h a l f b a k e r yEureka! Keeping naked people off the streets since 1999.
add, search, annotate, link, view, overview, recent, by name, random
news, help, about, links, report a problem
browse anonymously,
or get an account
and write.
register,
|
|
|
In order to explain this, I'll have to
illustrate the concept of sound design: All
high-quality digital audio is broken into
tiny bits called samples. Each sample is a
data point essentially telling the speaker
where to go at any given moment. Put
millions of points in a line and the speaker
will vibrate so fast that it produces a
sound which we then can hear. These
large files contain millions of samples
producing a CD quality sound we've come
to accept as high-quality.
Analog recordings are the original we use
to gauge quality with. Records are an
example. The unbroken vibrations of
audio essentialy contain an infinite
number of nuances too numerous to
record. It has to be simplified into finite
data points which then become etched
into a CD as binary data.
Scan a photo, and what you get is a similar
thing. The photo is broken into millions of
squares, each with a defined color, to give
an impression of the original. A high-
quality photo scan is usually a large sized
file with millions and millions of pixels.
Vector graphics are different. They are
bezier curve-based shapes that are
defined by a few points here and there
with handles to express the curve between
those points.
What if an audio waveform was defined
similarly? Only a fraction of points would
be necessary, with handles to define the
curve between them. The result would be
an unbroken mathematical curve with
infinite scalability i.e. a true LOSSLESS
audio format. Observe the attatched
illustration.
Ideas anyone?
Illustration
https://live.static...93_50e914463f_o.gif Vector Audio concept illustration [ophello, Sep 27 2005, last modified Oct 28 2024]
Maybe you could discuss it with [lawpoop]
vector-mapped_20waveforms [half, Sep 27 2005]
ADPCM
http://www-mobile.e...tandards/adpcm.html Adaptive Differential Pulse Code Modulation [csea, Sep 27 2005, last modified Nov 24 2011]
Fourier Series
http://mathworld.wo.../FourierSeries.html Recording a square wave will require infinite storage space to store the equation. [Worldgineer, Sep 27 2005]
[link]
|
|
Ok, so I didn't see this one yet. I still think
it's possible. |
|
|
What you are describing is very similar to ADPCM [link]. By the time it's truly lossless, you will have spent more bits than are necessary for linear PCM. But it is quite satisfactory for speech. |
|
|
A quality CD player will apply local bezier interpretation to the sampled data in order to output a waveform that's closer to smooth than the stair-step that a cheap player will output. |
|
|
What you're proposing sounds like generating the bezier from a very high sample rate first, and then saving only the minimum number of points required to accurately reproduce that waveform within a given error limit. I'm no audio compression expert, but I was under the impression that this is how some variable bitrate compression schemes worked. |
|
|
Variable bitrate is a modified version of
the mp3 codec. It doesn't rely on
waveform curvature mathematics. The
resulting waveform of a VBR file is not the
same as the analog. I've looked =) |
|
|
Your solution may work fine for long slow sine waves (compared to a high frequency digital recording, that is). As you approach higher frequency sound, your solution will either become lossy or require more data. |
|
|
If you get rid of the equation part of this and focus on the concept of vectors (descriptions of measurement and length), you'll find that digital recording is made up of vectors - just in a fixed format. |
|
|
Even high frequencies can be defined by
relativelly fewer points. It would be a
variable rate format, but still not VBR. |
|
|
//Even high frequencies can be defined by relativelly fewer points.// I disagree. If you're talking about equation-form sound files then then equation gets very long very fast if you deal with high and variable frequencies. |
|
|
I'm not hoping for an equation, but a
peak-to-trough locator. You'd only need
to record when the highs and lows occur,
then plot some kind of bezier curve
between those points. This wouldn't be
some longhand equation, but a semi-
sample based curve. |
|
|
Ah, so you're not really looking for lossless. I think you'll find minimal reduction in data except in simple sounds or if you're willing to lose a lot of the higher frequencies. |
|
|
The only flaw in your idea is that when an audio signal comes out of a CD player, it is an analog signal. The data on the CD does not determine the function of the speaker. When the CD player reads the disc, it takes the binary information and converts it into an analog signal through the use of the D/A converter (digital to analog) and then it goes through a filter thus making all of the samples into a continous wave. |
|
|
[Jscotty] well, that and your ears! I wonder if there's something there we can think about? |
|
|
If anyone uses CoolEdit (or whatever Adobe call it nowadays, Audition?) they display the information like this. |
|
|
I guess with a number of spline-curves it could be done. Hm, you could probably get your graphics processor to 'render' the sounds, too... |
|
|
[Jscotty],
>an audio signal comes out of a CD player, it is an analog signal. |
|
|
Unless your CD/DVD player has a S/PDIF (Sony/Philips Digital InterFace) signal output, many do. Physically, it may look like an RCA connector, or an optical TOSLINK connection. This may carry the digital audio information on the CD, or Dolby Digital bursts of data if from a DVD. |
|
|
These digital signals are generally routed to a receiver for decoding or external DACs. |
|
|
[Ian} I think you'll find that a non-periodic clock requires more data to specify the clock location than is saved by its non-preiodicity. |
|
|
I'm bunning this idea for being cool, not because it has any inherent advantage. What you need to remember is that sound is complex and contains a lot of information, and that information must still be stored and recalled regardless of the form in which it is held. Normal cd audio (16 bit/44.1kHz) is not really a 'lossless' format in itself. It simply takes the (theoretically) infinite amount of information contained in an audio wave and discards the frequencies above 22.05kHz that we cannot hear. Your vector system would also have to lose some of the infinite amount of information you could record, it's merely a question of how accurate you want to make it. My guess it that if you developed the system well, it would perform very similarly to the current digital system. |
|
|
The key here is smaller file size. Less data,
same sound. It isn't really geared for cd
audio. |
|
|
I was under the impression that MPEG encoding DID use this method. The FF/Wavelet Transforms convert the analog audio signal into a set of frequency components with amplitude / phase data. The fact that this is then converted BACK into samples indicates that a sample based format is a more efficient storage method. Perhaps the functional information could be optimised to be smaller but this could be very processor expensive. |
|
|
<aside> interesting user-page [QuantumMechanique] |
|
| |