2 ms·It's the real recording the model generates the voice with.by shellfishgene 4y agoIt's the real recording the model generates the voice with.