react-native-simple-note-pitch-detector 0.7.0 → 0.7.1

package/android/src/main/java/expo/modules/simplenotepitchdetector/PitchAnalyzer.kt

@@ -2,15 +2,19 @@ package expo.modules.simplepitchdetector
 
 import android.util.Log
 import be.tarsos.dsp.AudioDispatcher
+import be.tarsos.dsp.AudioEvent
 import be.tarsos.dsp.AudioProcessor
 import be.tarsos.dsp.io.android.AudioDispatcherFactory
 import be.tarsos.dsp.pitch.PitchDetectionHandler
+import be.tarsos.dsp.pitch.PitchDetectionResult
 import be.tarsos.dsp.pitch.PitchProcessor
 import be.tarsos.dsp.pitch.PitchProcessor.PitchEstimationAlgorithm
+import be.tarsos.dsp.util.fft.FFT
 import kotlin.math.log2
 import kotlin.math.round
-import kotlin.math.pow
-import kotlin.math.abs
+import kotlin.math.ln
+import kotlin.math.cos
+import kotlin.math.PI
 
 private const val TAG = "PitchAnalyzer"
 
@@ -22,6 +26,14 @@ data class PitchData(
     val offset: Float
 )
 
+/**
+ * Dual-dispatcher pitch detector:
+ * - HPS (Harmonic Product Spectrum, h=2,3,4,5) on its own AudioDispatcher for low notes (<200Hz)
+ * - MPM (McLeod Pitch Method) on a separate AudioDispatcher for mid/high notes (>=200Hz)
+ *
+ * Two completely separate microphone streams to avoid any shared state or buffer corruption.
+ * Previous experiments proved that ANY combination of HPS+MPM in the same dispatcher degrades HPS.
+ */
 class PitchAnalyzer {
 
     private val notes = arrayOf("C","C#","D","D#","E","F","F#","G","G#","A","A#","B")
@@ -30,74 +42,200 @@ class PitchAnalyzer {
     private var onStatus: ((String, String) -> Unit)? = null
     private var isRecording = false
     private var levelThreshold = -30f
-    // Configurable buffer size - can be changed from JS
-    // Larger buffer = better low frequency detection, more latency
-    // Smaller buffer = better high frequency detection, less latency
-    private var bufferSize = 2048
-    // Configurable algorithm - can be changed from JS
-    private var algorithm = PitchEstimationAlgorithm.MPM
-    private var algorithmName = "mpm"
-
-    private var dispatcher: AudioDispatcher? = null
-    private var processor: AudioProcessor? = null
-    private var runner: Thread? = null
-
-    private val handler = PitchDetectionHandler { res, e ->
-        val pitchInHz = res.pitch
-        val decibel = e.getdBSPL().toFloat()
-        // Only process if above the level threshold
-        if (decibel > levelThreshold) {
-            // Note: Not logging every pitch detection to JS to avoid flooding
-            // Use adb logcat for verbose pitch logs
-            Log.v(TAG, "Pitch detected: ${pitchInHz}Hz at ${decibel}dB")
-            process(pitchInHz, decibel)
+    private var sampleRate = 44100
+    private var bufferSize = 8192
+    private var algorithmName = "hps"
+
+    // HPS dispatcher (low notes)
+    private var hpsDispatcher: AudioDispatcher? = null
+    private var hpsThread: Thread? = null
+
+    // MPM dispatcher (mid/high notes)
+    private var mpmDispatcher: AudioDispatcher? = null
+    private var mpmThread: Thread? = null
+
+    // HPS pre-allocated buffers
+    private var fftSize = 0
+    private var freqResolution = 0f
+    private var hpsFft: FFT? = null
+    private var fftBuffer: FloatArray? = null
+    private var magnitudes: FloatArray? = null
+    private var hpsProduct: FloatArray? = null
+    private var hannWindow: FloatArray? = null
+
+    // HPS searches low range only: A0 (27.5Hz) to ~G3 (196Hz)
+    private val hpsMinHz = 26f
+    private val hpsMaxHz = 200f
+
+    // MPM crossover: only emit MPM results >= this frequency
+    private val mpmMinHz = 200f
+    // MPM max: reject garbage above piano range (C8 = 4186Hz)
+    private val mpmMaxHz = 4200f
+
+    private val hpsProcessor = object : AudioProcessor {
+        override fun process(audioEvent: AudioEvent): Boolean {
+            val buffer = audioEvent.floatBuffer
+            val dB = audioEvent.getdBSPL().toFloat()
+
+            if (dB > levelThreshold) {
+                try {
+                    val pitch = detectPitchHPS(buffer)
+                    if (pitch > 0f) {
+                        emitPitch(pitch, dB)
+                    }
+                } catch (ex: Exception) {
+                    Log.e(TAG, "HPS failed: ${ex.message}")
+                }
+            }
+            return true
         }
+
+        override fun processingFinished() {}
     }
 
