update- see revision.md

AudioTuner.cpp

 
 #include "AudioTuner.h"
 #include "utility/dspinst.h"
+#include "arm_math.h"
 
-#if SAMPLE_RATE == SAMPLE_RATE_44100
-    #define SAMPLE_RATE_EXACT AUDIO_SAMPLE_RATE_EXACT / 1
-#elif SAMPLE_RATE == SAMPLE_RATE_22050
-    #define SAMPLE_RATE_EXACT AUDIO_SAMPLE_RATE_EXACT / 2
-#elif SAMPLE_RATE == SAMPLE_RATE_11025
-    #define SAMPLE_RATE_EXACT AUDIO_SAMPLE_RATE_EXACT / 4
-#endif
-
-#define HALF_BUFFER NUM_SAMPLES / 2
+#define HALF_BLOCKS AUDIO_BLOCKS * 64
 
 #define LOOP1(a)  a
 #define LOOP2(a)  a LOOP1(a)
 #define LOOP3(a)  a LOOP2(a)
+#define LOOP4(a)  a LOOP3(a)
 #define LOOP8(a)  a LOOP3(a) a LOOP3(a)
+#define LOOP16(a) a LOOP8(a) a LOOP2(a) a LOOP3(a)
+#define LOOP32(a)  a LOOP16(a) a LOOP8(a) a LOOP1(a) a LOOP3(a)
+#define LOOP64(a)  a LOOP32(a) a LOOP16(a) a LOOP8(a) a LOOP2(a) a LOOP1(a)
 #define UNROLL(n,a) LOOP##n(a)
 
-/**
- *  Audio update function.
- */
+static void copy_buffer(void *destination, const void *source) {
+    const uint16_t *src = (const uint16_t *)source;
+    uint16_t *dst = (uint16_t *)destination;
+    for (int i=0; i < AUDIO_BLOCK_SAMPLES; i++) *dst++ = *src++;
+}
+
 void AudioTuner::update( void ) {
+    
     audio_block_t *block;
-    const int16_t *p, *end;
-    block = receiveReadOnly( );
     
-    if ( !block ) return;
+    block = receiveReadOnly();
+    if (!block) return;
     
     if ( !enabled ) {
         release( block );
         return;
     }
     
-    p = block->data;
-    end = p + AUDIO_BLOCK_SAMPLES;
-    
-    /*
-     * Double buffering, one fills while the other is processed
-     * 2x the throughput.
-    */
-    uint16_t *dst;
-    bool next = next_buffer;
-    if ( next ) {
-        //digitalWriteFast(6, HIGH);
-        dst = ( uint16_t * )buffer;
+    digitalWriteFast(2, HIGH);
+    if ( next_buffer ) {
+        blocklist1[state++] = block;
+        if ( !first_run && process_buffer ) process( );
+    } else {
+        blocklist2[state++] = block;
+        if ( !first_run && process_buffer ) process( );
     }
-    else {
-       //digitalWriteFast(6, LOW);
-       dst = ( uint16_t * )buffer + NUM_SAMPLES;
+    
+    if ( state >= AUDIO_BLOCKS ) {
+        if ( next_buffer ) {
+            if ( !first_run && process_buffer ) process( );
+            for ( int i = 0; i < AUDIO_BLOCKS; i++ ) copy_buffer( AudioBuffer+( i * 0x80 ), blocklist1[i]->data );
+            for ( int i = 0; i < AUDIO_BLOCKS; i++ ) release(blocklist1[i] );
+        } else {
+            if ( !first_run && process_buffer ) process( );
+            for ( int i = 0; i < AUDIO_BLOCKS; i++ ) copy_buffer( AudioBuffer+( i * 0x80 ), blocklist2[i]->data );
+            for ( int i = 0; i < AUDIO_BLOCKS; i++ ) release( blocklist2[i] );
+        }
+        process_buffer = true;
+        first_run = false;
+        state = 0;
+        //digitalWriteFast(LED_BUILTIN, !digitalReadFast(LED_BUILTIN));
     }
+}
+
+FASTRUN void AudioTuner::process( void ) {
+    //digitalWriteFast(0, HIGH);
     
-    // gather data/and release block
-    uint16_t count = count_global;
+    const int16_t *p;
+    p = AudioBuffer;
+    
+    uint16_t cycles = 64;;
+    uint16_t tau = tau_global;
     do {
-        *( dst+count++ ) = *( uint16_t * )p;
-        p += SAMPLE_RATE;
-    } while ( p < end );
-    release( block );
+        uint16_t x   = 0;
+        int64_t  sum = 0;
+        //uint32_t res;
+        do {
+            /*int16_t current1, lag1, current2, lag2;
+             int32_t val1, val2;
+             lag1 = *( ( uint32_t * )p + ( x + tau ) );
+             current1 = *( ( uint32_t * )p + x );
+             x += 32;
+             lag2 = *( ( uint32_t * )p + ( x + tau ) );
+             current2 = *( ( uint32_t * )p + x );
+             val1 = __PKHBT(current1, current2, 0x10);
+             val2 = __PKHBT(lag1, lag2, 0x10);
+             res = __SSUB16( val1, val2 );
+             sum = __SMLALD(res, res, sum);
+             //sum = __SMLSLD(delta1, delta2, sum);*/
+            int16_t current, lag, delta;
+            //UNROLL(16,
+                   lag = *( ( int16_t * )p + ( x+tau ) );
+                   current = *( ( int16_t * )p+x );
+                   delta = ( current-lag );
+                   sum += delta * delta;
+#if F_CPU == 144000000
+                   x += 8;
+#elif F_CPU == 120000000
+                   x += 12;
+#elif F_CPU == 96000000
+                   x += 16;
+#elif F_CPU < 96000000
+                   x += 32;
+#endif
+                   //);
+        } while ( x <= HALF_BLOCKS );
+
+        running_sum += sum;
+        yin_buffer[yin_idx] = sum*tau;
+        rs_buffer[yin_idx] = running_sum;
+        yin_idx = ( ++yin_idx >= 5 ) ? 0 : yin_idx;
+        tau = estimate( yin_buffer, rs_buffer, yin_idx, tau );
+
+        if ( tau == 0 ) {
+            process_buffer  = false;
+            new_output      = true;
+            yin_idx         = 1;
+            running_sum     = 0;
+            tau_global      = 1;
+            //digitalWriteFast(2, LOW);
+            //digitalWriteFast(0, LOW);
+            return;
+        }
+    } while ( --cycles );
     
-    /* 
-     * If buffer full switch to start filling next
-     * buffer and process the just filled buffer.
-     */
-    if ( count >= NUM_SAMPLES ) {
-        //digitalWriteFast(2, !digitalReadFast(2));
-        __disable_irq();
-        next_buffer = !next_buffer;
-        process_buffer  = true;
-        count_global    = 0;
-        tau_global      = 1;
+    if ( tau >= HALF_BLOCKS ) {
+        process_buffer  = false;
+        new_output      = false;
         yin_idx         = 1;
         running_sum     = 0;
-        count           = 0;
-        __enable_irq();
-    }
-    count_global = count;// update global count
-    
-    /*
-     * Set the number of cycles to be processed per receiving block.
-     */
-    uint16_t cycles;
-    const uint16_t usage_max = cpu_usage_max;
-    if ( AudioProcessorUsage( ) > usage_max ) {
-#if NUM_SAMPLES >= 8192
-        cycles = tau_global + 2;
-#elif NUM_SAMPLES == 4096
-        cycles = tau_global + 4;
-#elif NUM_SAMPLES == 2048
-        cycles = tau_global + 8;
-#elif NUM_SAMPLES <= 1024
-        cycles = tau_global + 32;
-#endif
-    }
-    else {
-#if NUM_SAMPLES >= 8192
-        cycles = tau_global + 8;
-#elif NUM_SAMPLES == 4096
-        cycles = tau_global + 16;
-#elif NUM_SAMPLES == 2048
-        cycles = tau_global + 32;
-#elif NUM_SAMPLES <= 1024
-        cycles = tau_global + 64;
-#endif
-    }
-    
-    if ( process_buffer ) {
-        //digitalWriteFast(0, HIGH);
-        uint16_t tau;
-        next = next_buffer;
-        tau = tau_global;
-        do {
-            int64_t sum  = 0;
-            const int16_t *end, *buf;
-            if ( next ) {
-                //digitalWriteFast(4, LOW);
-                buf = buffer + NUM_SAMPLES;
-            }
-            else {
-                //digitalWriteFast(4, HIGH);
-                buf = buffer;
-            }
-            end = buf + HALF_BUFFER;
-            
-            // TODO: How to make faster?
-            do {
-                int16_t current, lag, delta;
-                UNROLL( 8,
-                       lag = *( buf + tau );
-                       current = *buf++;
-                       delta = current - lag;
-                       //sum = multiply_accumulate_32x32_rshift32_rounded(sum, delta, delta);
-                       sum += delta*delta;
-                       );
-            } while ( buf < end );
-            
-            running_sum += sum;
-            yin_buffer[yin_idx] = sum*tau;
-            rs_buffer[yin_idx] = running_sum;
-            yin_idx = ( ++yin_idx >= 5 ) ? 0 : yin_idx;
-            
-            tau = estimate( yin_buffer, rs_buffer, yin_idx, tau );
-            
-            if ( tau == 0 ) {
-                process_buffer  = false;
-                new_output      = true;
-                //digitalWriteFast(0, LOW);
-                return;
-            }
-            else if ( tau >= HALF_BUFFER ) {
-                process_buffer  = false;
-                new_output      = false;
-                //digitalWriteFast(0, LOW);
-                return;
-            }
-            
-        } while ( tau <= cycles );
-        tau_global = tau;
+        tau_global      = 1;
         //digitalWriteFast(0, LOW);
+        return;
     }
+    tau_global = tau;
+    //digitalWriteFast(0, LOW);
 }
 
