Add AudioAnalyzeToneDetect

Audio.cpp

 	return out;
 }
 
+// computes (((int64_t)a[31:0] * (int64_t)b[31:0]) >> 32)
+static inline int32_t multiply_32x32_rshift32(int32_t a, int32_t b) __attribute__((always_inline));
+static inline int32_t multiply_32x32_rshift32(int32_t a, int32_t b)
+{
+	int32_t out;
+	asm volatile("smmul %0, %1, %2" : "=r" (out) : "r" (a), "r" (b));
+	return out;
+}
+
+// computes (((int64_t)a[31:0] * (int64_t)b[31:0] + 0x8000000) >> 32)
+static inline int32_t multiply_32x32_rshift32_rounded(int32_t a, int32_t b) __attribute__((always_inline));
+static inline int32_t multiply_32x32_rshift32_rounded(int32_t a, int32_t b)
+{
+	int32_t out;
+	asm volatile("smmulr %0, %1, %2" : "=r" (out) : "r" (a), "r" (b));
+	return out;
+}
+
+// computes sum + (((int64_t)a[31:0] * (int64_t)b[31:0] + 0x8000000) >> 32)
+static inline int32_t multiply_accumulate_32x32_rshift32_rounded(int32_t sum, int32_t a, int32_t b) __attribute__((always_inline));
+static inline int32_t multiply_accumulate_32x32_rshift32_rounded(int32_t sum, int32_t a, int32_t b)
+{
+	int32_t out;
+	asm volatile("smmlar %0, %2, %3, %1" : "=r" (out) : "r" (sum), "r" (a), "r" (b));
+	return out;
+}
+
+// computes sum - (((int64_t)a[31:0] * (int64_t)b[31:0] + 0x8000000) >> 32)
+static inline int32_t multiply_subtract_32x32_rshift32_rounded(int32_t sum, int32_t a, int32_t b) __attribute__((always_inline));
+static inline int32_t multiply_subtract_32x32_rshift32_rounded(int32_t sum, int32_t a, int32_t b)
+{
+	int32_t out;
+	asm volatile("smmlsr %0, %2, %3, %1" : "=r" (out) : "r" (sum), "r" (a), "r" (b));
+	return out;
+}
 
 // computes ((a[15:0] << 16) | b[15:0])
 static inline uint32_t pack_16x16(int32_t a, int32_t b) __attribute__((always_inline));
 
 
 
+/******************************************************************/
+
+
+static inline int32_t multiply_32x32_rshift30(int32_t a, int32_t b) __attribute__((always_inline));
+static inline int32_t multiply_32x32_rshift30(int32_t a, int32_t b)
+{
+	return ((int64_t)a * (int64_t)b) >> 30;
+}
+
+//#define TONE_DETECT_FAST
+
+void AudioAnalyzeToneDetect::update(void)
+{
+	audio_block_t *block;
+	int32_t q0, q1, q2, coef;
+	const int16_t *p, *end;
+	uint16_t n;
+
+	block = receiveReadOnly();
+	if (!block) return;
+	if (!enabled) {
+		release(block);
+		return;
+	}
+	p = block->data;
+	end = p + AUDIO_BLOCK_SAMPLES;
+	n = count;
+	coef = coefficient;
+	q1 = s1;
+	q2 = s2;
+	do {
+		// the Goertzel algorithm is kinda magical ;-)
+#ifdef TONE_DETECT_FAST
+		q0 = (*p++) + (multiply_32x32_rshift32_rounded(coef, q1) << 2) - q2;
+#else
+		q0 = (*p++) + multiply_32x32_rshift30(coef, q1) - q2;
+		// TODO: is this only 1 cycle slower?  if so, always use it
+#endif
+		q2 = q1;
+		q1 = q0;
+		if (--n == 0) {
+			out1 = q1;
+			out2 = q2;
+			q1 = 0;  // TODO: does clearing these help or hinder?
+			q2 = 0;
+			new_output = true;
+			n = length;
+		}
+	} while (p < end);
+	count = n;
+	s1 = q1;
+	s2 = q2;
+	release(block);
+}
+
+void AudioAnalyzeToneDetect::set_params(int32_t coef, uint16_t cycles, uint16_t len)
+{
+	__disable_irq();
+	coefficient = coef;
+	ncycles = cycles;
+	length = len;
+	count = len;
+	s1 = 0;
+	s2 = 0;
+	enabled = true;
+	__enable_irq();
+	Serial.printf("Tone: coef=%d, ncycles=%d, length=%d\n", coefficient, ncycles, length);
+}
+
+float AudioAnalyzeToneDetect::read(void)
+{
+	int32_t coef, q1, q2, power;
+	uint16_t len;
+
+	__disable_irq();
+	coef = coefficient;
+	q1 = out1;
+	q2 = out2;
+	len = length;
+	__enable_irq();
+#ifdef TONE_DETECT_FAST
+	power = multiply_32x32_rshift32_rounded(q2, q2);
+	power = multiply_accumulate_32x32_rshift32_rounded(power, q1, q1);
+	power = multiply_subtract_32x32_rshift32_rounded(power,
+		multiply_32x32_rshift30(q1, q2), coef);
+	power <<= 4;
+#else
+	int64_t power64;
+	power64 = (int64_t)q2 * (int64_t)q2;
+	power64 += (int64_t)q1 * (int64_t)q1;
+	power64 -= (((int64_t)q1 * (int64_t)q2) >> 30) * (int64_t)coef;
+	power = power64 >> 28;
+#endif
+	return sqrtf((float)power) / (float)len;
+}
+
+
+AudioAnalyzeToneDetect::operator bool()
+{
+	int32_t coef, q1, q2, power, trigger;
+	uint16_t len;
+
+	__disable_irq();
+	coef = coefficient;
+	q1 = out1;
+	q2 = out2;
+	len = length;
+	__enable_irq();
+#ifdef TONE_DETECT_FAST
+	power = multiply_32x32_rshift32_rounded(q2, q2);
+	power = multiply_accumulate_32x32_rshift32_rounded(power, q1, q1);
+	power = multiply_subtract_32x32_rshift32_rounded(power,
+		multiply_32x32_rshift30(q1, q2), coef);
+	power <<= 4;
+#else
+	int64_t power64;
+	power64 = (int64_t)q2 * (int64_t)q2;
+	power64 += (int64_t)q1 * (int64_t)q1;
+	power64 -= (((int64_t)q1 * (int64_t)q2) >> 30) * (int64_t)coef;
+	power = power64 >> 28;
+#endif
+	trigger = (uint32_t)len * thresh;
+	trigger = multiply_32x32_rshift32(trigger, trigger);
+
+	Serial.printf("bool: power=%d, trig=%d\n", power, trigger);
+	return (power >= trigger);
+}
+
+
+
 
