Audacity Forum

steve wrote:So did you try raising the noise floor setting to see what happens?

Noise floor at -28, always hear it, see it front and end of the file (comparing to the noise floor before compression is added.) No visible or audible sound before compression, obvious in both cases after.

As I approach -20, it is almost totally gone on the start of the sound (around -75 dB, close enough for me), but the tail end still barely has it. I can see/hear that end. I can see the front end noise if I zoom in enough, but for practical purposes it's silent. (Someone here on the board posted an "Amplitude..." plugin and I use that to check that I didn't have noise I couldn't see, plus I zoomed way in before applying the compressor.) Note that it's adding something, but for practical purposes it's not enough to matter.

I did lots of testing. It's inconsistent with itself between different files. The AMOUNT of the added noise is longer/shorter depending on how the audio starts/ends. Not by much, but there is some variation. That implies to me if that you have music, it will depend on how it starts and ends, what it's level is. I also didn't test this audio with it EQed, it's just raw voice with the noise removed via the noise tool.

It's still adding the sound at the beginning no matter what, but for practical purposes I have test cases where that is low enough to be practical IF I raise the floor to -20 (or higher), but I'm not clear that works for me. IF I get to -15 or so then I can't hear it or see it.

I was paying no attention to the quality of the compression with that floor on the voice, just checking out the noise based on this thread. I'll next compare some audio with the floor at -28 vs. -20 and see what that's like in terms of compression. I'm out of time for testing today (actually need to process some audio), but if the floor is high enough, the extra noise goes down.

BTW - Anybody else doing audiobook/podcast or other naked voice type processing: What settings are you using for Chris/compression? Anybody else using a noise floor that high? Maybe I'm missing something so please let me know what you are doing.

I have said for a while now that properly applied Chris Compressor can solve all your Spoken Word Standards problems in one swoop. I have never noticed the beginning and end problem, but I'm trying to remember if I cut both of those off anyway each time I use it. I know it was a very serious issue with an early version and I thought 1.2.6 was designed to solve that.... It's been a while.

Chris designed this compressor so he could listen to opera in the car. He said so. He designed it to be musical and still carefully remove all the enormous volume changes in each dramatic scene. That's why it doesn't work like other compressors. He started with Verdi and Puccini and worked up instead of starting with attack and release differential algorithms and working down.

I posted three waveforms. An original "podcast" performance, and two Chris variations: Standard Settings and Compression changed to 0.77 from 0.50.

http://kozco.com/tech/audacity/pix/Chri ... Afters.png

The 0.77 version is indistinguishable from the same show broadcast on the local FM station using broadcast compressors, and in fact, I can ride around in my truck/lorry without having to constantly adjust the music system volume.

It would not shock me that one of the two settings resulted in a show that conforms to Spoken Word sound standards, since they and FM Radio have similar requirements.

Koz

Koz,
You gave me some idea, couldn't Chris' Compressor be used for exactly the opposite, namely to bring broadcasted music back to its original dynamic range?
You could experiment to figure out which (negative) factor should be entered to just achieve this.
There's a floor when used as expander though, the gain is not adapted accordingly. Certainly a point for improvement, along with some input correction (some values produce division by zero).

kozikowski wrote:< snip>

I posted three waveforms. An original "podcast" performance, and two Chris variations: Standard Settings and Compression changed to 0.77 from 0.50.
http://kozco.com/tech/audacity/pix/Chri ... Afters.png
<snip>
Koz

As I mentioned, I'm fine with the sound quality of the compressor but I certainly am not going to assume someone couldn't create something stronger someday. I have no idea, but it works great for me except the extra noise it adds. And I've found workarounds, but still would love to find a way to eliminate them.

Next question:

Please cure my ignorance:
When you say "Standard Settings" I'm guessing/assuming that means "Defaults"? And if you changed Compression you mean "Compression Ratio", and you're leaving all the other settings per the default?

What settings do you like for spoken word?

I hope you're right and it's the best choice, since it's what I'm using regularly.

As a side bar: My wife and I are both musicians (she has a voice degree from University of Michigan) so I heard tons of operas as a young man along with paying some dues working in a recording studio when I was young too many decades ago.

Opera is not exactly what I would consider close to spoken word, and I'm not a huge fan of what FM stations do to most music, in terms of compressing. I understand why they do it, and it sucks the life out of great dynamic music, but that's a totally different discussion. I'm not saying a product designed for opera would be unsuitable for an audiobook, but in my mind they are in different leagues in terms of dynamic range and what is needed as an end product. So far Chris's IS the best I've found, but still I expect over time someone will find ways to make it even better or a different way of doing it. If I'm wrong that's fine, I'm already a huge fan of it other than the one issues I've outlined in previous posts.

I love hearing your insights into the original design. Great to have some additional depth. Thanks for sharing with those of us who are newer.

The Dolby Corporation made its name by designing a compressor/expander system that you couldn't hear working. It was exactly symmetrical and in its "A" version made tape overdubbing possible because the 10dB noise boost that resulted was precisely the 10dB suppression that Dolby offered. Whole worlds opened up. The first layer of Analog Stereo Television was Dolby processed to get rid of the FM Stereo Noise Penalty. Analog TV sound is (was) FM. It worked. TV Stereo appeared in living rooms without fffffffffff in the background.

