qwen3-asr-swift brings PersonaPlex duplex voice to Mac

This is a meaningful implementation milestone for on-device voice AI, but it still reads as an early-stage builder stack rather than a production-ready assistant platform.

• –The release shows full-duplex speech-to-speech can run locally on Apple Silicon with sub-real-time-factor performance claims.
• –Reusing shared MLX components across ASR, TTS, and speech-to-speech makes the toolkit unusually cohesive for Swift developers.
• –Community discussion highlights current gaps like tool use, reliability under long conversations, and controllability for real agent workflows.
• –Even with those limits, the open-source Swift-first path lowers experimentation friction for Mac and iOS voice products.