ref #22 : add "duration" option

Can be used to partially process a recording
pull/128/head
Georgi Gerganov 2 years ago
parent 8fdfb0ba92
commit c30bffc8a5
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735

@ -53,6 +53,7 @@ struct whisper_params {
int32_t n_processors = 1; int32_t n_processors = 1;
int32_t offset_t_ms = 0; int32_t offset_t_ms = 0;
int32_t offset_n = 0; int32_t offset_n = 0;
int32_t duration_ms = 0;
int32_t max_context = -1; int32_t max_context = -1;
int32_t max_len = 0; int32_t max_len = 0;
@ -95,6 +96,8 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
params.offset_t_ms = std::stoi(argv[++i]); params.offset_t_ms = std::stoi(argv[++i]);
} else if (arg == "-on" || arg == "--offset-n") { } else if (arg == "-on" || arg == "--offset-n") {
params.offset_n = std::stoi(argv[++i]); params.offset_n = std::stoi(argv[++i]);
} else if (arg == "-d" || arg == "--duration") {
params.duration_ms = std::stoi(argv[++i]);
} else if (arg == "-mc" || arg == "--max-context") { } else if (arg == "-mc" || arg == "--max-context") {
params.max_context = std::stoi(argv[++i]); params.max_context = std::stoi(argv[++i]);
} else if (arg == "-ml" || arg == "--max-len") { } else if (arg == "-ml" || arg == "--max-len") {
@ -154,6 +157,7 @@ void whisper_print_usage(int argc, char ** argv, const whisper_params & params)
fprintf(stderr, " -p N, --processors N number of processors to use during computation (default: %d)\n", params.n_processors); fprintf(stderr, " -p N, --processors N number of processors to use during computation (default: %d)\n", params.n_processors);
fprintf(stderr, " -ot N, --offset-t N time offset in milliseconds (default: %d)\n", params.offset_t_ms); fprintf(stderr, " -ot N, --offset-t N time offset in milliseconds (default: %d)\n", params.offset_t_ms);
fprintf(stderr, " -on N, --offset-n N segment index offset (default: %d)\n", params.offset_n); fprintf(stderr, " -on N, --offset-n N segment index offset (default: %d)\n", params.offset_n);
fprintf(stderr, " -d N, --duration N duration of audio to process in milliseconds (default: %d)\n", params.duration_ms);
fprintf(stderr, " -mc N, --max-context N maximum number of text context tokens to store (default: max)\n"); fprintf(stderr, " -mc N, --max-context N maximum number of text context tokens to store (default: max)\n");
fprintf(stderr, " -ml N, --max-len N maximum segment length in characters (default: %d)\n", params.max_len); fprintf(stderr, " -ml N, --max-len N maximum segment length in characters (default: %d)\n", params.max_len);
fprintf(stderr, " -wt N, --word-thold N word timestamp probability threshold (default: %f)\n", params.word_thold); fprintf(stderr, " -wt N, --word-thold N word timestamp probability threshold (default: %f)\n", params.word_thold);
@ -532,6 +536,7 @@ int main(int argc, char ** argv) {
wparams.n_threads = params.n_threads; wparams.n_threads = params.n_threads;
wparams.n_max_text_ctx = params.max_context >= 0 ? params.max_context : wparams.n_max_text_ctx; wparams.n_max_text_ctx = params.max_context >= 0 ? params.max_context : wparams.n_max_text_ctx;
wparams.offset_ms = params.offset_t_ms; wparams.offset_ms = params.offset_t_ms;
wparams.duration_ms = params.duration_ms;
wparams.token_timestamps = params.output_wts || params.max_len > 0; wparams.token_timestamps = params.output_wts || params.max_len > 0;
wparams.thold_pt = params.word_thold; wparams.thold_pt = params.word_thold;

@ -2339,6 +2339,7 @@ struct whisper_full_params whisper_full_default_params(enum whisper_sampling_str
/*.n_threads =*/ std::min(4, (int32_t) std::thread::hardware_concurrency()), /*.n_threads =*/ std::min(4, (int32_t) std::thread::hardware_concurrency()),
/*.n_max_text_ctx =*/ 16384, /*.n_max_text_ctx =*/ 16384,
/*.offset_ms =*/ 0, /*.offset_ms =*/ 0,
/*.duration_ms =*/ 0,
/*.translate =*/ false, /*.translate =*/ false,
/*.no_context =*/ false, /*.no_context =*/ false,
@ -2376,6 +2377,7 @@ struct whisper_full_params whisper_full_default_params(enum whisper_sampling_str
/*.n_threads =*/ std::min(4, (int32_t) std::thread::hardware_concurrency()), /*.n_threads =*/ std::min(4, (int32_t) std::thread::hardware_concurrency()),
/*.n_max_text_ctx =*/ 16384, /*.n_max_text_ctx =*/ 16384,
/*.offset_ms =*/ 0, /*.offset_ms =*/ 0,
/*.duration_ms =*/ 0,
/*.translate =*/ false, /*.translate =*/ false,
/*.no_context =*/ false, /*.no_context =*/ false,
@ -2496,11 +2498,12 @@ int whisper_full(
} }
const int seek_start = params.offset_ms/10; const int seek_start = params.offset_ms/10;
const int seek_end = seek_start + (params.duration_ms == 0 ? whisper_n_len(ctx) : params.duration_ms/10);
// if length of spectrogram is less than 1s (100 samples), then return // if length of spectrogram is less than 1s (100 samples), then return
// basically don't process anything that is less than 1s // basically don't process anything that is less than 1s
// see issue #39: https://github.com/ggerganov/whisper.cpp/issues/39 // see issue #39: https://github.com/ggerganov/whisper.cpp/issues/39
if (whisper_n_len(ctx) < 100 + seek_start) { if (seek_end < 100 + seek_start) {
return 0; return 0;
} }
@ -2533,7 +2536,7 @@ int whisper_full(
// main loop // main loop
int seek = seek_start; int seek = seek_start;
while (true) { while (true) {
int progress_cur = (100*seek)/whisper_n_len(ctx); const int progress_cur = (100*(seek - seek_start))/(seek_end - seek_start);
while (progress_cur >= progress_prev + progress_step) { while (progress_cur >= progress_prev + progress_step) {
progress_prev += progress_step; progress_prev += progress_step;
if (params.print_progress) { if (params.print_progress) {
@ -2541,7 +2544,7 @@ int whisper_full(
} }
} }
if (seek + 100 >= whisper_n_len(ctx)) { if (seek + 100 >= seek_end) {
break; break;
} }
@ -2622,7 +2625,7 @@ int whisper_full(
// end of text token // end of text token
if (token.id == whisper_token_eot(ctx)) { if (token.id == whisper_token_eot(ctx)) {
if (result_len == 0) { if (result_len == 0) {
if (seek + seek_delta + 100 >= whisper_n_len(ctx)) { if (seek + seek_delta + 100 >= seek_end) {
result_len = i + 1; result_len = i + 1;
} else { } else {
// TODO: figure out how to resolve this // TODO: figure out how to resolve this

@ -186,7 +186,8 @@ extern "C" {
int n_threads; int n_threads;
int n_max_text_ctx; int n_max_text_ctx;
int offset_ms; int offset_ms; // start offset in ms
int duration_ms; // audio duration to process in ms
bool translate; bool translate;
bool no_context; bool no_context;

Loading…
Cancel
Save