18
YuE
👩
OpenSource Music Generator
Nishith Jain
KingNish
AI & ML interests
AI is fun actually.
Busy till June 2025.
Recent Activity
updated
a Space
23 minutes ago
innova-ai/YuE-music-generator-demo
liked
a model
about 3 hours ago
black-forest-labs/FLUX.1-schnell-onnx
liked
a model
about 3 hours ago
retronic/colox-v1
Organizations
KingNish's activity
updated
a
Space
23 minutes ago
liked
2
models
about 3 hours ago
reacted to
retronic's
post with 🔥
about 3 hours ago
Post
865
Colox, a reasoning AI model. I am currently working on a model smarter than GPT o1 that thinks before it speaks. It is coming tomorrow in the afternoon.
- 1 reply
updated
a
Space
about 3 hours ago
upvoted
a
paper
about 3 hours ago
liked
a
Space
about 3 hours ago
reacted to
hexgrad's
post with 👍
about 4 hours ago
Post
3279
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.
- 1 reply
updated
a
Space
about 5 hours ago
Optimized for speed
1
#7 opened about 5 hours ago
by
KingNish
liked
a
dataset
about 15 hours ago
liked
a
dataset
about 16 hours ago
liked
a
model
1 day ago
liked
a
Space
1 day ago
upvoted
a
paper
2 days ago
liked
a
Space
2 days ago
upvoted
an
article
3 days ago
Article
Open-source DeepResearch – Freeing our search agents
•
701
liked
a
Space
3 days ago
updated
2
collections
4 days ago