Theme
AI Resources
AuK
AuK is a 1.5B speech model from Tencent Hunyuan that uses natural-language instructions for both generating speech and changing existing audio.
Its public code and weights cover text-to-speech, content and acoustic edits, emotion and timbre changes, speech cleanup, and source separation through command-line, Python, Gradio, and ComfyUI paths. Use this as a first read, not a recommendation. Open the original project before trusting details like terms, limits, privacy, cost, setup, or safety.
What it is
One model for speech generation and editing
AuK accepts an instruction with optional source or reference audio. The task can be new speech, a change to what was said, an acoustic adjustment, a voice-style edit, cleanup, or separation.
Why it stands out
Editing goes beyond text-to-speech
The same interface can replace or remove spoken content, adjust pitch, speed, or volume, change emotion or timbre, add or remove nonverbal sounds, enhance speech, and isolate a speaker or vocal track.
Availability
Base and faster Flash checkpoints
Tencent publishes the code, base weights, a four-step AuK-Flash checkpoint, demos, an inference cookbook, ComfyUI nodes, and a fine-tuning path. The repository identifies the released AuK materials as MIT-licensed.
Why it matters
What makes it useful
Speech projects often need more than a voice reading text. AuK puts generation, targeted edits, delivery changes, cleanup, and separation behind the same instruction format, so a builder can explore several audio jobs without changing to a different model interface each time.
What to know
Where it fits
It fits experiments that need programmatic speech generation or editing rather than a finished consumer voice studio. The repository includes ready-made examples, a local web demo, ComfyUI integration, and training materials for adapting the model to source-target audio pairs.
Notable points
What stands out
The accompanying paper reports results across generation, instruction-guided editing, and signal-level restoration, including a 4.5-times wall-clock speedup for AuK-Flash under its matched test conditions. Those are project-reported evaluations, not a guarantee for a different voice, recording, language, device, or workflow.
Before using
What to review
Plan for separate AuK or AuK-Flash weights, the required Qwen2.5-Omni-3B encoder, Python 3.10, the project's pinned PyTorch line, and a suitable local GPU setup.
Review the provider-declared terms for both the AuK materials and the required Qwen encoder instead of assuming every runtime component shares one license.
Test the exact task on your own recordings. Content edits, voice changes, cleanup, and separation can preserve or alter different details depending on the audio and instruction.
Use reference voices and recordings only in ways that respect the people represented, and keep generated or edited speech from being mistaken for an authentic recording.
Check the optional Prompt Enhancer and speech-recognition settings before supplying private audio: some configurations can call an OpenAI-compatible language model or cloud ASR, while a local ASR fallback is also documented.
Reader fit
Who may find it relevant
Builders exploring one model interface for speech generation, editing, enhancement, and separation.
Creators testing targeted changes to words, delivery, emotion, timbre, or nonverbal sounds in recorded speech.
Researchers who want public checkpoints, inference code, a cookbook, and a fine-tuning path for instruction-guided audio work.
Less relevant for people who need a lightweight CPU model, a finished voice-production app, or a hosted API with no local setup.
Editorial note
Why LifeHubber lists it
LifeHubber lists AuK because it treats speech generation, targeted editing, cleanup, and separation as parts of one instruction-driven model rather than only offering another text-to-speech checkpoint. That helps a reader decide whether a broader audio model is worth the heavier local setup and testing.
Source links
Source materials
Reader note
Before relying on this entry
LifeHubber lists entries to help readers inspect AI projects, not to endorse them or prove they are safe, suitable, accurate, maintained, or right for a specific use. We do not verify every entry in depth. Before relying on anything listed, review the original materials, terms, privacy practices, limits, and risks that matter for your situation.
What to explore next
Compare AuK with other voice and speech tools.
AuK combines generation, targeted edits, cleanup, and separation in one model. Use the wider guide to compare that setup with simpler text-to-speech models, realtime voice agents, transcription tools, and finished apps.
More in Speech Models
Keep browsing this category
Explore more speech model resources.
Fish Audio S2 Pro
fishaudio/s2-pro
A text-to-speech model with detailed control over prosody and emotional delivery.
OmniVoice
k2-fsa/OmniVoice
A multilingual zero-shot text-to-speech model for more than 600 languages, with voice cloning, text-guided voice design, local inference, training and fine-tuning materials, and public demos.
Kokoro-82M
hexgrad/Kokoro-82M
A compact 82M-parameter text-to-speech model from hexgrad, with model facts, usage examples, voice materials, samples, a demo Space, and a linked GitHub inference library.
For project maintainers
Listed here? You can use the badge.
If you maintain a project with a current LifeHubber listing, you may add the optional “Listed on LifeHubber AI Resources” badge to its README, docs, or website. No introduction or permission request is needed.