-    private fun prepare() {
-        try {
-            logStatus("debug", "prepare() called with bufferSize=$bufferSize, algorithm=$algorithmName")
-            processor = PitchProcessor(algorithm, 22050f, bufferSize, handler)
-            logStatus("debug", "PitchProcessor created successfully")
-            dispatcher = AudioDispatcherFactory.fromDefaultMicrophone(22050, bufferSize, 0)
-            logStatus("debug", "AudioDispatcher created successfully: $dispatcher")
-            dispatcher?.addAudioProcessor(processor)
-            logStatus("debug", "AudioProcessor added to dispatcher")
-        } catch (e: Exception) {
-            logStatus("error", "Error in prepare(): ${e.message}")
-            throw e
+    private fun detectPitchHPS(buffer: FloatArray): Float {
+        val fft = hpsFft ?: return -1f
+        val fftBuf = fftBuffer ?: return -1f
+        val mags = magnitudes ?: return -1f
+        val hps = hpsProduct ?: return -1f
+        val window = hannWindow ?: return -1f
+
+        // Window and zero-pad
+        for (i in buffer.indices) {
+            fftBuf[i] = buffer[i] * window[i]
+        }
+        for (i in buffer.size until fftSize) {
+            fftBuf[i] = 0f
+        }
+
+        // FFT
+        fft.forwardTransform(fftBuf)
+        fft.modulus(fftBuf, mags)
+
+        // HPS: multiply magnitudes at harmonics 2, 3, 4, 5
+        val hpsSize = mags.size / 5
+        for (i in 0 until hpsSize) {
+            hps[i] = mags[2 * i] * mags[3 * i] * mags[4 * i] * mags[5 * i]
+        }
+
+        // Search for peak in low range only
+        val minBin = (hpsMinHz / freqResolution).toInt().coerceAtLeast(1)
+        val maxBin = (hpsMaxHz / freqResolution).toInt().coerceAtMost(hpsSize - 1)
+
+        var peakBin = minBin
+        var peakVal = hps[minBin]
+        for (i in minBin + 1..maxBin) {
+            if (hps[i] > peakVal) { peakBin = i; peakVal = hps[i] }
+        }
+
+        // SNR check
+        val snrStart = (peakBin - 200).coerceAtLeast(minBin)
+        val snrEnd = (peakBin + 200).coerceAtMost(maxBin)
+        var sum = 0.0
+        for (i in snrStart..snrEnd) { sum += hps[i].toDouble() }
+        val mean = sum / (snrEnd - snrStart + 1)
+        val snr = if (mean > 0.0) peakVal.toDouble() / mean else 0.0
+        if (snr < 12.0) return -1f
+
+        return interpolatePeak(hps, peakBin, minBin, maxBin) * freqResolution
+    }
+
+    private fun interpolatePeak(spectrum: FloatArray, peakBin: Int, minBin: Int, maxBin: Int): Float {
+        if (peakBin <= minBin || peakBin >= maxBin) return peakBin.toFloat()
+        val prev = spectrum[peakBin - 1].coerceAtLeast(1e-30f)
+        val curr = spectrum[peakBin].coerceAtLeast(1e-30f)
+        val next = spectrum[peakBin + 1].coerceAtLeast(1e-30f)
+        val logPrev = ln(prev)
+        val logCurr = ln(curr)
+        val logNext = ln(next)
+        val denom = logPrev - 2f * logCurr + logNext
+        return if (denom != 0f) {
+            peakBin + 0.5f * (logPrev - logNext) / denom
+        } else {
+            peakBin.toFloat()
+        }
+    }
+
+    private fun prepareHPS() {
+        val overlap = (bufferSize * 3) / 4
+        logStatus("debug", "prepareHPS() sampleRate=$sampleRate, bufferSize=$bufferSize, overlap=$overlap")
+
+        // 4x zero-padding for frequency resolution
+        fftSize = bufferSize * 4
+        freqResolution = sampleRate.toFloat() / fftSize
+        hpsFft = FFT(fftSize)
+        fftBuffer = FloatArray(fftSize)
+        magnitudes = FloatArray(fftSize / 2)
+        hpsProduct = FloatArray(fftSize / (2 * 5))
+        hannWindow = FloatArray(bufferSize) { i ->
+            (0.5 * (1.0 - cos(2.0 * PI * i / (bufferSize - 1)))).toFloat()
         }
+
+        val maxHpsHz = (fftSize / (2 * 5)) * freqResolution
+        logStatus("debug", "HPS: fftSize=$fftSize, freqRes=${freqResolution}Hz/bin, search=${hpsMinHz}-${hpsMaxHz}Hz, maxHpsHz=$maxHpsHz")
+
+        hpsDispatcher = AudioDispatcherFactory.fromDefaultMicrophone(sampleRate, bufferSize, overlap)
+        hpsDispatcher?.addAudioProcessor(hpsProcessor)
+
+        logStatus("debug", "HPS dispatcher prepared (low notes <${hpsMaxHz}Hz)")
     }
 
