Tencent Hunyuan Open-Sources AuK, a Unified Speech Generation and Editing Foundation Model

On September 9–10, Tencent's Hunyuan team released AuK, an open-source 1.5B-parameter speech foundation model that merges speech generation and editing into a single framework driven by natural language instructions plus audio context. The weights are fully open, and the technical report is on Hugging Face, making the model reproducible. The team positions it as the "Nano Banana of audio"—editing speech with one-line instructions the way images are edited today.

Confirmed

• AuK is built on a multimodal language model backbone with a joint VAE and a hybrid rectified-flow Transformer.
• It covers five task families: speech generation (zero-shot TTS, voice cloning, instruction-controlled generation), content editing (word replacement, insertion, deletion), enhancement and separation, paralinguistic editing (emotion, pitch, speaking rate, volume, accent removal, speech↔whisper conversion), and acoustic editing (background noise repair, environment sound replacement while preserving speaker timbre), plus Whisper-style transcription conversion.
• A distilled variant, AuK-Flash, runs 4.5x faster.

Why it matters

• AuK replaces a full stack of separate speech tools with one model—a framing highlighted by both @realmrfakename and @solyarisoftware—cutting complexity and integration cost.
• Natural language as the unified interface lets non-experts perform fine-grained speech content and style editing, pushing the field from specialized tools toward general foundation models.
• Fully open weights and report enable community reproduction and secondary development.

2026-09-09 ~ 2026-09-10 · 6 related posts

• [source] Tencent Hunyuan open-sources AuK, a foundational model unifying speech generation and editing — Tencent-Hunyuan · 2026-09-09
• Tencent releases AuK, a unified speech generation and editing model with a 4.5x faster Flash version — _akhaliq · 2026-09-10
• [source] AuK: an open-source foundational model unifying speech generation and editing via natural language — _akhaliq · 2026-09-10
• Tencent open-sources 1.5B AuK audio model that replaces a whole stack of audio tools — solyarisoftware · 2026-09-10
• [source] Tencent releases AuK, an open-source audio foundation model unifying speech generation and editing — TencentHunyuan · 2026-09-10

1 near-duplicate retellings: realmrfakename