SurroundByUs.com

Noted - I shall retreat back into the shadows. Thank you for your time.

An idea that would straddle both worlds here would be for Phil to suggest a song for us, and a couple of us could tackle it and post our layout for it. I think it could shed some light on how differently some of us old-timers (and when I say that I, of course, speak mostly of the other River ) tackle something differently from one another.

I'll check it out, but I will say that I'm not responsible for what the surround output of an mp3 sounds like.

I'm sorry the input source is of low resolution but it I selected it because it's public domain, quite short and not terribly complex (being synthesizer based as opposed to orchestral).

Other than the obvious (GIGO) is there any other impact of using a lower resolution source? What I mean is, will the SPEC code react poorly to a having a lack of data to work with? Would a simple drum/symbol (thump, brush) be treated very differently by SPEC if input as a 160Kbps .mp3 verses the same segment as a .flac or 2,117Kbps .wav.

I would be happy with any source - if there is a more suitable candidate available then we should switch - the value for me is not in the choice of source track but in the method employed to convert.

Electronic music converts easily and well. Also was an easy win with SLICE, which was developed before we had ArcTan. As such, it might not be the best choise for learning the nuances.

The better the source, the better the results. Mp3, and other lossy formats already have artifacts, and SPEC just magnifies them, so we always suggest lossless sources.

Even at that, some 44.1 or 48 Khz lossless sources produce artifacts (typically a swishing or "backwards" snare or cymbal sound) and the solution is to upsample to 88.2 or 96 Khz, run through SPEC and then downsample to the rate of your target media.

FYI we would "oversample" if we could, so the above step would not be needed. Know any expert C++ programmers in the area of windowed FFTs (Specifically "Phase Vocoders")?

Thanks for the response Zee. I shall look for a better source example - not electronic, higher sample rate, including vocals - while still being public domain.

I have personally been struggling to do anything useful with "George Benson - This Masquerade" as taken from album:

I manage to get some percussion to the rear but that's about it. I also find that my efforts produce a harsh edge to lead vocals - perhaps they were always there but I would prefer not to highlight that but don't know how.

"Counting Crows - Big Yellow Taxi"
Came out quite nice but again the male lead sounds a little harsh and there are some fun components which I would have like to "steer" more towards the rear but didn't know how (or if I can).

As someone who only discovered SPEC recently I have no experience with the world "prior to ArcTan". While I saw the SPLICE option I viewed it mostly as something which was there for upward compatibility and not something to be used in place of ArcTan. -- So there's another preconception removed.

I found the description of what happens when you run a lossy sample through SPEC very illuminating - I have heard the problem you describe and now understand the cause. Also the possible work-around. Did I miss a "tips and tricks" or "trouble shooting" guide?

C++ programmers. I'll need to research "windowed FFTs" and "Phase Vocoders" so I can better understand how to correctly pose the question to my network (so they can re-pose the question). I'll try to spend some time on that this weekend. Any links to supporting material you could send me would be most helpful.

OK I think the "rivers" and fan51 can work from those examples.

Have you seen this?

viewtopic.php?f=8&t=221&p=1487&hilit=how+to+win+with+arctan#p1487

The reference to C++ is because that is what the Plogue Bidule SDK is written in (I see you've found the Plogue forum, so there's a SDK section there. No doc ). You can call c from c++ though.

Another approach would be using the Stienburg VST framework. Anyone who has developed a VST plugin (ideally an oversampling one) would be familiar with that.

Google "Phase Vocoder" for details on that.

What we are after is a two way transformation of high quality audio between the time domain and the frequency domain. It also needs to be VERY efficient code as we need two time to frequency and 5 frequency to time transforms to run simultaneously in essentially real time, on the audio streams.

The "FFT" and "iFFT" bidules in Plogue are not really pure FFTs but "Phase Vocoders". They sound good, and the performance is good enough for modern computers, but we'd like to take things further. Oversampling, and maybe the ability to bust out from plogue and into other platforms like other VST hosts or even plugins for foobar and/or other players. GPU acceleration is another interesting direction.

RE the Plogue Bidlule SDK, the phase vocoder transforms would need to mirror the functionality of the existing FFT and iFFT bidules, utilizing the magnitude and frequency signal types in Plogue.

Audio -->
Magnitudes of each Freq. bin, and absolute center freq. of each bin -->
existing SPEC algorithms -->
Audio

There are millions of songs in the history of music. In order to have this work for you, understanding what a centered vocal is when listening in stereo (especially over headphones), understanding what an echoey or reverbed vocal sounds like, etc., is going to be important for you. When listening to an album in stereo, hearing what sounds are in the front of the mix, what is panned hard right and left, and familiarizing yourself with the music beforehand is important. If these are not concepts which come easy to you, I myself learned through trial and error, but also through just listening a lot. I can assure you that, from 9 to 5, I am the furthest thing from an audio expert. As a matter of fact, I actually don't have a lot of technical ability.

The "How to win at ArcTan" guide is an excellent start, as well as much of what River136 suggested, with just about anything. If you really need an example......well, let's try "Houses of the Holy" or something.

I've attached the layout I used for the test track you sent. As Zeerround said, this was a pretty basic track that would convert well without much effort.