 /******************************************************************/
 

Audio.h

 
 
 
+
+class AudioAnalyzeToneDetect : public AudioStream
+{
+public:
+	AudioAnalyzeToneDetect(void)
+	  : AudioStream(1, inputQueueArray), thresh(6554), enabled(false) { }
+	void frequency(float freq, uint16_t cycles=10) {
+		set_params((int32_t)(cos((double)freq
+		  * (2.0 * 3.14159265358979323846 / AUDIO_SAMPLE_RATE_EXACT))
+		  * (double)2147483647.999), cycles,
+		  (float)AUDIO_SAMPLE_RATE_EXACT / freq * (float)cycles + 0.5f);
+	}
+	void set_params(int32_t coef, uint16_t cycles, uint16_t len);
+	bool available(void) {
+		__disable_irq();
+		bool flag = new_output;
+		if (flag) new_output = false;
+		__enable_irq();
+		return flag;
+	}
+	float read(void);
+	void threshold(float level) {
+		if (level < 0.01f) thresh = 655;
+		else if (level > 0.99f) thresh = 64881;
+		else thresh = level * 65536.0f + 0.5f;
+	}
+	operator bool();  // true if at or above threshold, false if below
+	virtual void update(void);
+private:
+	int32_t coefficient;	// Goertzel algorithm coefficient
+	int32_t s1, s2;		// Goertzel algorithm state
+	int32_t out1, out2;	// Goertzel algorithm state output
+	uint16_t length;	// number of samples to analyze
+	uint16_t count;		// how many left to analyze
+	uint16_t ncycles;	// number of waveform cycles to seek
+	uint16_t thresh;	// threshold, 655 to 64881 (1% to 99%)
+	bool enabled;
+	volatile bool new_output;
+	audio_block_t *inputQueueArray[1];
+};
+
+
+
+
+
+
+
 // TODO: more audio processing objects....
-//  N-channel mixer, adjustable gain on each channel
 //  sine wave with frequency modulation (phase)
-//  non-sine oscillators, ramp, triangle, square/pulse, etc
+//  waveforms with bandwidth limited tables for synth
 //  envelope: attack-decay-sustain-release, maybe other more complex?
-//  filters, low pass, high pass, bandpass, notch
-//  frequency analysis - FFT, single frequency (eg, filter for DTMF)
 //  MP3 decoding - it is possible with optimized code?
 //  other decompression, ADPCM, Vorbis, Speex, etc?