finding give the best balance between low latency, high reliability/accuracy, AND natural-sounding voice in these languages within Vapi? Side note: I hit validation-failed errors trying to use Anthropic/Claude so I couldn't even properly test its performance. Has anyone had success using Claude models smoothly in voice agents ? I would like to test Claude Sonet 3.7 Really appreciate any shared experiences or recommendations! Thanks!