Open tremblap opened 3 years ago
jamesb93
commented
3 years ago Hi PA,
It looks like fftsettings for the mfcc() function are not able to be sanitised because it expects to get 3 values as a list and it cannot index the last value.
Can you post the script here for reference?
tremblap
commented
3 years ago This is fun, since I am not changing them at all...
from ftis.world import World
from ftis.corpus import Corpus
from ftis.analyser.meta import ClusteredSegmentation
from ftis.analyser.slicing import FluidNoveltyslice
from ftis.adapter.reaper import render_tracks
from ftis.common.io import get_sr
from pathlib import Path
c = Corpus("/Volumes/machins/projets/newsfeed/sons/textes/Audio/synth/fromtexttospeech-AmE-George.wav")
nov = FluidNoveltyslice(threshold=0.05, minslicelength=1, fftsettings=[1024,128,1024], filtersize=5, kernelsize=29)
# if we dont set the minslicelength to 2 then we might generate slices too short for the analysis
seg = ClusteredSegmentation()
c >> nov >> seg
w = World(sink="~/Desktop/clustered_segmentation")
w.build(c)
if __name__ == "__main__":
w.run()
tracks = {}
for media, slices in seg.output.items():
pos = 0
sr = get_sr(media)
items = []
for start, end in zip(slices, slices[1:]):
start /= sr
end /= sr
item = {
"file" : media,
"length" : end - start,
"start" : start,
"position" : pos
}
pos += end - start
items.append(item)
tracks[media] = items
render_tracks(
Path("~/Desktop/clustered_segmentation/clustered_slices.rpp").expanduser(),
data = tracks
)Okay, I do remember we edited the FTIS class ClusteredSegmentation() though so could you dig into that class definition and see what the internal MFCC fftsettings are?
tremblap
commented
3 years ago I deleted the edits, this is a new env and a new pull. So the internals are as is in the commit:
fftsettings=[1024, -1 -1]
jamesb93
commented
3 years ago Gotchya, I shall investigate and report back
jamesb93
commented
3 years ago It is a missing comma. Fixed (I think). Can you test? dcd9f1f168dc3844fdf7dd612ffa54ddd5b88d17
tremblap
commented
3 years ago not working despite the comma, but a new bug:
Traceback (most recent call last):
File "clustered_segmentation.py", line 21, in <module>
w.run()
File "/Users/pa/Documents/documents@hudd/research/projects/fluid corpus navigation/research/ftis/ftis/ftis/world.py", line 109, in run
c.walk_chain()
File "/Users/pa/Documents/documents@hudd/research/projects/fluid corpus navigation/research/ftis/ftis/ftis/corpus.py", line 47, in walk_chain
forward_connection.walk_chain()
File "/Users/pa/Documents/documents@hudd/research/projects/fluid corpus navigation/research/ftis/ftis/ftis/common/analyser.py", line 167, in walk_chain
forward_connection.walk_chain()
File "/Users/pa/Documents/documents@hudd/research/projects/fluid corpus navigation/research/ftis/ftis/ftis/common/analyser.py", line 151, in walk_chain
self.run()
File "/Users/pa/Documents/documents@hudd/research/projects/fluid corpus navigation/research/ftis/ftis/ftis/analyser/meta.py", line 144, in run
singleproc(self.name, self.analyse, self.input)
File "/Users/pa/Documents/documents@hudd/research/projects/fluid corpus navigation/research/ftis/ftis/ftis/common/proc.py", line 28, in singleproc
process(x)
File "/Users/pa/Documents/documents@hudd/research/projects/fluid corpus navigation/research/ftis/ftis/ftis/analyser/meta.py", line 123, in analyse
stats = get_buffer(fluid.stats(mfccs, numderivs=self.numderivs), "numpy")
AttributeError: 'ClusteredSegmentation' object has no attribute 'numderivs'but I solved it, line 91
jamesb93
commented
3 years ago Okay, thanks. That is also pushed to: 7fe83d124a1ca9b3748002e8c8258105c1398e3f
As you can see these classes are not great for extending 👎 and their interface sucks quite a lot but thanks for sticking with it
commit f2945963d
trying it with a valid file, works for the FluidNoveltySlice but then aborts: