manual/include/rhythm-ferret.html

<figure>
  <img src="/images/rhythm-ferret.png" alt="The Rhythm Ferret window">
  <figcaption>
    The Rhythm Ferret window
  </figcaption>
</figure>

<p>
  The Rhythm Ferret is a dedicated tool to speed up the usually labor intensive
  task of slicing and adjusting a sound region to match a specific time grid. It is
  especially useful for drum tracks, either to match a different tempo, or to
  adjust a slightly out of tempo performance.
</p>
<p>
  It is not limited to this use though, as it supports both percussive and note
  type detection, and can be used on melodic material too.
</p>

<h2>Accessing the Rhythm Ferret</h2>

<p>
  The Rhythm Ferret window can be accessed by <kbd class="mouse">right</kbd> clicking
  any audio region, then <kbd class="menu"><em>Name_Of_The_Region</em> &gt; Edit
  &gt; Rhythm Ferret</kbd>.
</p>
<p>
  Once the window is open, selecting any region will make it the focus of the
  Rhythm Ferret's detection, hence allowing to process multiple regions sequentially
  without reopening the window each time.
</p>
<p>
  The window itself is made of:
</p>
<ul>
    <li>a "mode" selection</li>
    <li>some parameters for this mode</li>
    <li>an operation selection, that for now only allows to <kbd class="menu">Split
      regions</kbd>.</li>
</ul>


<h2>The "Mode" selection</h2>

<p>
  As the Rhythm Ferret is able to detect both percussive hits and melodic notes,
  it is important to choose the best suited mode for the considered material,
  so that Ardour can perform the detection with the greatest accuracy :
</p>
<ul>
    <li><dfn>Percussive Onset</dfn> will detect the start of each hit based
    on the sudden change in energy (&equals; volume) of the waveform</li>
    <li><dfn>Note Onset</dfn> will detect the start of each note based on the changes
    in the frequency domain.</li>
</ul>



<h2>The Percussive Onset mode</h2>

<p>
  In this mode, only two parameters are active:
</p>
<table class="dl">
  <tr><th><dfn>Sensitivity</dfn> (%)</th><td>The proportion of the samples that must exceed the
    energy rise threshold in order for an onset to be detected (at frames in which
    the detection function peaks). This roughly corresponds to how "noisy" a percussive
    sound must be in order to be detected.</td></tr>
  <tr><th><dfn>Cut Pos Threshold</dfn> (dB)</th><td>The rise in energy amongst a group of samples
    that is required for that to be counted toward the detection function's count.
    This roughly corresponds to how "loud" a percussive sound must be in order to
    be detected.</td></tr>
</table>
<p>
  As those parameters are very material-related, there is no recipe for a perfect
  match, and a good peak detection is a matter of adjusting those two parameters
  by trial and error, and trying using the <kbd class="menu">Analyze</kbd> button
  after each try.
</p>
<p>
  Vertical grey markers will appear on the selected region, showing where Ardour
  detects onsets as per the parameters. This markers can be manually adjusted, see
  below.
</p>

<h2>The Note Onset Mode</h2>

<p>
  In the Note Onset mode, more parameters are active:
</p>
<table class="dl">
  <tr><th><dfn>Detection function</dfn></th><td>The method used to detect note changes. More on
    this below.</td></tr>
  <tr><th><dfn>Trigger gap (postproc)</dfn> (ms)</th><td>Set the minimum inter-onset interval,
    in milliseconds, i.e. the shortest interval between two consecutive onsets.
    </td></tr>
  <tr><th><dfn>Peak threshold</dfn></th><td>Set the threshold value for the onset peak picking.
    Lower threshold values imply more onsets detected. Increasing this threshold
    should reduce the number of incorrect detections.</td></tr>
  <tr><th><dfn>Silence threshold</dfn> (dB)</th><td>Set the silence threshold, in dB, under which
    the onset will not be detected. A value of -20.0 would eliminate most onsets
    but the loudest ones. A value of -90.0 would select all onsets.</td></tr>
</table>

<p>
  The Detection function, used in Note Onset mode to choose the mathematical strategy
  used to detect the note changes, is user-selectable:
</p>
<table class="dl">
  <tr><th><dfn>Energy based</dfn></th><td>This function calculates the local energy of the input
    spectral frame</td></tr>
  <tr><th><dfn>Spectral Difference</dfn></th><td>Spectral difference onset detection function
    based on Jonathan Foote and Shingo Uchihashi's "The beat spectrum: a new
    approach to rhythm analysis" (2001)</td></tr>
  <tr><th><dfn>High-Frequency Content</dfn></th><td> This method computes the High Frequency
    Content (HFC) of the input spectral frame. The resulting function is efficient
    at detecting percussive onsets. Based on Paul Masri's "Computer modeling
    of Sound for Transformation and Synthesis of Musical Signal" (1996)</td></tr>
  <tr><th><dfn>Complex Domain</dfn></th><td>This function uses information both in frequency and
    in phase to determine changes in the spectral content that might correspond
    to musical onsets. It is best suited for complex signals such as polyphonic
    recordings.</td></tr>
  <tr><th><dfn>Phase Deviation</dfn></th><td>This function uses information both energy and in
    phase to determine musical onsets.</td></tr>
  <tr><th><dfn>Kullback-Liebler</dfn></th><td>Kulback-Liebler onset detection function based on
    Stephen Hainsworth and Malcom Macleod's "Onset detection in music audio
    signals" (2003)</td></tr>
  <tr><th><dfn>Modified Kullback-Liebler</dfn></th><td>Modified Kulback-Liebler onset detection
    function based on Paul Brossier's "Automatic annotation of musical audio for
    interactive systems" (2006)</td></tr>
</table>

<p>
  Ardour defaults to Complex Domain, which usually gives good result for harmonic
  material.
</p>

<h2>Manual adjustment</h2>

<figure>
  <img src="/images/rhythm-ferret-demo-a.png" alt="The Rhythm Ferret: analysing">
  <img src="/images/rhythm-ferret-demo-b.png" alt="The Rhythm Ferret: Splitting">
  <img src="/images/rhythm-ferret-demo-c.png" alt="The Rhythm Ferret: Snapping to grid">
  <figcaption>
    The Rhythm Ferret: Analyzing, Splitting regions, and snapping to grid
  </figcaption>
</figure>

<p>
  Using the Rhythm Ferret consists usually in finding the right parameters to
  split the audio, by adjusting them and clicking the
  <kbd class="menu">Analyze</kbd> button. Each time an analysis is run, Ardour
  erases the previous results, and creates grey markers on the region according
  to the parameters. Those markers can be manually dragged with the
  <kbd class="mouse">LEFT</kbd> mouse button to adjust their positions.
</p>
<p>
  Once the markers are suitably placed, the second button in the down hand side
  of the Rhythm Ferret window allows to <kbd class="menu">Apply</kbd> the operation.
  At the moment of writing, only the <kdb class="menu">Split Region</kdb> is
  available, which will split the region at the markers.
</p>
<p>
  Those regions can then be manually aligned, or have their sync points set to
  the closest grid (as per the <a href="@@grid-controls">Grid settings</a> in
  effect), by selecting all the regions, and using the
  <kbd class="mouse">right</kbd> click then <kbd class="menu">Selected Regions &gt; Position &gt;
  Snap position to grid</kbd>.
</p>