I am trying to use the DeepFry model with my own dataset.
My data is downsampled to 16k, has TextGrids, and uses the same arpabet phone set.
However, when it is time to write the new TextGrids in add_tier() in utils.py, at the end of the tier there is a difference between the calculated end of the interval (prev_end) and the overall end of textgrid (max_time). In a lot of my recordings, this difference is larger than
the window size (0.005 seconds), easily twice as large. Currently, the code does not adjust the interval end time because the difference is too large, which means the interval can't be written to the textgrid, which doesn't accept intervals with max_time's after the textgrid's max_time, resulting in a ValueError.
Is this something to worry about (e.g. all the boundaries are incorrect or imprecise) or that can be fixed (other than removing the requirement that the difference between prev_end and max_time is less than f_window?
Thanks,
Emelie
Hi,
Could you provide the full error?
Alternatively, could you check if the waveform end time is greater than the textgrids max time? If so, changing the textgrid's max time will prevent the exception.
Hi,
I am trying to use the DeepFry model with my own dataset. My data is downsampled to 16k, has TextGrids, and uses the same arpabet phone set.
However, when it is time to write the new TextGrids in
add_tier()
inutils.py
, at the end of the tier there is a difference between the calculated end of the interval (prev_end
) and the overall end of textgrid (max_time
). In a lot of my recordings, this difference is larger than the window size (0.005 seconds), easily twice as large. Currently, the code does not adjust the interval end time because the difference is too large, which means the interval can't be written to the textgrid, which doesn't accept intervals withmax_time
's after the textgrid'smax_time
, resulting in aValueError
.Is this something to worry about (e.g. all the boundaries are incorrect or imprecise) or that can be fixed (other than removing the requirement that the difference between
prev_end
andmax_time
is less thanf_window
? Thanks, Emelie