Junchuan Zhao

Achieving expressive TTS now hinges on effectively optimizing non-verbal vocalizations, with design choices impacting NV fidelity more than previously understood.

Achieve more robust and higher-quality speech synthesis from neural codec language models without any retraining, simply by pruning unnatural token sequences during inference.

Achieve high-fidelity bandwidth extension by operating directly in the latent space of neural audio codecs, sidestepping the computational costs and fidelity limitations of spectrogram or waveform-based methods.