html/information.html

<p>This tool is based on the "aero" audio upscaling project (<a href="https://github.com/slp-rl/aero">original version</a>, <a href="https://github.com/pokepress/aero">modified version used here</a>). It takes digitized radio recordings and attempts to reduce noise/interference while recreating lost frequencies.</p>

<B>The FM model should be able to:</B>
<ul>
<li>Reduce stereo hiss</li>
<li>Reduce certain kinds of static/interference</li>
<li>Reduce 19khz stereo pilot</li>
<li>Reduce overly warm bass frequencies (boost them 3 or so dB if you miss them)</li>
<li>Restore lost high & low frequencies (generally only up to 16 khz since most training data is MP3-based)</li>
</ul>

<B>The AM model should be able to:</B>
<ul>
<li>Reduce certain kinds of static/interference</li>
<li>Restore lost high & low frequencies (capacity tends to diminish after 10 khz)</li>
</ul>

<B>The Telephone model should be able to:</B>
<ul>
<li>Reduce certain kinds of distortion</li>
<li>Restore lost high & low frequencies (capacity tends to diminish after 10 khz)</li>
</ul>


<B>Known Issues:</B>
<ul>
<li>The FM model is much better at removing noise from the stereo (L-R) signal than the mono (L+R) signal</li>
<li>The AM model tends to produce artifacts/sibilance in the 6-9 khz range, particularly in the 7-8 khz range</li>
</ul>