Unable to accept concurrent api calls for /tts_to_audio since out.wav will be shared

[IguteChung]

In /tts_to_audio api, if USE_CACHE=False, it will use default file_name_or_path="out.wav" for XTTS.process_tts_to_file.

However it get error if concurrent api calls happen.

torchaudio.save(output_file, torch.tensor(out["wav"]).unsqueeze(0), 24000)

[jdola]

I also get an error when processing 2 or more requests at the same time, I'm trying to fix it.