Unformatted text preview: imate the noise
After getting a noiseprint, if the source has non-noisy material like voice in it, you will be
able to see the peaks in the graph corresponding to “harmonic” audio. In many cases, the
noise follows a constant decreasing linear slope and does not contain peaks. The very small dips and peaks can be usually ignored when creating a noiseprint shape, but
the very large peaks are usually part of the material that you want to keep. Your main goal is
to have the envelope points be below what seems like the peaks from the non-noisy audio
and above what looks like the sloping noise spectrum. In many cases, you can get away with
a noiseprint of just 2 points, with the right-hand side point lower than the left-hand side
one, depending on the amount of high frequencies content in the noise.
To create a noiseprint envelope manually, do the following:
1. On the Noiseprint page, set the Fit size to 6 points. (The envelope will have 6 points.)
2. Click the button to create the envelope. The points will appear above the
noiseprint on the graph. 3. Move the envelope points so that they mirror the noise spectrum. We recommend that you move the envelope points below the peaks from non-noisy audio
and above the sloping noise spectrum.
4. Adjust the Noise bias control to achieve the noise reduction you want. APPENDIX A TIPS AND FREQUENTLY ASKED QUESTIONS 54 Modifying the amount of high-frequency reduction with envelope points
This technique is often used after you have created your own noiseprint envelopes. You may
raise or lower envelope points and change the amount of high-frequency dulling.
For example, to ensure that high frequencies are removed, raise all the envelope points in
the higher frequencies. (This action creates a low-pass filter effect.) By raising the envelope
points all the way to the top, you ensure that those frequencies will be attenuated by the
Reduce Noise by amount.
Conversely, to ensure that low frequencies of the spectrum are not affected, lower all the
envelope points (the ones on the left side of the noiseprint graph). In many cases, however,
removing only high-frequency noise is all that is necessary to improve the material. Why am I having trouble removing hiss from 8-bit files?
Files stored with only 8 bits per sample (as opposed to 16 bit or 24 bit) have an inherent
noise floor that is very high. Removing the noise is possible during 16-bit playback, but
when the files get saved back to 8-bit, the re-quantization puts the noise right back into the
file. The only way to remove the noise is to run a noisegate that mutes the “silent” regions to
zero. However, the noise will still be present during non-silent regions.
We recommend that instead of using 8-bit PCM files, a compressed format such as ADPCM
or MPEG be used that requires the same amount of storage (or less) than 8-bit PCM but has
a much lower noise floor. Why do I hear flange-type artifacts during processing?
When using Noise Reduction, trying to remove too much noise at once can sometimes
create high-pitched flanging-type artifacts in the processed signal. There are a couple of
things you may try to avoid flanging.
• On the General page, lower the Reduce noise by slider. Some files may require you to make multiple passes of Noise Reduction rather than trying
to make one large pass. Reduction of 6 to 10 dB is usually a very safe amount to try for
almost any sound file.
• Use Modes 2 or 3 to minimize the flangy chirps.
• Adjust the Noise bias setting with its slider.
• On the Noiseprint page, manually adjust the controls as they apply to the noiseprint. TIPS AND FREQUENTLY ASKED QUESTIONS APPENDIX A 55 Sometimes, especially for voice, lowering the Noiseprint envelope in the low frequencies
can prevent the bulk of the artifacts. The effectiveness of this adjustment, however, is
very much determined by the nature of the noise. For example, if the most objectionable
portion of the noise is high frequency material, as in tape hiss, the Noiseprint envelope can
be effectively lowered in the lower frequencies. Sound Forge
The following section relates to tips and questions as they apply when using Noise
Reduction with Sound Forge 4.5. Using the plug-ins with the Sound Forge Audio Plug-in Chainer
In Sound Forge, the Audio Plug-in Chainer allows you to run multiple DirectX effects at a
time. For example, you could run two instances for Noise Reduction, each with different
settings and noiseprint parameters. Or, create a chain that performs Click and Crackle
Removal followed by Vinyl Restoration, and then finish up with Noise Reduction, all in one
pass, based on your settings.
The advantage of the Audio Plug-in Chainer is that you can create a chain of effects that an
audio file will pass through one time. This can speed up the restoration process significantly.
However, you will need a fast computer to preview these processing chains in real-time.
To run multiple effects, do the following:
1. From the DirectX menu, choose Audio Plug-in C...
View Full Document
- Winter '10