Voice Quality issue using Librispeech

[Anshu-Kumar-1]

Hello, I'm found difference between the generated audios from the provided demo notebook using Librispeech and the audios available on the web page. The generated audios lack naturalness compared to the ones on the demo; they seem to lack pauses and exhibit an unnatural tone.

I've tried adjusting various parameters, including diffusion steps ([15, 20, 25]), embedding scale ([1, 2]), and altering the audio speed (110%) using duration = model.predictor.duration_proj(x) / speed. Despite these adjustments, the issue persists. Any insights or suggestions on how to improve the generated audios to match the quality of those on the demo page would be greatly appreciated.

Generated Audio: https://drive.google.com/file/d/1xhYaB_1jzabYbkPwt6aP3KSNDkjdwT7X/view?usp=drive_link

Thank you for your time and impressive work!