am_pitchshift_1433.xml

<?xml version="1.0"?>
<!DOCTYPE ladspa SYSTEM "ladspa-swh.dtd">
<?xml-stylesheet href="ladspa.css" type="text/css"?>

<ladspa>
  <global>
    <meta name="maker" value="Steve Harris &lt;steve@plugin.org.uk&gt;"/>
    <meta name="copyright" value="GPL"/>
    <meta name="properties" value="HARD_RT_CAPABLE"/>
    <code><![CDATA[
      #include <stdlib.h>
      #include <math.h>

      #include "ladspa-util.h"

      /* Beware of dependcies if you change this */
      #define DELAY_SIZE 8192      
    ]]></code>
  </global>

  <plugin label="amPitchshift" id="1433" class="PitchPlugin">
    <name>AM pitchshifter</name>
    <p>This plugin works by running a single write pointer (monotonic) and two read pointers (pitchscaled) over a ringbuffer.</p>
    <p>The output is faded between the two readpointers according to the sine of the distance from the write pointer. The design is based on the mechanism of a mechanical pitchshifter I saw in the Gemeentemuseum in Den Haag, though I'm sure it is a common enough algorithm.</p>

    <callback event="instantiate"><![CDATA[
      delay = calloc(DELAY_SIZE, sizeof(LADSPA_Data));
      rptr.all = 0;
      wptr = 0;
      last_size = -1;
      delay_mask = 0xFF;
      delay_ofs = 0x80;
      last_gain = 0.5f;
      count = 0;
      last_inc = 0.0f;
    ]]></callback>

    <callback event="cleanup"><![CDATA[
      free(plugin_data->delay);
    ]]></callback>

	<callback event="run"><![CDATA[
    
    	unsigned long pos;
    	
		if (enabled == 0)
		{
			//Send input direct to output (no processing)
            for (pos = 0; pos < sample_count; pos++)
            	buffer_write(output[pos], input[pos]); 
		}
		else
		{
    	    fixp16 om;
	        float gain = last_gain, gain_inc = last_inc;
          	unsigned int i;

			float p = pitch;

			if (cents != 0)
				p = pow(2, cents/1200); /* Convert cents to frequency multiplyer */

          	om.all = f_round(p * 65536.0f);

			if (size != last_size) {
    			int size_tmp = f_round(size);

        		if (size_tmp > 7) {
        	  		size_tmp = 5;
        		} else if (size_tmp < 1) {
        	  		size_tmp = 1;
        		}
        		plugin_data->last_size = size;

        		/* Calculate the ringbuf parameters, the magick constants will need
        	 	* to be changed if you change DELAY_SIZE */
        		delay_mask = (1 << (size_tmp + 6)) - 1;
        		delay_ofs = 1 << (size_tmp + 5);
          	}

		  for (pos = 0; pos < sample_count; pos++) {
			  float out = 0.0f;

			  if (count++ > 14) {
				  float tmp;
				  count = 0;
				  tmp = 0.5f * (float)((rptr.part.in - wptr + delay_ofs/2) &
					  delay_mask) / (float)delay_ofs;
				  tmp = sinf(M_PI * 2.0f * tmp) * 0.5f + 0.5f;
				  gain_inc = (tmp - gain) / 15.0f;
			  }
			  gain += gain_inc;

			  delay[wptr] = input[pos];

			  /* Add contributions from the two readpointers, scaled by thier
				  * distance from the write pointer */
			  i = rptr.part.in;
			  out += cube_interp((float)rptr.part.fr * 0.0000152587f,
                             delay[(i - 1) & delay_mask], delay[i],
                             delay[(i + 1) & delay_mask],
                             delay[(i + 2) & delay_mask]) * (1.0f - gain);
			  i += delay_ofs;
			  out += cube_interp((float)rptr.part.fr * 0.0000152587f,
                             delay[(i - 1) & delay_mask], delay[i & delay_mask],
                             delay[(i + 1) & delay_mask],
                             delay[(i + 2) & delay_mask]) * gain;
	
			  buffer_write(output[pos], out);

			  /* Increment ringbuffer pointers */
			  wptr = (wptr + 1) & delay_mask;
			  rptr.all += om.all;
			  rptr.part.in &= delay_mask;
          }

          plugin_data->rptr.all = rptr.all;
          plugin_data->wptr = wptr;
          plugin_data->delay_mask = delay_mask;
          plugin_data->delay_ofs = delay_ofs;
          plugin_data->last_gain = gain;
          plugin_data->count = count;
          plugin_data->last_inc = gain_inc;

          *(plugin_data->latency) = delay_ofs/2;
		}
    
    ]]></callback>

    <port label="enabled" dir="input" type="control" hint="toggled,default_1">
      <name>Enabled</name>
      <range min="0" max="1"/>
    </port>

    <port label="pitch" dir="input" type="control" hint="logarithmic,default_1">
      <name>Pitch shift (Frequency)</name>
      <p>The multiple of the output pitch, eg. 2.0 will increase the pitch by one octave.</p>
      <range min="0.25" max="4.0"/>
    </port>

    <port label="cents" dir="input" type="control" hint="integer,default_0">
      <name>Pitch shift (Cents)</name>
      <p>Pitch shift in cents. Overrides frequency control.</p>
      <range min="-100" max="100"/>
    </port>

    <port label="size" dir="input" type="control" hint="integer,default_middle">
      <name>Buffer size</name>
      <p>The order of magnitude of the buffer size. Small buffers will sound fluttery, large buffers will have flangy sounding echos.</p>
      <p>I recommend a buffer size of 3 for a reasonable compromise, with wideband material at around 48KHz. For drums you might have to lower it, and for voiced background noises it can go higher.</p>
      <range min="1" max="7"/>
    </port>

    <port label="input" dir="input" type="audio">
      <name>Input</name>
    </port>

    <port label="output" dir="output" type="audio">
      <name>Output</name>
    </port>

    <port label="latency" dir="output" type="control">
      <name>latency</name>
    </port>

    <instance-data label="delay" type="LADSPA_Data *" />
    <instance-data label="rptr" type="fixp16" />
    <instance-data label="wptr" type="unsigned int" />
    <instance-data label="last_size" type="int" />
    <instance-data label="delay_mask" type="unsigned int" />
    <instance-data label="delay_ofs" type="unsigned int" />
    <instance-data label="last_gain" type="float" />
    <instance-data label="last_inc" type="float" />
    <instance-data label="count" type="unsigned int" />
  </plugin>
</ladspa>