🎙️AuK turns natural-language instructions into generated, edited, restored, or separated speech with one 1.5B model. 📜 MIT License. 🤖 📄 🏆 AuK delivers leading results on zero-shot and instruction-controlled speech generation, plus general speech editing benchmarks including MMAE-Speech and SpeechEditBench. 🗣️ Clone or design a voice, rewrite speech or lyrics, change emotion, timbre, accent, pitch, speed, and volume, or add and remove laughs, breaths, and coughs. 🎧 It also denoises and dereverberates speech, separates speakers or vocals, and extracts a target speaker from a mixture. All tasks use the same instruction interface. 🧠 Training spans 3.03B instruction-audio pairs and 1.95M hours of effective supervision.