/* Command line front end: a 16 kHz mono WAV in, segments or frame scores out. */ #include "pyseg.h" #include #include #include #include "wav.h" static const char *usage = "usage: pyseg_cli [options] file.wav\n" "\n" " --mode speech|overlap|count|frames what to print (default speech)\n" " --step SECONDS window hop, default 1.0 (a tenth of the window)\n" " --threads N windows to run in parallel, default 1\n" " --batch N windows per engine call, default 4\n" " --weights FILE weight blob; default is the embedded copy\n" " --onset X score above which a region opens, default 0.5\n" " --offset X score below which it closes, default 0.5\n" " --min-on SECONDS drop regions shorter than this, default 0\n" " --min-off SECONDS fill gaps shorter than this, default 0\n" " --scalar portable C kernels instead of SIMD\n" "\n" "speech and overlap print [start end] pairs in seconds; count and frames\n" "print one line per frame.\n"; enum { MODE_SPEECH, MODE_OVERLAP, MODE_COUNT, MODE_FRAMES }; static int parse_mode(const char *s, int *mode) { if (strcmp(s, "speech") == 0 || strcmp(s, "vad") == 0) *mode = MODE_SPEECH; else if (strcmp(s, "overlap") == 0 || strcmp(s, "osd") == 0) *mode = MODE_OVERLAP; else if (strcmp(s, "count") == 0) *mode = MODE_COUNT; else if (strcmp(s, "frames") == 0) *mode = MODE_FRAMES; else return -1; return 0; } int main(int argc, char **argv) { pyseg_slide_params slide; pyseg_binarize_params bin; pyseg_slide_params_default(&slide); pyseg_binarize_params_default(&bin); int mode = MODE_SPEECH; const char *path = NULL; for (int i = 1; i < argc; ++i) { const char *a = argv[i]; const char *v = (i + 1 < argc) ? argv[i + 1] : NULL; #define NEED(name) \ if (!v) { \ fprintf(stderr, "%s needs a value\n", name); \ return 2; \ } if (strcmp(a, "--mode") == 0) { NEED("--mode"); if (parse_mode(v, &mode) != 0) { fprintf(stderr, "unknown mode: %s\n", v); return 2; } ++i; } else if (strcmp(a, "--step") == 0) { NEED("--step"); slide.step = atof(v); ++i; } else if (strcmp(a, "--threads") == 0) { NEED("--threads"); slide.threads = atoi(v); ++i; } else if (strcmp(a, "--batch") == 0) { NEED("--batch"); slide.batch = atoi(v); ++i; } else if (strcmp(a, "--weights") == 0) { NEED("--weights"); slide.weights = v; ++i; } else if (strcmp(a, "--onset") == 0) { NEED("--onset"); bin.onset = atof(v); ++i; } else if (strcmp(a, "--offset") == 0) { NEED("--offset"); bin.offset = atof(v); ++i; } else if (strcmp(a, "--min-on") == 0) { NEED("--min-on"); bin.min_duration_on = atof(v); ++i; } else if (strcmp(a, "--min-off") == 0) { NEED("--min-off"); bin.min_duration_off = atof(v); ++i; } else if (strcmp(a, "--scalar") == 0) { slide.flags |= PYSEG_SCALAR_KERNELS; } else if (strcmp(a, "-h") == 0 || strcmp(a, "--help") == 0) { fputs(usage, stdout); return 0; } else if (a[0] == '-') { fprintf(stderr, "unknown option: %s\n%s", a, usage); return 2; } else { path = a; } #undef NEED } if (!path) { fputs(usage, stderr); return 2; } wav_t wav; const char *err = NULL; if (wav_read_mono_s16(path, &wav, &err) != 0) { fprintf(stderr, "%s: %s\n", path, err ? err : "read failed"); return 1; } if (wav.sample_rate != PYSEG_SAMPLE_RATE) { fprintf(stderr, "%s: expected %d Hz, got %d\n", path, PYSEG_SAMPLE_RATE, wav.sample_rate); wav_free(&wav); return 1; } const size_t count = wav.count; const size_t frames = pyseg_slide_frames(count, slide.step); float *tracks = malloc(frames * PYSEG_TRACKS * sizeof(float)); if (!tracks) { wav_free(&wav); return 1; } const size_t got = pyseg_slide_s16(wav.samples, count, &slide, tracks); wav_free(&wav); if (got == 0) { fprintf(stderr, "inference failed: no engine could be created%s\n", slide.weights ? " from the given weight file" : "; this build may carry no embedded weights, " "in which case pass --weights pyseg_weights.bin"); free(tracks); return 1; } if (mode == MODE_FRAMES || mode == MODE_COUNT) { for (size_t f = 0; f < got; ++f) { const float *row = tracks + f * PYSEG_TRACKS; if (mode == MODE_COUNT) printf("%.3f %.4f\n", pyseg_frame_time(f), row[PYSEG_TRACK_COUNT]); else printf("%.3f %.4f %.4f %.4f\n", pyseg_frame_time(f), row[PYSEG_TRACK_SPEECH], row[PYSEG_TRACK_OVERLAP], row[PYSEG_TRACK_COUNT]); } free(tracks); return 0; } const int track = mode == MODE_SPEECH ? PYSEG_TRACK_SPEECH : PYSEG_TRACK_OVERLAP; const size_t n = pyseg_binarize(tracks, got, track, &bin, NULL, 0); pyseg_segment *segments = n ? malloc(n * sizeof *segments) : NULL; if (n && !segments) { free(tracks); return 1; } const size_t written = pyseg_binarize(tracks, got, track, &bin, segments, n); for (size_t i = 0; i < written; ++i) printf("%.3f %.3f\n", segments[i].start, segments[i].end); free(segments); free(tracks); return 0; }