my DAC dataset again managed to only have some utterances with only 8 of 9 RVQ levels, this fixes an oversight from it
This commit is contained in:
parent
2a1794c084
commit
054d28573a
|
@ -682,7 +682,7 @@ def pad_codes_with_silence( codes, size=1 ):
|
|||
duration = codes.shape[0] * get_framerate()
|
||||
difference = math.ceil( duration + size ) - duration
|
||||
|
||||
silence = get_silence( difference, device=codes.device )
|
||||
silence = get_silence( difference, device=codes.device )[:, :codes.shape[-1]]
|
||||
|
||||
half = math.floor(difference / 2 * get_framerate())
|
||||
|
||||
|
|
Loading…
Reference in New Issue
Block a user