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

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

Fix: Clean text conditioning in generate() matching forward_train exactly (eliminating random noise injection)

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

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

Fix: Vocos 80-to-100 mel channel adaptation in NeuralVocoder to prevent robotic buzz fallback

Master SOTA Upgrade: in-context voice conditioning, decoupled studio vocoder, EMA tracking, warmup scheduler

Update: training fixes, LICENSE + requirements + scripts

Initial commit: ViuAI_TTS_200M architecture, prosody engine, and weights