Scale of Intelligibility Score

[fakerybakery]

Hi,
What's the scale of Intelligibility Score? 1-1000? 1-10000? 1-100? Is higher or lower better?
Thank you!

[fakerybakery]

Right now, I'm getting varied results ranging from 1000 to 10,000

[CamellIyquitous]

Hello, do you know what the reference value range should be? I have also obtained some values ​​around 5000 for CLVP FD, but I don't know whether this is large or small?

from tts_scores.clvp import CLVPMetric
cv_metric = CLVPMetric(device='cuda')
score = cv_metric.compute_fd('<generated_audio_path>', '<real_audio_path>')

---

In addition, I have another question: For TTS synthesis tasks, the length of the synthesized speech is generally not the same as the length of the ground truth speech. In this case, should I directly calculate the CLVP FD of the two audio sequences? Or do I need to do DTW alignment first, get the two aligned audio sequences, and then calculate the CLVP FD?