Marcus Leich wrote:
>just ran your tests. I have that noise on both VMs. Had to put on
>Headphones to really hear it. There is some strange noise that always
>appears when the guy is talking. In the silence periods it's not there. I
>don't know what exactly that "robotic" effect should sound like, but what I
>hear seems to fit your description. I can not hear that sound when playing
>the file on QuickTime Player or VLC
If the sound you hear is a more or less pure tone or whistle at 8 KHz, then
that is sampling noise. It appears at the sampling frequency.
The examples have only an 8 KHz sample-rate, and even at 16-bits, sampling
noise will be perceptible to many people. This has as much to do with the
frequency, 8 KHz, as anything else.
You'll only hear sampling noise when there's a "signal", not when there's
silence, since silence is a flat DC level.
You don't hear it with QT Player because I'm almost certain that QT Player
(or QT itself) upsamples and/or filters slow-rate signals so they sound
better.
This could be further tested by synthesizing a sine wave at, say, 440 Hz,
with an 8 KHz sample-rate. The whistle should be apparent, but at a low
level. The sampling noise will be more obvious with 8-bit samples, and
using ULAW vs. PCM will also change its amplitude. If you don't know what
8 KHz sounds like, synthesize its sine-wave at a 44 KHz sample-rate, as a
listening reference.
When I first read the original post, the "robotic" characterization made me
think the samples were exhibiting aliasing, which is the transposition of
high frequencies into low frequencies. This is usually a result of
inadequate input filtering, when the samples are digitized.
I did not hear any aliasing when I played the examples originally posted,
so I discounted that as an explanation. I did hear the 8 KHz sampling
noise, but it wasn't "robotic", so I discounted that as what the OP heard.
Also, the 8 KHz sampling noise didn't vary between 1.4 and 1.5. That left
me with no perceptible "robotic" sound, and no plausible explanation, so I
said nothing.
Aliasing "folds" high frequencies down, so an increasing pitch in the
original signal results in a decreasing pitch of the aliased signal. This
gives a strange "metallic" or whining effect to voice recordings. If you
can listen to the original intonations, i.e. the rising and falling pitch,
rather than the words, and you notice that the "robotic" effect mirrors
those pitch changes in reverse, then it almost certainly is aliasing. Once
you hear it, you'll know it; it has a very different character from
sampling noise, which is constant pitch but varying amplitude.
I don't know how to eliminate aliasing that only appears in playback. As I
said, it usually appears in the original sampling, and once it's happened
it can't be removed. There are ways aliasing can appear only in playback,
but AFAIK not when playing the signal at its original sample-rate.
-- GG
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Java-dev mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden
