33.1 GB

Ctrl+K

1 contributor

History: 94 commits

ViuAI

Fix: Peak audio normalization (-1 dB), robust speaker search, and tuned CFG for crystal-clear benchmark generation

d62a3fe verified 7 days ago

checkpoints
Upload checkpoints/viuai_tts_latest.pt with huggingface_hub 7 days ago
config
Update: training fixes, LICENSE + requirements + scripts 9 days ago
dataset
Turbo Speed: Vectorized FFT F0 pitch, cuDNN autotune, Tensor Core precision, speaker RAM cache & prefetch DataLoader 8 days ago
eval_benchmarks
Upload eval_benchmarks/epoch_60/02_hindi_female.wav with huggingface_hub 7 days ago
models
Fix: Peak audio normalization (-1 dB), robust speaker search, and tuned CFG for crystal-clear benchmark generation 7 days ago
scripts
Master Fix: Real F0 pitch, acoustic Mel-80 to 100 filterbank vocoder projection, soundfile multi-format loader, safe weights_only 8 days ago
.gitattributes

4.92 kB
Fix: Peak audio normalization (-1 dB), robust speaker search, and tuned CFG for crystal-clear benchmark generation 7 days ago
LICENSE

747 Bytes
Update: training fixes, LICENSE + requirements + scripts 9 days ago
README.md

4.09 kB
Initial commit: ViuAI_TTS_200M architecture, prosody engine, and weights 10 days ago
ROADMAP.md

2.32 kB
Initial commit: ViuAI_TTS_200M architecture, prosody engine, and weights 10 days ago
inference.py

9.18 kB
Fix: Peak audio normalization (-1 dB), robust speaker search, and tuned CFG for crystal-clear benchmark generation 7 days ago
requirements.txt

185 Bytes
Performance: 32 parallel download workers and hf_transfer for 10x faster dataset downloading 8 days ago
test_viuai_tts_kaggle.ipynb

6.89 kB
Fix: Peak audio normalization (-1 dB), robust speaker search, and tuned CFG for crystal-clear benchmark generation 7 days ago
train.py

45.9 kB
Fix: Peak audio normalization (-1 dB), robust speaker search, and tuned CFG for crystal-clear benchmark generation 7 days ago
train_colab.ipynb

3.65 kB
Fix: Peak audio normalization (-1 dB), robust speaker search, and tuned CFG for crystal-clear benchmark generation 7 days ago
training_epoch60_completed_memory.png

279 kB
xet

Fix: Peak audio normalization (-1 dB), robust speaker search, and tuned CFG for crystal-clear benchmark generation 7 days ago
upload_to_hf.py

2.36 kB
Fix: Peak audio normalization (-1 dB), robust speaker search, and tuned CFG for crystal-clear benchmark generation 7 days ago
verify_model.py

4.47 kB
Master Fix: Real F0 pitch, acoustic Mel-80 to 100 filterbank vocoder projection, soundfile multi-format loader, safe weights_only 8 days ago