🚨🚧 FeatureExtractor → AudioProcessor #44394
eustlb
wants to merge
70 commits into
main
from
audio-processor
refacto to introduce ProcessingMixin
4c0db24d
draft
9a0e2848
draft update
2d7c1ca9
keep drafting
8dbd2cab
starting to look like something
57e18b66
update
b069eada
update
aebfee9e
update
2563bf0c
torch equal on audio processor vs feature extractor outputs
486f56b1
update passing test
ae96c82b
_preprocess merged for both backends
90664fc6
frequency_bin_mode and refacto
7cbb70e3
ensure BC + deprecate
c162c962
ensure BC + deprecate
cbaf55bd
update audio processors
4bdf844b
remove test files for another repo
37ad7bbd
add computation_dtype to have matching torch/ numpy implems
017d179b
use 5.5 for deprecation
233e769f
lasr update
cfa48c92
gemma3n update
ec84bc98
temporarily use separate backends files
2573ade8
all tests passing
1147317a
udpates
faf0d554
another round of updates
9eaefc02
some more updates
6fb8b16b
the closert I get the further it gets
e2161aaf
caching window
7ff5c42d
cleaning backend
64dde305
finish merging
c9f30161
finish merging cleanly
e61f94ea
fix
27ca26b7
refacto math helpers
282f1a03
restructure BaseAudioProcessor
a56bef62
tmp refacto
392696a8
test_feature_extraction_xx -> test_audio_processing + EVERY model wit…
0a9ac343
add audio_processing_numpy classes
bc4e5cf7
audio processors for gemma4 and cohere_asr
27cbbadd
audio processors updates
c2ba974b
WIP: audio processor updates before merging main
e0fd4d30
Merge remote-tracking branch 'origin/main' into audio-processor
5aecdc6b
feat(audio): map parakeet_rnnt/tdt + granite_speech_plus to existing …
22769893
feat(audio): add NemotronAsrStreamingAudioProcessor (bit-exact with l…
1fa6ac33
feat(audio): NemotronAsrStreaming legacy FE -> deprecation shim; drop…
c6f30d83
feat(audio): register NemotronAsrStreamingAudioProcessor in auto + init
afb1de6f
feat(audio): add InklingAudioProcessor (parity with legacy FE within …
c7f6bc10
feat(audio): Inkling legacy FE -> deprecation shim
abd18727
feat(audio): register InklingAudioProcessor in auto + init
95cbdca4
chore(audio): drop redundant auto_mappings import + unused FeatureExt…
775d3aa7
fix(audio): let processor/docstring machinery handle AudioProcessor b…
0f96424e
refactor(audio): hoist waveform-level preemphasis into the base pipeline
da7a2d44
chore(audio): remove unused torch_mel_spectrogram / numpy_mel_spectro…
e848c169
update
3fd06bbf
nits
93b39506
mel filters backend
19c07ed7
migrate new models
dbb3143e
factorize image/audio processing
8f7c666e
revert to sampling_rate
451d33d6
revert to sampling_rate
11252caa
nits
4799d1a7
more nits
1617609d
clean
3b105b64
factorize
f238c6d0
factorize more
945ef372
unbloat comment
350d10e2
deprecate FeatureExtractionMixin and SequenceFeatureExtractor
a10a66f7
rmv tests/test_wav2vec2_whisper.py
194f4ed5
rm unused
14494061
fix sample_rate -> sampling_rate typo in backends equivalence tests
bc1bdd90
simplify numpy framing: full pad + stride tricks replaces segmented f…
e465f88f
udpates
d5ad6e8f
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub