* Voiced/unvoiced detection using 2 criteria * * Uses two criteria: * 1. Low energy: originalEnergy < minEnergyThreshold → unvoiced (always c

Bind a NoteCmd sequence to a track/voice. pitch_shift >= 1

* Get encoded parameters for this frame * Returns a map of parameter names to quantized values

* Helper to create default settings with all required fields

* Create a minimal valid WAV file in memory * PCM, 8000 Hz, 16-bit, mono

* Encode WAV file to LPC hex data

* Leroux-Gueguen algorithm for finding reflection coefficients * This is a more numerically stable variant of Levinson-Durbin * * @param codi

* Calculate autocorrelation for a specific lag

* Calculate frame energy for voiced/unvoiced detection * Can calculate from original or emphasized signal

* Format RMS from residual energy * Scale to 16-bit amplitude space

Use voiced override on a frame if present

* Helper to get max absolute value from Float32Array

* Create a stop frame (silence)

* Apply unvoiced multiplier to all unvoiced frames (BlueWizard RMSNormalizer.m lines 39-44) * This is ALWAYS applied, regardless of normalization s

* Center-clipping to emphasize periodic excitation and reduce formant influence

* Sub-multiple detection for octave error correction * Checks if the detected pitch is actually a harmonic of the true pitch

* Find the period with maximum autocorrelation

Returns true if more samples are available (override from AudioStream)

* Parabolic interpolation for sub-sample pitch precision * Uses three points around the peak to estimate true maximum

* Load and resample WAV for preview (without full encoding)

* Convert MIDI note number to frequency in Hz * A4 = 440 Hz at MIDI note 69

---------------------------------- Soft clip and bias to 8-bit sample ----------------------------------

* Generate the next audio sample at 8KHz. * * This implements a 10-stage lattice filter driven by either: * - Voiced excitation: Periodic chirp wav

* Convert binary string to array of 4-bit nibbles

* Normalize unvoiced RMS using peak normalization (BlueWizard RMSNormalizer.m lines 24-37) * Finds max RMS across all unvoiced frames and scales to

* Normalize voiced RMS using peak normalization (BlueWizard RMSNormalizer.m lines 9-22) * Finds max RMS across all voiced frames and scales to rmsL

* Convert notes to NoteCmd format * Rests are represented as notes with period_us = 0

--------------------------------------- Utility: integer-only freq -> phase_inc phase_inc = (freq * 2^32) / FS_HZ where freq = 1e6 / period_us so phas

* Plays the music by iterating through note commands and toggling the GPIO pin * * @param max_len_us Maximum playback time in microseconds (used to

* Process audio samples into LPC frames * Following BlueWizard's algorithm with selective buffer usage: * - Voiced/unvoiced detection uses BOTH

* Read and convert PCM samples to normalized float32 mono

* Reset the speech synthesizer to start from the beginning of the data. * Clears all filter state and synthesis parameters. * The first frame will b

* Reverse bits within a 4-bit nibble * Example: 0010 (2) → 0100 (4)

* Start speaking encoded speech data * @param address Pointer to LPC-encoded speech data * @param mode Speech chip to emulate (TMS5220 or TMS510

* Split a track's notes into monophonic streams (no overlapping notes per stream) * Notes with small overlaps will have their duration adjusted rathe

* Convert a value to binary string with specified bit width * Clamps value to prevent overflow

* Minimal YIN implementation (difference function + CMND) returning period or 0 if none