Wenming Tu

[Wenming Tu @Kevin_Tu0604]

Edit audio like you edit text: remove applause, keep the speaker; swap a lyric, keep the singer.

The @ieeeICASSP 2027 Audio Editing Challenge is open. 2 tracks, $7,000 (Tencent), open-source baselines.

Details & register 👇
https://t.co/3ateZWvDed
#ICASSP2027 #AudioEditing

[Wenming Tu @Kevin_Tu0604]

Key dates (deadlines 23:59 US Pacific):
Oct 1: Challenge starts
Nov 10: Data + leaderboard
Nov 25: Submission deadline
Dec 8: Final results

Register: https://t.co/9D1PlrHG2l
Slack: https://t.co/liqqslGKGJ
#ICASSP2027 #AudioEditing

[Wenming Tu @Kevin_Tu0604]

Edit audio like you edit text: remove applause, keep the speaker; swap a lyric, keep the singer.

The @ieeeICASSP 2027 Audio Editing Challenge is open. 2 tracks, $7,000 (Tencent), open-source baselines.

Details & register 👇
https://t.co/3ateZWvDed
#ICASSP2027 #AudioEditing

[Wenming Tu @Kevin_Tu0604]

$7,000 in prizes, sponsored by Tencent, for the top 3 teams in each track.

Top teams are invited to submit a 2-page ICASSP 2027 paper; on-site presenters can extend it for the IEEE OJ-SP journal.

[Wenming Tu @Kevin_Tu0604]

Both tracks ship with open-source baselines so you can start fast:
🔹 Single Model: AuK-based, end-to-end
🔹 Agent: an LLM calling audio tools

Baselines: https://t.co/eMPSLCUKFL
AuK: https://t.co/jHSZwGDy6N

[Wenming Tu @Kevin_Tu0604]

General audio editing across speech, music, and environmental sound: understand the audio, follow the instruction, make the target edit, and leave everything else untouched.

Two tracks, pick one or both:
🔹 Single Model (end-to-end)
🔹 Agent (tool-using)

[Wenming Tu @Kevin_Tu0604]

Edit audio like you edit text: remove the applause but keep the speaker, or swap a lyric but keep the singer's voice.

The @ieeeICASSP 2027 Audio Editing Challenge is open for registration. 🧵
🌐 https://t.co/3ateZWvDed

[Wenming Tu @Kevin_Tu0604]

@wildmindai Thanks for bringing AuK to ComfyUI! For the emotion/pitch issues you mentioned, our Prompt Enhancer (PE) helps make them much more stable: https://t.co/wIOL7gOnbp

[Wenming Tu @Kevin_Tu0604]

@Marble68 @TeksEdge Here: https://t.co/rgb30n0e3C

[Andrew Ng @AndrewYNg]

With AI Engineering skills, you actively shape the build: You influence what gets built, and drive the build loop. Here're key skills to do this. https://t.co/sysOYdzuZY

[Wenming Tu @Kevin_Tu0604]

@Marble68 @TeksEdge You can try it at Huggingface Space 🤗:
https://t.co/rKUoQQHR6G

[Guitar Cat + LLM @GenAI_is_real]

SGLang Omni 全天候完成对混元 AuK 模型的强烈支持！！！

完成了对 AuK 模型的 day 0 支持，欢迎大家使用～clone 最新的 commit，然后：

python -m sglang_omni.cli serve --model-path tencent/AuK

坦诚说这个目前做的还是比较粗糙，接下来会有一些时间专门来优化下 auk，也欢迎大家自己 profile，自己提出想法，来做优化

相关的 roadmap 在 issue 2064～

https://t.co/vse3jikgJK

https://t.co/v8HBJtuqW3

https://t.co/QspqutmcBW

[Wenming Tu @Kevin_Tu0604]

@GenAI_is_real 🥳 Cool !!!

[Wenming Tu @Kevin_Tu0604]

@ivanfioravanti @TencentHunyuan Thanks! 🙌 You can try it out here:
🤗 HF Space: https://t.co/rgb30n0e3C
Let us know what you think!

[Wenming Tu @Kevin_Tu0604]

@nrol_ling @TencentHunyuan Good question!🤩
Give it a try and let us know how it does. 👇
https://t.co/rgb30n0e3C

[Wenming Tu @Kevin_Tu0604]

@Jaidcel @TencentHunyuan Yes, some capabilities can still be a bit unstable. When that happens, trying a different seed may give you a much better result. Enabling Prompt Enhancer (PE) usually makes the generation more stable as well. We’ll keep gathering feedback and improving the model.❤️❤️

[Wenming Tu @Kevin_Tu0604]

@Jaidcel @TencentHunyuan Cool~!!! 🙌

[Wenming Tu @Kevin_Tu0604]

@_banzaan @TencentHunyuan 🎤🍌 Just try it!!! Have fun!! 😀

[Wenming Tu @Kevin_Tu0604]

@liscat @TencentHunyuan Haha, not copying 😄 We just want to build a “Nano Banana🍌 for audio��� — there isn’t really an equivalent for audio yet.

[Wenming Tu @Kevin_Tu0604]

@Jaidcel @TencentHunyuan we have provided many examples at the bottom, just try it!!! Click one example and then click 'generate' button. Have fun!!! 🥹