Mirrors
/
riffusion-inference
mirror of https://github.com/riffusion/riffusion.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Disable compression by default, too slow 

Topic: disable_compression
This commit is contained in:
Hayk Martiros 2022-12-27 15:44:39 +00:00
parent 266af6b3e0
commit f7288f8cd3
4 changed files with 23 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
README.md
View File

@ -108,9 +108,9 @@ Execute:
python -m riffusion.cli image-to-audio --image spectrogram_image.png --audio clip.wav python -m riffusion.cli image-to-audio --image spectrogram_image.png --audio clip.wav
``` ```
## Streamlit playground ## Riffusion Playground
Riffusion also has a streamlit app for interactive use and exploration. Riffusion also has a [streamlit](https://streamlit.io/) app for interactive use and exploration.
This app is called the Riffusion Playground. This app is called the Riffusion Playground.
Run with: Run with:

3
riffusion/server.py
View File

@ -155,8 +155,9 @@ def compute_request(
) )
# Reconstruct audio from the image # Reconstruct audio from the image
# TODO(hayk): It may help performance to cache this object # TODO(hayk): It may help performance a bit to cache this object
converter = SpectrogramImageConverter(params=params, device=str(pipeline.device)) converter = SpectrogramImageConverter(params=params, device=str(pipeline.device))
segment = converter.audio_from_spectrogram_image( segment = converter.audio_from_spectrogram_image(
image, image,
apply_filters=True, apply_filters=True,

5
riffusion/spectrogram_converter.py
View File

@ -155,7 +155,10 @@ class SpectrogramConverter:
# Optionally apply post-processing filters # Optionally apply post-processing filters
if apply_filters: if apply_filters:
segment = audio_util.apply_filters(segment) segment = audio_util.apply_filters(
segment,
compression=False,
)
return segment return segment

28
riffusion/util/audio_util.py
View File

@ -32,7 +32,7 @@ def audio_from_waveform(
return pydub.AudioSegment.from_wav(wav_bytes) return pydub.AudioSegment.from_wav(wav_bytes)
def apply_filters(segment: pydub.AudioSegment) -> pydub.AudioSegment: def apply_filters(segment: pydub.AudioSegment, compression: bool = False) -> pydub.AudioSegment:
""" """
Apply post-processing filters to the audio segment to compress it and Apply post-processing filters to the audio segment to compress it and
keep at a -10 dBFS level. keep at a -10 dBFS level.
@ -40,20 +40,22 @@ def apply_filters(segment: pydub.AudioSegment) -> pydub.AudioSegment:
# TODO(hayk): Come up with a principled strategy for these filters and experiment end-to-end. # TODO(hayk): Come up with a principled strategy for these filters and experiment end-to-end.
# TODO(hayk): Is this going to make audio unbalanced between sequential clips? # TODO(hayk): Is this going to make audio unbalanced between sequential clips?
segment = pydub.effects.normalize( if compression:
segment, segment = pydub.effects.normalize(
headroom=0.1, segment,
) headroom=0.1,
)
segment = segment.apply_gain(-10 - segment.dBFS) segment = segment.apply_gain(-10 - segment.dBFS)
segment = pydub.effects.compress_dynamic_range( # TODO(hayk): This is quite slow, ~1.7 seconds on a beefy CPU
segment, segment = pydub.effects.compress_dynamic_range(
threshold=-20.0, segment,
ratio=4.0, threshold=-20.0,
attack=5.0, ratio=4.0,
release=50.0, attack=5.0,
) release=50.0,
)
desired_db = -12 desired_db = -12
segment = segment.apply_gain(desired_db - segment.dBFS) segment = segment.apply_gain(desired_db - segment.dBFS)