 /**
  *  @return tau
  */
 uint16_t AudioTuner::estimate( int64_t *yin, int64_t *rs, uint16_t head, uint16_t tau ) {
-    const int64_t *p = ( int64_t * )yin;
+    const int64_t *y = ( int64_t * )yin;
     const int64_t *r = ( int64_t * )rs;
     uint16_t _tau, _head;
+    const float thresh = yin_threshold;
     _tau = tau;
     _head = head;
     
         idx2 = ( idx2 >= 5 ) ? 0 : idx2;
         
         float s0, s1, s2;
-        s0 = ( ( float )*( p+idx0 ) / *( r+idx0 ) );
-        s1 = ( ( float )*( p+idx1 ) / *( r+idx1 ) );
-        s2 = ( ( float )*( p+idx2 ) / *( r+idx2 ) );
+        s0 = ( ( float )*( y+idx0 ) / *( r+idx0 ) );
+        s1 = ( ( float )*( y+idx1 ) / *( r+idx1 ) );
+        s2 = ( ( float )*( y+idx2 ) / *( r+idx2 ) );
         
-        if ( s1 < yin_threshold && s1 < s2 ) {
+        if ( s1 < thresh && s1 < s2 ) {
             uint16_t period = _tau - 3;
             periodicity = 1 - s1;
             data = period + 0.5f * ( s0 - s2 ) / ( s0 - 2.0f * s1 + s2 );
             return 0;
         }
-        
-        //if ( s1 > 2.4 ) return _tau + 2;
-        //else return _tau + 1;
     }
     return _tau + 1;
 }
  *  @param threshold Allowed uncertainty
  *  @param cpu_max   How much cpu usage before throttling
  */
-void AudioTuner::initialize( float threshold, float cpu_max ) {
+void AudioTuner::initialize( float threshold ) {
     __disable_irq( );
-    cpu_usage_max = cpu_max*100;
-    yin_threshold = threshold;
     process_buffer = false;
+    yin_threshold  = threshold;
     periodicity    = 0.0f;
     next_buffer    = true;
     running_sum    = 0;
-    count_global   = 0;
+    tau_global     = 1;
+    first_run      = true;
     yin_idx        = 1;
-    data           = 0;
     enabled        = true;
+    state          = 0;
+    data           = 0.0f;
     __enable_irq( );
 }
 
     __disable_irq( );
     float d = data;
     __enable_irq( );
-    return SAMPLE_RATE_EXACT / d;
+    return AUDIO_SAMPLE_RATE_EXACT / d;
 }
 
