We want to train a custom Turkish TTS model with speech quality comparable to ElevenLabs. The model must achieve ultra-low latency suitable for real-time phone call interactions. We have tried open models such as PiperTTS, xTTS-v2, MMS-TTS, Chatterbox, community fine-tuned Orpheus models etc. and have found them to be lacking in either quality or speed or both. We want something that will both be high quality and be serveable with very low latency.