-    private fun process(pitchInHz: Float, decibel: Float) {
-        if (pitchInHz <= 0 || pitchInHz.isNaN()) {
-            return
+    private fun prepareMPM() {
+        val mpmBufferSize = 2048
+        val mpmOverlap = (mpmBufferSize * 3) / 4
+        logStatus("debug", "prepareMPM() sampleRate=$sampleRate, bufferSize=$mpmBufferSize, overlap=$mpmOverlap")
+
+        val pitchHandler = PitchDetectionHandler { result: PitchDetectionResult, event: AudioEvent ->
+            val pitch = result.pitch
+            if (pitch > 0 && pitch >= mpmMinHz && pitch <= mpmMaxHz && result.isPitched) {
+                val dB = event.getdBSPL().toFloat()
+                emitPitch(pitch, dB)
+            }
         }
 
-        // Calculate MIDI note number (A4 = 440Hz = MIDI 69)
+        // Use YIN instead of MPM — McLeodPitchMethod.peakPicking throws AssertionError
+        // when two AudioRecord instances compete for the microphone
+        val yinPitchProcessor = PitchProcessor(PitchEstimationAlgorithm.YIN, sampleRate.toFloat(), mpmBufferSize, pitchHandler)
+        val safeYinProcessor = object : AudioProcessor {
+            override fun process(audioEvent: AudioEvent): Boolean {
+                try {
+                    return yinPitchProcessor.process(audioEvent)
+                } catch (e: Exception) {
+                    Log.w(TAG, "YIN error (ignored): ${e.message}")
+                    return true
+                }
+            }
+            override fun processingFinished() {
+                yinPitchProcessor.processingFinished()
+            }
+        }
+
+        mpmDispatcher = AudioDispatcherFactory.fromDefaultMicrophone(sampleRate, mpmBufferSize, mpmOverlap)
+        mpmDispatcher?.addAudioProcessor(safeYinProcessor)
+
+        logStatus("debug", "YIN dispatcher prepared (mid/high notes ${mpmMinHz}-${mpmMaxHz}Hz)")
+    }
+
+    private fun emitPitch(pitchInHz: Float, decibel: Float) {
+        if (pitchInHz <= 0 || pitchInHz.isNaN()) return
+
         val midiNote = 12 * log2(pitchInHz / 440f) + 69
         val roundedMidiNote = round(midiNote).toInt()
-
-        // Calculate note index (0-11) and octave
         val noteIndex = ((roundedMidiNote % 12) + 12) % 12
         val octave = (roundedMidiNote / 12) - 1
-
-        // Calculate offset from perfect pitch (in cents, then convert to percentage)
-        // 100 cents = 1 semitone, so we express as percentage of a semitone
         val centsOff = (midiNote - roundedMidiNote) * 100
-        val offsetPercentage = centsOff // Already in a reasonable range (-50 to +50)
 
-        val note = notes[noteIndex]
+        // Pre-filter: JS discards offset > 25% anyway, so skip detections > 40%
+        // to reduce noise in the JS majority voting window
+        if (kotlin.math.abs(centsOff) > 40f) return
 