 /**

AudioTuner.h

 #define AudioTuner_h_
 
 #include "AudioStream.h"
-/****************************************************************/
-#define SAMPLE_RATE_44100  1      // 44100    sample rate
-#define SAMPLE_RATE_22050  2      // 22050    sample rate
-#define SAMPLE_RATE_11025  4      // 11025    sample rate
-/****************************************************************/
-
 /****************************************************************
  *              Safe to adjust these values below               *
  *                                                              *
- *  These two parameters define how this object works.          *
+ *  This parameter defines the size of the buffer.              *
  *                                                              *
- *  1.  NUM_SAMPLES - Size of the buffer. Since object uses     *
- *      double buffering this value will be 4x in bytes of      *
- *      memory.  !!! Must be power of 2 !!!!                    *
+ *  1.  AUDIO_BLOCKS -  Buffer size is 128 * AUDIO_BLOCKS.      *
+ *                      The more AUDIO_BLOCKS the lower the     *
+ *                      frequency you can detect. The defualt   *
+ *                      (24) is set to measure down to 29.14    *
+ *                      Hz or B(flat)0.                         *
  *                                                              *
- *  2.  SAMPLE_RATE - Just what it says.                        *
- *                                                              *
- *  These two parameters work hand in hand. For example if you  *
- *  want a high sample rate but do not allocate enough buffer   *
- *  space, you will be limit how low of a frequency you can     *
- *  measure. If you then increase the buffer you use up         *
- *  precious ram and slow down the system since it takes longer *
- *  to processes the buffer.                                    *
- *                                                              *
- *  Play around with these values to find what best suits your  *
- *  needs. The max number of buffers you can have is 8192 bins. *
  ****************************************************************/
-// !!! Must be power of 2 !!!!
-#define NUM_SAMPLES 2048 // make a power of two
-
-// Use defined sample rates above^
-#define SAMPLE_RATE SAMPLE_RATE_22050
+#define AUDIO_BLOCKS  24
 /****************************************************************/
-
-class AudioTuner : public AudioStream
-{
+class AudioTuner : public AudioStream {
 public:
     /**
      *  constructor to setup Audio Library and initialize
      *
      *  @return none
      */
-    AudioTuner( void ) : AudioStream( 1, inputQueueArray ), enabled( false ), new_output(false) {}
+    AudioTuner( void ) : AudioStream( 1, inputQueueArray ), enabled( false ), new_output(false) {
+    
+    }
     
     /**
      *  initialize variables and start conversion
      *
      *  @param threshold Allowed uncertainty
      *  @param cpu_max   How much cpu usage before throttling
+     *
+     *  @return none
      */
-    void initialize( float threshold, float cpu_max);
+    void initialize( float threshold );
     
     /**
      *  sets threshold value
      *
      *  @param thresh
+     *  @return none
      */
     void threshold( float p );
     
     
     /**
      *  Audio Library calls this update function ~2.9ms
+     *
+     *  @return none
      */
     virtual void update( void );
-    
+
 private:
     /**
      *  check the sampled data for fundamental frequency
      */
     uint16_t estimate( int64_t *yin, int64_t *rs, uint16_t head, uint16_t tau );
     
-    int16_t  buffer[NUM_SAMPLES*2] __attribute__ ( ( aligned ( 4 ) ) );
-    float    periodicity, yin_threshold, data, cpu_usage_max;
-    int64_t  rs_buffer[5], yin_buffer[5];
+    /**
+     *  process audio data
+     *
+     *  @return none
+     */
+    void process( void );
+    
+    /**
+     *  Variables
+     */
     uint64_t running_sum;
-    uint16_t tau_global, count_global, tau_cycles;
-    uint8_t  yin_idx;
-    bool     enabled, process_buffer, next_buffer;
-    volatile bool new_output;
+    uint16_t tau_global;
+    int64_t  rs_buffer[5], yin_buffer[5];
+    int16_t  AudioBuffer[AUDIO_BLOCKS*128] __attribute__ ( ( aligned ( 4 ) ) );
+    uint8_t  yin_idx, state;
+    float    periodicity, yin_threshold, cpu_usage_max, data;
+    bool     enabled, next_buffer, first_run;
+    volatile bool new_output, process_buffer;
+    audio_block_t *blocklist1[AUDIO_BLOCKS];
+    audio_block_t *blocklist2[AUDIO_BLOCKS];
     audio_block_t *inputQueueArray[1];
 };
 #endif

README.md

 <p align="center">
-    <b>Guitar and Bass Tuner Library v2.2</b><br>
+    <b>Guitar and Bass Tuner Library v2.3</b><br>
     <b>Teensy 3.1/2</b><br>
 </p>
 
                                                        *---<\ P /
                                                              \_/
 
->Many optimizations have been done to the [YIN] algorithm for frequencies between 29-360Hz. 
+>Many optimizations have been done to the [YIN] algorithm for frequencies between 29-400Hz. 
 >>While its still using a brute force method ( n<sup>2</sup> ) for finding the fundamental frequency f<sub>o</sub>, it is tuned to skip certain <b>tau</b> (<img src="http://latex.numberempire.com/render?%5Cinline%20%5Chuge%20%5Cmathbf%7B%5Ctau%7D&sig=845639da85c0dd8e2de679817b06639c"/></img>) values and focus mostly on frequencies found in the bass and guitar. 
 >>>The input is double buffered so while you are processing one buffer it is filling the other to double throughput. 
->>>>There are a few parameters that can be adjusted to "dial in" the algorithm for better estimations located in AudioTuner.h. The defaults below are what I found that have the best trade off for speed and accuracy.
+>>>>The parameter AUDIO_BLOCKS below can be adjusted but its default of 24 I found to be best to work with the guitar and bass frequency range (29- 400)Hz. 
+>>>>Looking into finding the Auto Correlation using FFT and IFFT to speed up processing of data! Not that simple because the YIN algorithm uses a squared difference tweak to the Auto Correlation.
 
