As far as I can tell, Analyze → Label Sounds isn’t paying attention to the audio in the project. No matter how I adjust the settings it sees my selection as one continuous sound sample and only adds one label at the start of my selection.
To try and make it idiot proof (in case I was doing something wrong) I highlighted a few chunks of an audiobook, used Generate ->Silence to create the silent spots noted in the screen-shot. I then ran the plugin once on the default settings and once on “Point before sound” and based on the labels it outputted, it seems like it’s interpreting the volume level as constant?
If I mis-read something I’d love to know what I missed in the manual but these were not the results I was expecting.
Results obtained using:
Audacity 3.0.0
Windows 10 Version 20H2 (OS Build 19042.906)
There is a bug in Audacity 3.0.0 that prevents Nyquist effects from working on long selections.
“Label Sounds” is a Nyquist effect.
For most Nyquist effects, “long selection” means about 45 minutes.
This has been fixed in Audacity 3.0.1 which I’m hoping will be released around the end of the month.
The workaround for now is to apply “Label Sounds” to shorter sections.
I guess I’ll reserve too many comments on this until we are told the bug is fixed, but when I started trying to use Label Sounds, I found it VERY fussy, confusing, and unreliable. After many many tests with different settings, I did get it to properly mark most of the tracks on one album. In contrast, Silence Finder has always been very simple and worked very well for me.
I agree that it is more complex, but it is a lot more flexible, more accurate, and very reliable (if it were not for the bug in Audacity).
The addition of “RMS” and “Average” options for Threshold measurement are particularly useful when labeling vinyl recordings. The old effects were very prone to adding extra labels when encountering a clicks or crackles between tracks, which is largely avoided by using “Average” (or RMS) instead of “Peak”.
Once you’ve found some useful settings, you may find it helpful to save them as “user presets” so that you then have a starting point for similar tasks. To access presets, click the “Manage” button.
(Personally I wanted to retain the old “Silence Finder” (because it is so simple) in addition to the new effect, but I was outvoted.)
If you frequently do automatic labeling, then it’s definitely worth persevering with the new effect because it is much better than the old effects in many ways. I’m disappointed that for many, the first impression of this effect will be a bad experience because of that bug.
Thanx Steve - your comments give me hope. I have written down the settings that I finally found to work on that one test album, so I look forward to doing more testing after the fix comes out. Before this discussion, I wasn’t sure I would ever touch it again, so yeah, first impression was really bad.
I am not sure if it was just the bug or what (since I DID get it to run eventually on the full album), but I am concerned about the default settings. I expected it to perform at least moderately well with the defaults, but instead it was a complete waste of computer time. Due to the extra complexity of this tool, I think it is critical for the default settings to produce at least fair results on the vast majority of recordings, just to keep from scaring the users away.
I am glad I still have the old Silence Finder on all of my computers. It may not have been perfect, but the ease of use had a lot of value.
The defaults seem reasonable to me. They are based on the defaults for the old Sound / Silence Finder effects.
In my very limited tests so far, they are completely and totally unusable. But I am not trying to say they are just flat bad default settings, just that what I have seen so far makes me worried that they perhaps could be better.
Even on your sample shown above, I have to wonder what might be going on around labels 2-3, 4-5, and 17-19. But if any of my tests had even been that close, I would not be complaining.
At this time, I have only tested the Label Sounds Effect against album files recorded from streaming sources, but for five separate files, the results have been wildly idiotic; ALWAYS producing way more tracks than actually exist in the source. In one case, where there were actually 11 tracks, this effect labeled 35! The last one I tried had 10 tracks, and Label Sounds default settings marked 19 tracks. All of these albums have been pop and rock albums.
Although I have had to change every one of the individual settings in Label Sounds to get usable results, the absolute worst one for this source material is the Threshold. These recordings have a very quiet base noise level, and I have to set the threshold over -40 to get even moderately acceptable results.
I do realize that Audacity is intended for use with many different types of recorded material, and perhaps a low-noise digital source file may be the absolute worst thing to try with the Label Sounds Effect. But that is where I started, and for me so far, it has been an almost total failure. For this material, I do also have to change the default settings in Silence Finder, but its simplicity makes finding something that works there so much easier.
Maybe I will see better results when I try it against a Tape or LP copy. I will get around to that eventually.
Just wanted to update this thread with my current results (since I posted above about how poor I found this tool on initial testing). I am now very pleased with this tool and no longer even think about using the old Silence Finder.
Bottom line is that I have found settings that work VERY well for me when analyzing recordings from all three sources I use: streaming, Tape, and Vinyl. I still believe that the default settings are way off, especially for the Threshold level.
For anyone struggling with getting good results from Label Sounds, Here are some default settings that have produced very good results for me, with generally only minor tweaks for different files:
I prefer Point before sound label types, and find that Peak level works best for very silent recordings, RMS often better for Vinyl (but I have not tested yet with any really noisy recordings).
Threshold= -46
Minimum silence duration=1.4s
Minimum label interval=30s
Maximum leading silence=1s
Maximum trailing- n/a for point labels
Are you aware that you can save your favourite settings as “presets”?
(Click on the “Manage” button)
Unfortunately Nyquist plug-ins don’t yet support shipping the effect with more than one (default) preset. I hope that they will in the future, and if they do I’ll add a few presets for common tasks such as CD albums, Cassette recordings, Vinyl albums, Speech recordings, …
Yes, those look like a good starting position for cassette recordings of music (and similar).
The problem is that the tool could be used for almost anything.
If, for example, you have a casual recording of two people speaking (perhaps a Skype recording), then the noise level is likely to be much higher, and the “minimum label interval” would probably need to be much shorter. On the other hand, if you have a digital transfer from a commercial CD, then you can probably drop the threshold level down to around -80 dB.
One aim of the defaults is to produce at least “some” labels, regardless of where the audio came from. It’s unlikely that the defaults will be optimal for any specific task, but they should at least give you something.
In the absence of presets, perhaps the manual could suggest some ball-park settings for a few different tasks. What do you think?
In the absence of presets, perhaps the manual could suggest some ball-park settings for a few different tasks. What do you think?
I think that would be an excellent idea. I do realize that my very limited use of Audacity is a small part of all the many different things that other users are doing. That is why I tried to be very specific to point out I was only dealing with typical music recordings from three sources (Streaming, Tape, and Vinyl).
