PaddlePaddle/PP-DocLayoutV3
Image Segmentation β’ Updated β’ 38.2k β’ 126
Mayowa Daniel
mayowadan
Β·
AI & ML interests
None yet
Recent Activity
liked a model 1 day ago
PaddlePaddle/PP-DocLayoutV3 new activity 9 months ago
deepseek-ai/DeepSeek-OCR-2:Made a Demo for people who want to quickly try the model liked a Space 11 months ago
ResembleAI/Chatterbox-Multilingual-TTSOrganizations
None yet
liked a
model 1 day ago
Made a Demo for people who want to quickly try the model
16
#2 opened 9 months ago
by 
ysong21
liked a
Space 11 months ago
Hallucinations
4
#39 opened about 1 year ago
by 
ntsarb
Why is granite-docling-258M so slow?
π€ 2
35
#37 opened about 1 year ago
by 
hgarp-prozis
liked a
model about 1 year ago
Amazing work!
Where can I find the unminified code for the demo?
reacted to hexgrad's post with β over 1 year ago
Post
4,244
IMHO, being able & willing to defeat CAPTCHA, hCaptcha, or any other reasoning puzzle is a must-have for any Web-Browsing / Computer-Using Agent (WB/CUA).
I realize it subverts the purpose of CAPTCHA, but I do not think you can claim to be building AGI/agents without smoothly passing humanity checks. It would be like getting in a self-driving car that requires human intervention over speed bumps. Claiming AGI or even "somewhat powerful AI" seems hollow if you are halted by a mere CAPTCHA.
I imagine OpenAI's Operator is *able* but *not willing* to defeat CAPTCHA. Like their non-profit status, I expect that policy to evolve over timeβand if not, rival agent-builders will attack that opening to offer a better product.
I realize it subverts the purpose of CAPTCHA, but I do not think you can claim to be building AGI/agents without smoothly passing humanity checks. It would be like getting in a self-driving car that requires human intervention over speed bumps. Claiming AGI or even "somewhat powerful AI" seems hollow if you are halted by a mere CAPTCHA.
I imagine OpenAI's Operator is *able* but *not willing* to defeat CAPTCHA. Like their non-profit status, I expect that policy to evolve over timeβand if not, rival agent-builders will attack that opening to offer a better product.
- 2 replies
reacted to hexgrad's post with π over 1 year ago
Post
6,379
I wrote an article about G2P: https://hf.co/blog/hexgrad/g2p
G2P is an underrated piece of small TTS models, like offensive linemen who do a bunch of work and get no credit.
Instead of relying on explicit G2P, larger speech models implicitly learn this task by eating many thousands of hours of audio data. They often use a 500M+ parameter LLM at the front to predict latent audio tokens over a learned codebook, then decode these tokens into audio.
Kokoro instead relies on G2P preprocessing, is 82M parameters, and thus needs less audio to learn. Because of this, we can cherrypick high fidelity audio for training data, and deliver solid speech for those voices. In turn, this excellent audio quality & lack of background noise helps explain why Kokoro is very competitive in single-voice TTS Arenas.
G2P is an underrated piece of small TTS models, like offensive linemen who do a bunch of work and get no credit.
Instead of relying on explicit G2P, larger speech models implicitly learn this task by eating many thousands of hours of audio data. They often use a 500M+ parameter LLM at the front to predict latent audio tokens over a learned codebook, then decode these tokens into audio.
Kokoro instead relies on G2P preprocessing, is 82M parameters, and thus needs less audio to learn. Because of this, we can cherrypick high fidelity audio for training data, and deliver solid speech for those voices. In turn, this excellent audio quality & lack of background noise helps explain why Kokoro is very competitive in single-voice TTS Arenas.
reacted to Jaward's post with π over 1 year ago
Post
3,526
Awesome intro to LLM course "Language Modeling from Scratch" by stanford. love the aesthetics behind the lecture notes, notes-in-code genius ideaπ
Course site: https://stanford-cs336.github.io/spring2025/
Repo: https://github.com/stanford-cs336/spring2025-lectures
Videos: https://www.youtube.com/playlist?list=PLoROMvodv4rOY23Y0BoGoBGgQ1zmU_MT_
Course site: https://stanford-cs336.github.io/spring2025/
Repo: https://github.com/stanford-cs336/spring2025-lectures
Videos: https://www.youtube.com/playlist?list=PLoROMvodv4rOY23Y0BoGoBGgQ1zmU_MT_
- 2 replies
upvoted 2
articles over 1 year ago
Article
Upgrading Kokoro: natural TTS for short bursts
hexgrad
β’ β’ 33 Article
G2P Shrinks Speech Models
hexgrad
β’ β’ 100 commented on G2P Shrinks Speech Models over 1 year ago
Love this explainer
Way more approachable than an academic paper
Noticed an issue with βliveβ in particular so I look forward to the neural augmentation for homonyms youβve mentioned.
Let me know if youβd like any help with it whatsoever
commented on Upgrading Kokoro: natural TTS for short bursts over 1 year ago
This makes a huge difference.
I noticed kokoro handles like single Roman numerals better than anything OpenAI has to offer.
Thanks so much!
liked a
Space over 1 year ago
liked a
model over 1 year ago
