Query about the inference speed

[Loydian]

I want to use this method in a on-the fly training scheme. So I am curious about how long it takes to process an audio. Can you provide me with a reference speed on a specific GPU?

[SarinaMeyer]

This depends on which exact version you want to use. I do not have these statistics for all versions and currently do not have the time to run the necessary experiments. In our multilingual paper, which the code in the main branch corresponds to, we reported the following statistics:

The total system consists of roughly 1.6 billion parameters for
ASR (1.5B), speaker extraction (2M), speaker anonymization
(50k), and speech synthesis (46M). The anonymization of one
10-second utterance takes 2 minutes on CPU ( Intel Xeon, E5-
2687W v4), and 35 seconds on GPU (NVIDIA RTX 3090).

I hope this helps!