XTTS Fix: conditioning length float error fix

[SinanAkkoyun]

When generating the speaker reference latent and supplying a floating point conditioning length, get_gpt_cond_latents throws an error, as when it calculates the samples it needs, it does not cast to an int.

for i in range(0, audio.shape[1], 22050 * chunk_length):

->

for i in range(0, audio.shape[1], int(22050 * chunk_length)):

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}