You only get that kind of transparency if you can control both ends. Nobody has managed to "de-Dolby" an audio performance except Dolby (several posts on the forum) and I suspect a look-ahead compressor with semi-unknown audio characteristics is unlikely to uncompress an existing performance. There's no way to know what the trigger points are without knowing what the original compressor was thinking about. Even worse is the assumption of only one compressor or processor. FM Broadcast systems have two. One overall level setter and a second peak grabber -- and later ones were even worse. I don't know any way to reverse one of those.

What were you going to use as a test clip?

Koz

The particular application I have is a spoken word podcast/broadcast (plus telephone voices and music) by two people who have an enormous range between "mumbling in one's beer" and a thermo-nuclear laugh. Whereas the squashed radio broadcast is perfectly enjoyable in the car, the podcast is not. The podcast has all the common complaints of people who fail to conform to Spoken Word requirements. The sound peaks are all over the map, the RMS value is too low, the voice volume varies too much, etc. etc. etc. This on top of room noise and echo problems common to people trying to record in their kitchen.

http://kozco.com/tech/audacity/pix/Chri ... Afters.png

So yes, I do have Spoken Word performance examples where Chris worked very well and continues to. That waveform example is this particular show which returns from a musical bumper (the level segment) into voices. Note the variation in the upper one is all but completely eliminated in the bottom (Compression Ratio 0.77), plus the overall volume increases without clipping or overload damage. Further, because of look-ahead, the sound neither pumps nor breathes.

It will be very difficult to beat that.

Koz

And yes, Look-Ahead is the problem at the top and bottom of the clip. At the beginning, Look Ahead presets the compression compared to a clip that isn't there yet, and at the end the gain used to blast off because the show sound would "mysteriously" drop to zero. I don't remember any more than that, other than a fuzzy idea that this is the solution. Legacy versions were much worse.

Koz

Unlikely wrote:Maybe I'm missing something so please let me know what you are doing.

When starting a recording it should be:
"Record,... 3...2...1...go..."
rather than
"3...2...1...Record..."

The very start of a recording can be a bit iffy for many reasons. Allowing a few seconds at each end of the recording allows space to tidy up when editing. Some recording programs even have a time line that starts at -2 seconds so as to encourage a 2 second lead in.

kozikowski wrote:Look Ahead presets the compression compared to a clip that isn't there yet,

Quite so. Compressors can use "lookahead" or not.
Compressors work by changing the gain (amplifying up/down) according to the input level. If the gain changes too rapidly then it is very noticeable (commonly described as "pumping").

Without lookahead, the gain will change only in response to the input level. If the input level goes up suddenly then it takes a while ("attack time") for the compressor to catch up and reduce the gain. Real time compressors tend to do this because they don't know what is coming until it happens. In order to prevent sudden peaks from going too high, a "limiter" may be used after the compressor. A limiter is similar to a compressor but has a very fast "attack" so that it can limit sudden peaks.

Compressors that are used for processing recordings may use "lookahead". The compressor looks at what the level will be doing next. If a sudden peak is approaching then the compressor will start reducing the gain a short time ("lookahead time") before the peak arrives. This prevents "overshoot" of the output level, but if there are a lot of sudden peaks (transients) then it will tend to push the level low and sound too quiet. For recordings of percussive sounds it is often better to use a compressor without lookahead (or a short lookahead time) and follow it with a peak limiter.

It is also important that after a high peak has passed that returning to normal gain does not occur too quickly. This is the "release time" (the Audacity compressor calls this "decay"). If this is too fast then you can hear "someone has turned the volume up". If it is too slow then the sound after the peak is likely to be too quiet until the compressor has "released" the attenuation.

As Koz wrote, there is a problem at the start/end of the recording because the best that the compressor can do is to guess what the signal level might have been before the start and after the end of the recording. If the compressor guesses that the signal level was 0 dB before the start then the gain should start at minimum and then "release" to the current level, which will tend to make the start too quiet. If the compressor guesses that the signal level was silence before the start, then the gain should start at unity (1:1) which may allow the input to overshoot until the compressor has caught up. Similarly at the end of the recording. Probably the "best guess" is that the level before the start is the same as the level at the start, and the level after the end is the same as the level at the end, but even this will not always provide the desired result.
The only solution guaranteed to work is if there is sufficient lead in / out to allow the compressor the time that it needs to change the gain to the correct level.

I just processed a Saturday radio show and I didn't hear anything wrong with the beginning or end.

You should post the damage you're hearing on the forum. You may have to do it in multiple passes due to forum restrictions.

http://forum.audacityteam.org/viewtopic ... 49&t=72887

Koz

kozikowski wrote:I just processed a Saturday radio show and I didn't hear anything wrong with the beginning or end.

You should post the damage you're hearing on the forum. You may have to do it in multiple passes due to forum restrictions.

http://forum.audacityteam.org/viewtopic ... 49&t=72887

Koz

I'll upload some samples with the issues ASAP. I used the compressor last night on a 30 min clip and had the issues front and back. (To be clear: This is minor, and not exactly a show stopper...)

I'll probably just run that same clip, see if I can duplicate it with 10-30 seconds of audio. My gut says it has to do with how much room tone/silence is at the front/back of the cuts. We ask our narrators to leave at least 5 seconds up front and at the end. I've duplicated the issue 20 times, so no problem uploading samples.

(I have a project due by tomorrow, so you'll see the upload within a day or two.)

I appreciate all your insights!

Don