11025hz project rate inaccurate?

[>porridge<]

Good evening,

This is my first post but bear with me. I'm a widely unknown game modder (not an occupation) and have used Audacity since the 1.2.6 days to do bit of this and bit of that. Today I began exporting samples in ogg format for my latest silly project, but got KO'd in the process because apparently Audacity 2.x's 11025hz project rate isn't accurate anymore, meaning that what I hear in Audacity is definitely not what I'll get.

The specs:
Audacity 2.0.3 (the official installer from the official source)
Windows XP Pro SP3
I use Foobar for listening to ready ogg samples, it's good enough for me + miles better than VLC with oggs.

I went back to a portable version of 1.3.12 (a version that I could still find) to compare, and it was easy to spot the difference between the old and current generation indeed.

Unsurprisingly, all the higher project rates are fine, for instance with 16000hz audio is previewed accurately, but once going down to 11025hz - let alone 8000hz - everything changes.

Since words don't mean a thing and I have audio material to showcase the issue, I share it with you.

http://www.mediafire.com/?a8ab81wv4k5by80

Short description of the file: it's a 96khz WAV sample of a certain rally co-driver saying "one left". Open it in Audacity and you'll find that with higher project rates "ft" has some unwanted noise - let's call it a squeak - and so it should be, but in 2.0.3 the noise practically disappears with 11025hz. But once you export it as 11025hz OGG (the highest quality) or quite possibly any format really, the reality hits in the face and you'll hear the squeak in the end result. That should be the case with any proper playback software, at least.

Hopefully I was clear enough. I tried the search and checked the buglist but I didn't see this very issue mentioned anywhere.

[Gale Andrews]

Thanks for the report but I am not sure what you are trying to do or what you are hoping to prove.

Clearly the recording you attached is clipped and has the "whistling" distortion you refer to (though that part is not clipped).

If you mean that if you change the project rate from 960000 Hz to 110025 Hz then you hear less whistling on playback, this is to be expected. If you reduce it to 6000 Hz you will hear less whistling (and less of all the other content).

OGG export is lossy. If you want a copy of the original, export as WAV.

I imported your file into 2.0.4 alpha, changed the project rate to 11025 Hz and exported as OGG at Quality 10. I repeated those steps in 1.3.11 (I don't have 1.3.12 right now) then imported both exported files into 2.0.4 alpha. They sound the same to me (though I have middle-aged ears). If I invert one then play, there is still some audio to be heard (though not much, which suggests the files are very similar). Is this what you mean? I don't think I would expect absolute silence after inverting one file, as this is lossy encoding.

Or are you comparing the file you attached played at 11025 Hz with an OGG exported at 11025 Hz? If so I think the OGG has marginally more distortion on the "ft" but what are you expecting at such a low rate with a lossy codec?


Gale

[>porridge<]

Ok, I made a mistake bringing OGGs into my report, terribly sorry about that. The formats are now totally irrelevant, forget all of that. I may have said those OGG things, but I didn't.

Also, I could have said the same in two lines instead of that rambling. So, let's try again.

Audacity 1.3.x - project rate 11025hz - the whistling can be heard - therefore true to reality. Whereas...

Audacity 2.0.3 - project rate 11025hz - the whistling cannot be heard - not true to reality when you export the sample in whatever format.

So I'm not here to seek any technical help, as I know what I'm doing and I'm not only saying that. My aim was only to prove something, and this is how I hear things.

---------------

Even though this is utterly tedious, not relevant in the slightest, and I bet no-one's even mildly curious or interested, I'd better stress that the sample I shared is not an actual recording from scratch. It is derived from an in-car video from which I have removed the engine sounds with the brilliant Noise Removal tool among other things like upping the sound pressure immensely with Hard Limiter and so on. And as everyone in here knows, removing noise is totally different kettle of porridge in terms of audio work than anything else you can think of. I HAVE TO work with Audacity's 11025hz project rate from the very beginning not only because the game (Richard Burns Rally, the best game of all time) is using 11025hz ogg samples, but also that way I know all the time where things are heading and I don't remove too much by accident. In a way, I'm thankful that I have to work with such low quality for the sample quality itself does a lot of work for me.

(I'm very skilled at this noise removal business even though by saying that I may turn out quite weird and even sad individual. In fact, I have an excellent track record in this kind of modding.)

And finally: the whistling is basically part of car engine sound that's left in there only because while I was working with 2.0.3 I didn't know it was there. I thought I had removed the engine sounds more or less completely. And then when I exported the sample listened to it with foobar, I began looking at this issue retrospectively.

Hopefully this effort cleared things up.

[Robert J. H.]

It may be that you have different Settings for Interpolation and dithering (on playback, not on Export) in your two Audacity configurations.
Take a look at the preferences menu - quality.

[steve]

I can reproduce your results, but I think your analysis of the "problem" is incorrect.
The Audacity 2.0.3 version is more accurate than the 1.3.12 version. You can prove this by resampling the high sample rate original down to 11025 and then back up to 96 kHz and comparing the spectrum with the original.

The whistling is quite a strange phenomena and is due to multiple factors. The original audio has a lot of high frequency noise above 5512.5 Hz (the Nyquist frequency for a sample rate of 11025 Hz). Ideally these frequencies should be entirely cut without touching any frequencies below 5512.5 Hz. In practice that is impossible, though modern digital filters can get very close. By default Audacity 2.0.3 uses extremely high quality conversion when mixing, and medium quality when playing (medium quality conversion is faster than the highest quality, so performs better when real time conversion is required. The "squeak" is due to frequencies that are very close to the 5512.5 Hz cut-off point of an ideal sample rate converter. In Audacity 1.3.x when set to the highest available conversion quality, the digital filter is slightly less accurate than in the 2.0.3 version and the squeak is filtered out. The greater accuracy of the highest quality conversion in 2.0.3 allows the squeak to be preserved.

To produce a sound in Audacity 2.0.3 that is similar to 1.3.12, try reducing the the "High Quality Conversion" setting in "Edit > Preferences > Quality" to "Medium Quality".

[>porridge<]

Thanks for the help, guys. Really appreciate your willingness to waste time with my troubles.

I did find the solution via Quality settings. However, and interestingly enough, the settings were the same initially in both versions (the factory settings by the way, i.e. High Quality Conversion was Medium in both), and to me, INCREASING that setting to the best possible in 2.0.3 made it sound the same as in "true-to-reality" as the ancient 1.3.12 did in that "medium quality". Yes, it's mind-boggling, but it's sorted now and that's the main thing.

I admit It was pretty pathetic and amateurish display from me to blindly trust the default settings to begin with, I mean why would I want to have medium quality in anything, when the best is the best?

See, I told you I needed technical help. That wasn't a bug report. I may have disguised it as one, but it wasn't.

[>porridge<]

the settings were the same initially in both versions (the factory settings by the way, i.e. High Quality Conversion was Medium in both), and to me, INCREASING that setting to the best possible in 2.0.3 made it sound the same as in "true-to-reality" as the ancient 1.3.12 did in that "medium quality". Yes, it's mind-boggling, but it's sorted now and that's the main thing.

And I take this back. It was on High in 1.3.12. I must have got confused switching back and forth couple of times.

I may not be the sharpest ball in the box, but this was truly something else. The great first impression, eh, let's face it, I'd probably be better off changing my nick worldwide.

Now I stop.

Except to say that I'm thankful to the Audacity team for their work over the years.

[steve]

I must have got confused switching back and forth couple of times.

No problem - it's really easy to get confused when switching between different versions and different preferences. I've done the same myself many times