How can I change the duration of the output audio?

[ChloeL19]

Hi, This is really a lovely repository. But how can I change the duration of the generated audio?
Thanks!!

[ChloeL19]

Okay, so I just kind of forced a different latent embedding size. I wanted one second of output, so I divided the original latent dimension (256) by 10 and then rounded up.

def prepare_latents(self, batch_size, inference_scheduler, num_channels_latents, dtype, device):
    # EDIT: they are hardcoding the latent size here!! to 256! I want to change this!
    shape = (batch_size, num_channels_latents, 256, 16)
    shape = (batch_size, num_channels_latents, 26, 16) # scaled to one second???

Indeed, the inference script now outputs audio files that are 1 second in length. Is this....okay??

[ChloeL19]

I suppose duration could be introduced as a training argument, and then saved as part of the training config and used in this way to adjust the lengths of the audio generated during the inference process...

[cvillela]

Would really like an audio sample duration feature as well!