The problem with effect "Normalize".

[юра00]

Sound the same. But after some training it is possible to differ even 0,01 db. May be, develop additional plugin, nearly "Approved normalize" with 2 or more decimal places? I should download!

[steve]

But after some training it is possible to differ even 0,01 db

No. I have many years of professional experience in sound engineering, and I can clearly hear a 1dB difference, but no way can I hear a 0.1 dB difference, let alone 0.01 dB difference. The developer of the updated effects was intending to make it whole numbers only, but I persuaded him to add one decimal place. I doubt that I could persuade him to make it 2 decimal places unless someone can come up with a good "use case" that demonstrates the benefit of more decimal places.

[юра00]

Thank you! Please, consider my note in this topic for next, now developed, I hope, version of Audacity, 2.1.2 or somelike...

[Robert J. H.]

But after some training it is possible to differ even 0,01 db

No. I have many years of professional experience in sound engineering, and I can clearly hear a 1dB difference, but no way can I hear a 0.1 dB difference, let alone 0.01 dB difference. The developer of the updated effects was intending to make it whole numbers only, but I persuaded him to add one decimal place. I doubt that I could persuade him to make it 2 decimal places unless someone can come up with a good "use case" that demonstrates the benefit of more decimal places.

I guess it depends highly on the context.

Simple example:
- Import a clip
- duplicate it
- amplify one by 0.1 dB


The audio is now clearly audible, even with a amplification of 0.01 dB.
I get a mixed level of -62 dB peak with my test clip.

There are other effects that do suffer from the newly introduced validation with predefined decimal places.
For instance, Paul stretch allows only a time resolution down to 0.1 seconds.
Imagine a reverb with such a setting - it's a cathedral.
I often use it with stretch factor 1.0 and Time resolution 0.025 - it gives a nice touch to some pads while keeping some transitions. Not possible anymore.

Robert

[Gale Andrews]

Moved to "Adding Features".

Presumably there is a "scientific" use case for two decimal places of accuracy? Is that why Amplify uses two places, as a more "advanced" tool?

Gale

[Gale Andrews]

There are other effects that do suffer from the newly introduced validation with predefined decimal places.
For instance, Paul stretch allows only a time resolution down to 0.1 seconds.
Imagine a reverb with such a setting - it's a cathedral.
I often use it with stretch factor 1.0 and Time resolution 0.025 - it gives a nice touch to some pads while keeping some transitions. Not possible anymore.

That Paulstretch example seems a poor decision. One decimal place is not sufficient for length in seconds. Three decimal places is allowed for the durations in Truncate Silence.

Are there any other examples where accuracy has been reduced?


Gale

[steve]

If you wish to normalize to a very low level, you can normalize down to -99.9 dB (which is extremely quiet and well below the minimum amplitude for 16-bit audio).

Presumably there is a "scientific" use case for two decimal places of accuracy?

For "scientific" cases, there is a "scientific" tool - "Nyquist". With Nyquist you can have 32-bit float precision (about 15 decimal places).
For example, to amplify by -0.12345 dB

Code: Select all

(mult *track* (db-to-linear -0.12345))

Is that why Amplify uses two places, as a more "advanced" tool?

There are a few cases where amplifying to greater precision than 0.1 dB is a requirement. One example is when attempting "phase cancellation" ("cancelling" one sound by adding its inverse).

For both Normalize and Amplify, whole number dB is probably adequate for 99% of use cases. 0.1 dB steps is probably adequate for 99.99% of use cases. Two decimal places in the Amplify effect is probably adequate for (guess your own figure) 99,9999% of use cases? For the remaining 0.00...01% of use cases, there is Nyquist. I can't think of any practical use cases that requires more than 1 decimal place in the Normalize effect. I think there is a stronger case for giving the Amplify effect 3 decimal places, but even that is very much a fringe case.

There are other effects that do suffer from the newly introduced validation with predefined decimal places.

Yes, there are a few such cases - particularly with Nyquist effects that previously provided a slider range for the most common values and an extended text input range. The extended text input range is now not available.

For instance, Paul stretch allows only a time resolution down to 0.1 seconds.

Another example is "Bass and Treble". I'd like to increase the range to say +/- 30 dB
The other possible solution would be to change the validation to allow an extended range beyond the slider range.

[steve]

That Paulstretch example seems a poor decision

There are many cases, but generally it is not due to a recent "design decision" to change the accuracy, but rather that the default accuracy is now strictly enforced (validated) whereas previous it was not.

[Robert J. H.]

That Paulstretch example seems a poor decision

There are many cases, but generally it is not due to a recent "design decision" to change the accuracy, but rather that the default accuracy is now strictly enforced (validated) whereas previous it was not.

Shouldn't be hard to fix those issues.
The main task is to list the effects in question and to agree on the needed accuracy.


Another accuracy/validation bug I've encountered:

I changed the gain on a track and wanted to reset it, so I opened the gain dialog and entered "0".
Nothing happened, the gain was still at -6 dB.
I had to move the slider to zero and the value displayed was 1-06. Same with Pan.

However, I'm not able to reproduce it anymore - it now takes the 0. Strange thing.

Robert

[Gale Andrews]

Presumably there is a "scientific" use case for two decimal places of accuracy?

For "scientific" cases, there is a "scientific" tool - "Nyquist". With Nyquist you can have 32-bit float precision (about 15 decimal places).
For example, to amplify by -0.12345 dB

Code: Select all

(mult *track* (db-to-linear -0.12345))

How many users will research to find that out, the more so without a published library of code snippets?

I can't think of any practical use cases that requires more than 1 decimal place in the Normalize effect. I think there is a stronger case for giving the Amplify effect 3 decimal places, but even that is very much a fringe case.

If" there is a use case for normalizing stereo channels independently to two decimal places of accuracy or to remove DC offset while normalizing to two decimal places, it is more convenient to do that with Normalize than Amplify.

But in any case it would look more consistent if the permitted accuracy was the same in both effects, and I don't see bad side effects of doing that. Do you?

For instance, Paul stretch allows only a time resolution down to 0.1 seconds

Another example is "Bass and Treble". I'd like to increase the range to say +/- 30 dB

The other possible solution would be to change the validation to allow an extended range beyond the slider range.

Why was that extended range not allowed in the first place? I thought the general feeling was that extended range was good, so that not too many steps were required in the sliders?

In any case as a sighted user I find the validation error message box that you have to dismiss very annoying. It would be more civil (and produce no ding noise) to grey out OK and display the error as inline text. Is such a solution too confusing for VI users?


Gale