Yongyi Zang
@yongyi_zang
Followers
132
Following
18
Statuses
29
Research Scientist @ Smule. Most works are done as an independent researcher, opinions are my own.
Redmond, WA
Joined December 2013
Okay, so if the lightning to HDMI adapter can accept at maximum a 256 MB runtime with 400 MHz clock, this is actually good for a bunch of neural codecs and speech enhancement stuff right?
Haywire itself doesn’t store any firmware (well, except for SecureROM), so iOS has to upload it every time. Firmware bundle for it is very tiny, around 25MB uncompressed. Shipped as preinstalled asset with iOS and/or can be downloaded. That’s how its folder structure looks like:
0
0
1
RT @RuiboLiu: my mum tried DeepSeek and told me she loved it because it could generate beautiful literature-level Chinese. she has zero i…
0
16
0
Glad to have contributed to such an ambitious project. Happy lunar new year! (Possibly) some follow up works to come (soon?)
1/n: 🚀 Announcing YuE (乐) – the most powerful open-source full-song music generation model! 🎵 Tackle the lyrics-to-song task (like with support for diverse genres, stunning vocals, & multiple languages. Bonus? It’s Hugging Face & LLAMA-compatible for easy fine-tuning. 🛠️ Code: Demo:
0
0
2
It's ICASSP season again and I have one collab work with @DichuchengLi and @QiuqiangK on Piano Transcription! We found that using a pre-trained frame-wise encoder trained and separate decoding stages for onset, offset and velocity helps significantly with performance. #ICASSP2025
0
0
2
🤯
New Anthropic research: Alignment faking in large language models. In a series of experiments with Redwood Research, we found that Claude often pretends to have different views during training, while actually maintaining its original preferences.
0
0
0
RT @YapengTian: Very offensive! After seeing her apology, I’m not convinced it was sincere. I’m also extremely sad that the conference itse…
0
12
0
RT @xiaolonw: Very sad to see this. Just reminds me some random dude in the bay drove by and shouted to me “go back to your own country”. R…
0
35
0
RT @ZhiyuChen4: I'm shocked to see racism happening in academia again, at the best AI conference @NeurIPSConf. Targeting specific ethnic gr…
0
296
0
RT @sunjiao123sun_: Mitigating racial bias from LLMs is a lot easier than removing it from humans! Can’t believe this happened at the bes…
0
905
0
RT @Yixiao_Zhang_: We have a paper - it discusses the interpretation gap between human and music generation models. @yongyi_zang will prese…
0
2
0
😅
#icassp2025 new template is a mess. The default font of latex template is 10pt yet 9pt is acceptable. @ieeeICASSP also allows the author sec similar to past years. This means, 9pt+old author sec can give a whole extra page compared to 10pt+new author sec if there are many authors
0
0
0
RT @jiatongshi: 🎉 Excited to announce that the SVDD2024 challenge is now online! 🚀 Check out the details here: W…
0
2
0
The WildSVDD track will be part of the MIREX Challenge this year ( with the amazing @yzyouzhang and @Yixiao_Zhang_ as task captions. If you’re interested, stay tuned for more details! (2/2)
Singing Voice Deepfake Detection (SVDD) challenge summary paper is out: Thanks to all 47 participating teams! We have a lowest EER of 1.65%, while generalization to an unseen commercial set remains challenging (almost all teams at ~50%). (1/2)
0
2
4
Singing Voice Deepfake Detection (SVDD) challenge summary paper is out: Thanks to all 47 participating teams! We have a lowest EER of 1.65%, while generalization to an unseen commercial set remains challenging (almost all teams at ~50%). (1/2)
0
3
10
RT @nicolaus625: 📢📢Glad to share our latest survey on Foundation Models for Music! 🎶For both computer music researchers and ML researchers:…
0
15
0
RT @sahar_abdelnabi: NeurIPS reviewer: solution is not new and was heavily studied. @NeurIPSConf #NeurIPS24
0
63
0
RT @jxmnop: if you have language model weights on your computer, you also are in possession of powerful compression software just put the…
0
58
0