-        val pitchData = PitchData(
-            note = note,
+        onPitchDetected(PitchData(
+            note = notes[noteIndex],
             octave = octave,
             frequency = pitchInHz,
             amplitude = decibel,
-            offset = offsetPercentage
-        )
-
-        onPitchDetected(pitchData)
+            offset = centsOff
+        ))
     }
 
     fun setOnPitchDetectedListener(listener: (PitchData) -> Unit) {
@@ -122,53 +260,63 @@ class PitchAnalyzer {
         this.levelThreshold = threshold
     }
 
+    fun setSampleRate(rate: Int) {
+        this.sampleRate = rate
+    }
+
+    fun getSampleRate(): Int = this.sampleRate
+
     fun setBufferSize(size: Int) {
         this.bufferSize = size
     }
 
-    fun getBufferSize(): Int {
-        return this.bufferSize
-    }
+    fun getBufferSize(): Int = this.bufferSize
 
     fun setAlgorithm(name: String) {
         this.algorithmName = name.lowercase()
-        this.algorithm = when (this.algorithmName) {
-            "yin" -> PitchEstimationAlgorithm.YIN
-            "fft_yin" -> PitchEstimationAlgorithm.FFT_YIN
-            "mpm" -> PitchEstimationAlgorithm.MPM
-            "fft_pitch" -> PitchEstimationAlgorithm.FFT_PITCH
-            "dynamic_wavelet" -> PitchEstimationAlgorithm.DYNAMIC_WAVELET
-            "amdf" -> PitchEstimationAlgorithm.AMDF
-            else -> PitchEstimationAlgorithm.FFT_YIN // Default
-        }
     }
 
-    fun getAlgorithm(): String {
-        return this.algorithmName
-    }
+    fun getAlgorithm(): String = this.algorithmName
 
     fun start() {
         try {
-            logStatus("debug", "start() called")
-            prepare()
-            runner = Thread(dispatcher)
-            logStatus("debug", "Thread created, starting...")
-            runner?.start()
+            logStatus("debug", "start() called — dual dispatcher mode (HPS + MPM)")
+
+            // Start HPS dispatcher first
+            prepareHPS()
+            hpsThread = Thread(hpsDispatcher)
+            hpsThread?.name = "HPS-Thread"
+            hpsThread?.start()
+            logStatus("debug", "HPS thread started")
+
+            // Start MPM dispatcher second
+            prepareMPM()
+            mpmThread = Thread(mpmDispatcher)
+            mpmThread?.name = "MPM-Thread"
+            mpmThread?.start()
+            logStatus("debug", "MPM thread started")
+
             isRecording = true
-            logStatus("debug", "Recording started successfully")
+            logStatus("debug", "Dual dispatcher recording started successfully")
         } catch (e: Exception) {
             logStatus("error", "Error in start(): ${e.message}")
+            // Clean up whatever was started
+            stop()
             isRecording = false
         }
     }
 
     fun stop() {
-        dispatcher?.stop()
-        runner?.interrupt()
+        hpsDispatcher?.stop()
+        mpmDispatcher?.stop()
+        hpsThread?.interrupt()
+        mpmThread?.interrupt()
+        hpsDispatcher = null
+        mpmDispatcher = null
+        hpsThread = null
+        mpmThread = null
         isRecording = false
     }
 
-    fun isRecording(): Boolean {
-        return isRecording
-    }
-}
+    fun isRecording(): Boolean = isRecording
+}

package/android/src/main/java/expo/modules/simplenotepitchdetector/ReactNativeSimpleNotePitchDetectorModule.kt

@@ -58,6 +58,14 @@ class ReactNativeSimpleNotePitchDetectorModule : Module() {
       pitchAnalyzer.setLevelThreshold(threshold.toFloat())
     }
 
+    Function("setSampleRate") { rate: Int ->
+      pitchAnalyzer.setSampleRate(rate)
+    }
+
+    Function("getSampleRate") {
+      pitchAnalyzer.getSampleRate()
+    }
+
     // Allow JS to configure the buffer size
     // Must be called before start() to take effect
     // Common values: 1024 (better for high frequencies), 2048 (balanced), 4096 (better for low frequencies)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "react-native-simple-note-pitch-detector",
-  "version": "0.7.0",
+  "version": "0.7.1",
   "description": "a simple react native library to detect the pitch of the input recording",
   "main": "build/index.js",
   "types": "build/index.d.ts",