AuK: Open-Source Model for Generating and Editing Speech

Melvin Vivas · X video post · 2026-09-18 · 1:53 · 277 views · Open on X

Topics: Industry Trends & Job Market, AI Dev Tools & Productivity · Level: beginner

Summary

Melvin Vivas shares a demo of AuK, a new open-source foundation model that both generates and edits speech. It takes natural-language instructions plus reference audio through a single interface, and the quoted post calls it "Nano Banana for audio." The video is a reel of audio samples with no spoken explanation. It is mostly a model announcement, not a tutorial.

Key points

• AuK is described as an open-source foundation model for both speech generation and speech editing.
• Input: natural-language instructions plus a reference audio clip, all through one interface.
• Zero-shot TTS: it clones a voice from reference audio with no training on that speaker.
• Instruction-controlled generation: you control style or delivery with text instructions.
• Content editing: you can change words in existing speech. The demo plays original and edited clips in pairs (0:39-1:15).
• Whisper conversion: it turns normal speech into whispered speech. This is a voice style, not OpenAI's Whisper model.
• The quoted post calls it 'Nano Banana for audio', meaning instruction-based editing like Google's image model, but for speech.

Resources mentioned

• AuK · tool · github.com · free
  Open-source foundation model for unified speech generation and editing (zero-shot TTS, instruction control, content editing, whisper conversion).
  Also in: AuK: Tencent Hunyuan's Unified Audio Generation and Editing Model (TTS Demo) (Melvin Vivas on X · notes)
• Nano Banana (Gemini 2.5 Flash Image) · tool · developers.googleblog.com · free
  Google's instruction-based image generation and editing model, used here as a comparison point for AuK.