Spaces:

SegerEnd
/

basic-pitch-wasm

Running

Basic Pitch without TensorFlow: C/WebAssembly runtime and demo

de6631a verified 13 days ago

2.57 kB

// bp.h - Spotify's Basic Pitch in C. Needs GCC or Clang. One model per thread.
//
// Audio: mono floats at BP_SAMPLE_RATE. Output: a row per frame, frames[row * BP_NOTES + key]
// (key 0 is MIDI 21), the same for onsets, and optional contours (3 bins a semitone).
// Errors are < 0: -1 no model, -2 bad plan, -3 out of memory.

#ifndef BP_H
#define BP_H

#define BP_SAMPLE_RATE 22050
#define BP_HOP         256     // samples per frame
#define BP_NOTES       88
#define BP_BINS        264
#define BP_WINDOW      43844   // samples per window
#define BP_WINDOW_ROWS 142     // rows per bp_window

typedef struct bp_model bp_model;

// plan.bin and weights.bin, kept alive while the model is used. NULL if invalid.
bp_model *bp_load(const int *plan, int plan_words, const float *weights);
bp_model *bp_load_files(const char *plan_path, const char *weights_path);
void bp_unload(bp_model *model);

// A whole recording, bp_frame_count(samples) rows. Returns the rows.
int bp_frame_count(int samples);
int bp_transcribe(bp_model *model, const float *audio, int samples, float *frames, float *onsets, float *contours);

// One window, rows index * BP_WINDOW_ROWS onwards, for sharing the work between threads.
int bp_window_count(int samples);
int bp_window(bp_model *model, const float *audio, int samples, int index, float *frames, float *onsets, float *contours);

// One raw window of BP_WINDOW samples, all 172 rows.
int bp_run(bp_model *model, const float *window, float *frames, float *onsets, float *contours);

// basicpitch.spotify.com's settings, BP_SETTINGS has its defaults.
typedef struct {
    double onset_threshold;   // note segmentation
    double frame_threshold;   // model confidence
    double min_note_ms;
    double min_frequency;     // Hz, 0 for no limit
    double max_frequency;     // Hz, 0 for no limit
    int boost_onsets;
} bp_settings;
#define BP_SETTINGS { 0.5, 0.3, 128, 0, 3000, 1 }

typedef struct { int pitch; double start, duration, amplitude; } bp_note;   // MIDI pitch, seconds, 0 to 1

// The notes (*notes malloc'ed). Returns their count.
int bp_find_notes(const float *frames, const float *onsets, int rows, const bp_settings *settings, bp_note **notes);

// A MIDI file (*bytes malloc'ed). The tempo only sets the beat grid. Returns its size.
int bp_midi(const bp_note *notes, int count, double tempo, unsigned char **bytes);

// All of it at once. settings NULL for the defaults.
int bp_audio_to_midi(bp_model *model, const float *audio, int samples, const bp_settings *settings, double tempo, unsigned char **bytes);

#endif

	// bp.h - Spotify's Basic Pitch in C. Needs GCC or Clang. One model per thread.
	//
	// Audio: mono floats at BP_SAMPLE_RATE. Output: a row per frame, frames[row * BP_NOTES + key]
	// (key 0 is MIDI 21), the same for onsets, and optional contours (3 bins a semitone).
	// Errors are < 0: -1 no model, -2 bad plan, -3 out of memory.

	#ifndef BP_H
	#define BP_H

	#define BP_SAMPLE_RATE 22050
	#define BP_HOP 256 // samples per frame
	#define BP_NOTES 88
	#define BP_BINS 264
	#define BP_WINDOW 43844 // samples per window
	#define BP_WINDOW_ROWS 142 // rows per bp_window

	typedef struct bp_model bp_model;

	// plan.bin and weights.bin, kept alive while the model is used. NULL if invalid.
	bp_model bp_load(const int plan, int plan_words, const float *weights);
	bp_model bp_load_files(const char plan_path, const char *weights_path);
	void bp_unload(bp_model *model);

	// A whole recording, bp_frame_count(samples) rows. Returns the rows.
	int bp_frame_count(int samples);
	int bp_transcribe(bp_model model, const float audio, int samples, float frames, float onsets, float *contours);

	// One window, rows index * BP_WINDOW_ROWS onwards, for sharing the work between threads.
	int bp_window_count(int samples);
	int bp_window(bp_model model, const float audio, int samples, int index, float frames, float onsets, float *contours);

	// One raw window of BP_WINDOW samples, all 172 rows.
	int bp_run(bp_model model, const float window, float frames, float onsets, float *contours);

	// basicpitch.spotify.com's settings, BP_SETTINGS has its defaults.
	typedef struct {
	double onset_threshold; // note segmentation
	double frame_threshold; // model confidence
	double min_note_ms;
	double min_frequency; // Hz, 0 for no limit
	double max_frequency; // Hz, 0 for no limit
	int boost_onsets;
	} bp_settings;
	#define BP_SETTINGS { 0.5, 0.3, 128, 0, 3000, 1 }

	typedef struct { int pitch; double start, duration, amplitude; } bp_note; // MIDI pitch, seconds, 0 to 1

	// The notes (*notes malloc'ed). Returns their count.
	int bp_find_notes(const float frames, const float onsets, int rows, const bp_settings settings, bp_note *notes);

	// A MIDI file (*bytes malloc'ed). The tempo only sets the beat grid. Returns its size.
	int bp_midi(const bp_note notes, int count, double tempo, unsigned char *bytes);

	// All of it at once. settings NULL for the defaults.
	int bp_audio_to_midi(bp_model model, const float audio, int samples, const bp_settings settings, double tempo, unsigned char *bytes);

	#endif