 <h4>AudioTuner.h</h4>
 
 ```
-/****************************************************************/
-#define SAMPLE_RATE_44100  1      // 44100    sample rate
-#define SAMPLE_RATE_22050  2      // 22050    sample rate
-#define SAMPLE_RATE_11025  4      // 11025    sample rate
-/****************************************************************/
-
 /****************************************************************
 *              Safe to adjust these values below               *
 *                                                              *
-*  These two parameters define how this object works.          *
+*  This parameter defines the size of the buffer.              *
 *                                                              *
-*  1.  NUM_SAMPLES - Size of the buffer. Since object uses     *
-*      double buffering this value will be 4x in bytes of      *
-*      memory.  !!! Must be power of 2 !!!!                    *
+*  1.  AUDIO_BLOCKS -  Buffer size is 128 * AUDIO_BLOCKS.      *
+*                      The more AUDIO_BLOCKS the lower the     *
+*                      frequency you can detect. The default   *
+*                      (24) is set to measure down to 29.14    *
+*                      Hz or B(flat)0.                         *
 *                                                              *
-*  2.  SAMPLE_RATE - Just what it says.                        *
-*                                                              *
-*  These two parameters work hand in hand. For example if you  *
-*  want a high sample rate but do not allocate enough buffer   *
-*  space, you will be limit how low of a frequency you can     *
-*  measure. If you then increase the buffer you use up         *
-*  precious ram and slow down the system since it takes longer *
-*  to processes the buffer.                                    *
-*                                                              *
-*  Play around with these values to find what best suits your  *
-*  needs. The max number of buffers you can have is 8192 bins. *
 ****************************************************************/
-// !!! Must be power of 2 !!!!
-#define NUM_SAMPLES 2048 // make a power of two
-
-// Use defined sample rates above^
-#define SAMPLE_RATE SAMPLE_RATE_22050
+#define AUDIO_BLOCKS  24
 /****************************************************************/
 ```
 
 </ol>
 </div>
 
-[YIN]:http://recherche.ircam.fr/equipes/pcm/cheveign/pss/2002_JASA_YIN.pdf
+[YIN]:http://recherche.ircam.fr/equipes/pcm/cheveign/pss/2002_JASA_YIN.pdf
+[Teensy Audio Library]:http://www.pjrc.com/teensy/td_libs_Audio.html

examples/Sample_Guitar_Tunning_Notes/Sample_Guitar_Tunning_Notes.ino

 }
 //---------------------------------------------------------------------------------------
 void setup() {
-    AudioMemory(4);
+    AudioMemory(30);
     /*
      *  Initialize the yin algorithm's absolute
      *  threshold, this is good number.
-     *
-     *  Percent of overall current cpu usage used
-     *  before making the search algorithm less
-     *  aggressive (0.0 - 1.0).
      */
-    tuner.initialize(.15, .99);
+    tuner.initialize(.15);
     pinMode(LED_BUILTIN, OUTPUT);
     playNoteTimer.begin(playNote, 1000);
 }

examples/Simple_Sine/Simple_Sine.ino

 char buffer[10];
 
 void setup() {
-    AudioMemory(4);
+    AudioMemory(30);
     /*
      *  Initialize the yin algorithm's absolute
      *  threshold, this is good number.
-     *
-     *  Percent of overall current cpu usage used 
-     *  before making the search algorithm less
-     *  aggressive (0.0 - 1.0).
      */
-    tuner.initialize(.15, .99);
+    tuner.initialize(.15);
     
     sine.frequency(30.87);
     sine.amplitude(1);

library.properties

 name=AudioTuner
-version=2.2
+version=2.3
 author=Colin Duffy
 maintainer=Colin Duffy
 sentence=Yin algorithm

revision.md

+><b>Updated (11/23/15 v2.3)</b><br>
+* Totally new method to gather and process data, data is available after 24 Blocks of data have been collected (~69.6ms) for all frequencies.<br>
+* Double buffer to collect Audio data, while one collects the other buffer is processed.<br>
+
 ><b>Updated (10/12/15 v2.2)</b><br>
 * Fixed yin cpu usage throttling code in update function.<br>
 * Function initialize second param takes a float (0.0 - 1.0).<br>