← Writing

Series · 3 articles

TTS on OpenVINO

Qwen3-TTS graph export, KV caches, and streaming batch scheduling on Intel hardware.

Trace Qwen3-TTS from graph export on Intel hardware to streaming audio. The series covers model stages, stateful KV caches, quantization boundaries and scheduling without relying on a vLLM runtime.

Read in the order below: start with the constraints, then follow the implementation and debugging details.

In this series

  1. Qwen3-TTS on Intel: OpenVINO inference without CUDA

    14 min read
  2. Qwen3-TTS on OpenVINO: Talker, Subcode and streaming decode

    14 min read
  3. OpenVINO Qwen3-TTS: Paged-KV, U8 cache and continuous batching

    14 min read