dsh-tts-flash

Reading long AI replies straight off the screen is quietly exhausting. dsh-tts-flash turns every LLM response in DeepSeek Harness Desktop into a speak-as-you-read audio stream: the moment a sentence is generated, it is synthesized and played locally, so your ears catch up before your eyes do. During the brief pause while the model is still thinking, a floating bar cycles through playful catchphrases spoken aloud in sync, replacing dead-air waiting with a small moment of fun.

Under the hood the plugin ships with Microsoft edge-tts out of the box: dozens of voices, zero configuration, works as long as you have an internet connection. When you need a private or higher-fidelity engine, drop a baseURL, API key and model name into the settings panel and any OpenAI-compatible TTS service (OpenAI, Xiaomi MiMo, SiliconFlow, and more) plugs in instantly with dual-protocol auto-detection. Voice speed, volume, subtitle glow, font size, and the draggable floating bar all respond in real time, and every setting survives restarts.

It is built for writers, learners, and developers who spend hours in AI-assisted sessions and want to rest their eyes; and for tinkerers who want a local or cloud TTS layer in DSH without extra accounts or scripts. MIT-licensed, drop in and go.