OzTianlu/A_Reasoning_Critique_of_Diffusion_Models
Viewer ā¢ Updated ā¢ 6 ā¢ 135 ā¢ 12
šļø Building on HF
Zixi "Oz" Li
OzTianlu
AI & ML interests
My research focuses on deep reasoning with small language models, Transformer architecture innovation, and knowledge distillation for efficient alignment and transfer.
Recent Activity
reacted to theirpost with š 27 days ago
zixi-51M: Not a stock voice. A voice. š¾
https://huggingface.co/NoesisLab/zixi-51m
Everyone is shipping 500B-parameter models. I shipped 51 million. With one voice. On purpose. š¤
What: open-weights Chinese TTS. VITS2 lineage. 51.43M params. Runs on your laptop CPU ā no cloud, no API key, no GPU bill, no meter running.šø
The voice: not a stock voice, not a 5,000-speaker average. One real human's voice ā mine. Timbre, rhythm, breath included. The sample on themodel card is one take, default parameters, recorded by absolutely no one. š§
What's NOT inside:
ā¢ ā BERT (zero. the text encoder vibes on phones + tones alone)
ā¢ ā language routing (Chinese-exclusive: 77 symbols, 6 tones, one job)
ā¢ ā my training data (the corpus is my own writing and stays mine ā open weights, closed corpus, that's the deal š¤)
The trick: embedding transplant. Phone, tone and speaker rows inherited from the MeloTTS-Chinese base by symbol name ā no random init. Themodel woke up already knowing Chinese. š§
Built solo, corpus to packaging, by a writer who is also a violinist. Prosody is my day job. š»
Use it, fork it, fine-tune your own voice. Just don't put words in a real person's mouth. š
š Weights: https://huggingface.co/NoesisLab/zixi-51m
š» Code: https://github.com/lizixi-0x2F/zixi-51M posted an update 27 days ago
zixi-51M: Not a stock voice. A voice. š¾
https://huggingface.co/NoesisLab/zixi-51m
Everyone is shipping 500B-parameter models. I shipped 51 million. With one voice. On purpose. š¤
What: open-weights Chinese TTS. VITS2 lineage. 51.43M params. Runs on your laptop CPU ā no cloud, no API key, no GPU bill, no meter running.šø
The voice: not a stock voice, not a 5,000-speaker average. One real human's voice ā mine. Timbre, rhythm, breath included. The sample on themodel card is one take, default parameters, recorded by absolutely no one. š§
What's NOT inside:
ā¢ ā BERT (zero. the text encoder vibes on phones + tones alone)
ā¢ ā language routing (Chinese-exclusive: 77 symbols, 6 tones, one job)
ā¢ ā my training data (the corpus is my own writing and stays mine ā open weights, closed corpus, that's the deal š¤)
The trick: embedding transplant. Phone, tone and speaker rows inherited from the MeloTTS-Chinese base by symbol name ā no random init. Themodel woke up already knowing Chinese. š§
Built solo, corpus to packaging, by a writer who is also a violinist. Prosody is my day job. š»
Use it, fork it, fine-tune your own voice. Just don't put words in a real person's mouth. š
š Weights: https://huggingface.co/NoesisLab/zixi-51m
š» Code: https://github.com/lizixi-0x2F/zixi-51M updated a model 27 days ago
NoesisLab/zixi-51mOrganizations
