Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Does square needed in compute_power? #383

Open
mosheman5 opened this issue Aug 4, 2021 · 1 comment
Open

Does square needed in compute_power? #383

mosheman5 opened this issue Aug 4, 2021 · 1 comment

Comments

@mosheman5
Copy link

The amplitude_to_db moves the rms energy from the amplitude to log scale.
Does the square operation needed in

ddsp/ddsp/spectral_ops.py

Line 338 in 3f90736

power_db = amplitude_to_db(rms_energy**2, use_tf=True)
?
Doesn't multiplying by 20 instead of 10 compensates for it?
@jesseengel jesseengel mentioned this issue Aug 19, 2021
Closed
@jesseengel
Copy link
Contributor

Good catch, thanks! I go more into the plan in #361

copybara-service bot pushed a commit that referenced this issue Aug 20, 2021
@jesseengel
Fix bugs in compute_power and compute_loudness. Fixes #361 and #383.
d190d56 
Breaking Change (-> v2.0.0):
* Perform loudness averaging in linear scale, not dB.
* Remove extra square from compute power.
* Set reference db to 0.0 to align with librosa instead of magic 20.7 number for white noise reference. That way for power, dB=0.0 corresponds to amplitude = 1.0. Correspondingly, set dB range to [-80, 0] instead of [-120, 0] to align with librosa.
* Provide convenience functions for `power_to_db` and `db_to_power` that are aligned to librosa. Clearer docstrings.
* Moved db functions to core with other scaling functions.
* Tests to make sure the db functions match librosa.
* Default to `use_tf=True`.
* Moved diff() to core (not spectral specific)
* Default loudness, power, and rms to much more reasonable frame size default of 512 (instead of 1024/2048) [better time resolution].
* Compute Loudness on the fly by default (it's very fast and doesn't hurt training speed).

Still TODO:
Retrain and swap out models in Colab demo.

PiperOrigin-RevId: 391661045
@copybara-service copybara-service bot mentioned this issue Aug 20, 2021
Merged
copybara-service bot pushed a commit that referenced this issue Jan 14, 2022
@jesseengel
Fix bugs in compute_power and compute_loudness. Fixes #361 and #383.
58ca359 
Breaking Change (-> v2.0.0):
* Perform loudness averaging in linear scale, not dB.
* Remove extra square from compute power.
* Set reference db to 0.0 to align with librosa instead of magic 20.7 number for white noise reference. That way for power, dB=0.0 corresponds to amplitude = 1.0. Correspondingly, set dB range to [-80, 0] instead of [-120, 0] to align with librosa.
* Provide convenience functions for `power_to_db` and `db_to_power` that are aligned to librosa. Clearer docstrings.
* Moved db functions to core with other scaling functions.
* Tests to make sure the db functions match librosa.
* Default to `use_tf=True`.
* Moved diff() to core (not spectral specific)
* Default loudness, power, and rms to much more reasonable frame size default of 512 (instead of 1024/2048) [better time resolution].
* Compute Loudness on the fly by default (it's very fast and doesn't hurt training speed).

Still TODO:
Retrain and swap out models in Colab demo.

PiperOrigin-RevId: 391661045
copybara-service bot pushed a commit that referenced this issue Jan 14, 2022
@jesseengel
Fix bugs in compute_power and compute_loudness. Fixes #361 and #383.
d5552bf 
Breaking Change (-> v2.0.0):
* Perform loudness averaging in linear scale, not dB.
* Remove extra square from compute power.
* Set reference db to 0.0 to align with librosa instead of magic 20.7 number for white noise reference. That way for power, dB=0.0 corresponds to amplitude = 1.0. Correspondingly, set dB range to [-80, 0] instead of [-120, 0] to align with librosa.
* Provide convenience functions for `power_to_db` and `db_to_power` that are aligned to librosa. Clearer docstrings.
* Moved db functions to core with other scaling functions.
* Tests to make sure the db functions match librosa.
* Default to `use_tf=True`.
* Moved diff() to core (not spectral specific)
* Default loudness, power, and rms to much more reasonable frame size default of 512 (instead of 1024/2048) [better time resolution].
* Compute Loudness on the fly by default (it's very fast and doesn't hurt training speed).

Future TODO:
Retrain and swap out models in Colab demo.

PiperOrigin-RevId: 391661045
copybara-service bot pushed a commit that referenced this issue Jan 15, 2022
@jesseengel
Fix bugs in compute_power and compute_loudness. Fixes #361 and #383.
fc99b4a 
Breaking Change (-> v2.0.0):
* Perform loudness averaging in linear scale, not dB.
* Remove extra square from compute power.
* Set reference db to 0.0 to align with librosa instead of magic 20.7 number for white noise reference. That way for power, dB=0.0 corresponds to amplitude = 1.0. Correspondingly, set dB range to [-80, 0] instead of [-120, 0] to align with librosa.
* Provide convenience functions for `power_to_db` and `db_to_power` that are aligned to librosa. Clearer docstrings.
* Moved db functions to core with other scaling functions.
* Tests to make sure the db functions match librosa.
* Default to `use_tf=True`.
* Moved diff() to core (not spectral specific)
* Default loudness, power, and rms to much more reasonable frame size default of 512 (instead of 1024/2048) [better time resolution].
* Compute Loudness on the fly by default (it's very fast and doesn't hurt training speed).

Future TODO:
Retrain and swap out models in Colab demo.

PiperOrigin-RevId: 391661045
copybara-service bot pushed a commit that referenced this issue Jan 15, 2022
@jesseengel
Fix bugs in compute_power and compute_loudness. Fixes #361 and #383.
a260cde 
Breaking Change (-> v2.0.0):
* Perform loudness averaging in linear scale, not dB.
* Remove extra square from compute power.
* Set reference db to 0.0 to align with librosa instead of magic 20.7 number for white noise reference. That way for power, dB=0.0 corresponds to amplitude = 1.0. Correspondingly, set dB range to [-80, 0] instead of [-120, 0] to align with librosa.
* Provide convenience functions for `power_to_db` and `db_to_power` that are aligned to librosa. Clearer docstrings.
* Moved db functions to core with other scaling functions.
* Tests to make sure the db functions match librosa.
* Default to `use_tf=True`.
* Moved diff() to core (not spectral specific)
* Default loudness, power, and rms to much more reasonable frame size default of 512 (instead of 1024/2048) [better time resolution].
* Compute Loudness on the fly by default (it's very fast and doesn't hurt training speed).

Future TODO:
Retrain and swap out models in Colab demo.

PiperOrigin-RevId: 421939115
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@jesseengel @mosheman5