ruby : transcribe without GVL, accept more MemoryViews, Windows support, fix memory size report, improve document (#3775)

* Change MemoryView example using NDAV

* Add note on audio attributes for #full and #full_parallel

* Support more variants of MemoryView

* Use IO.popen instead of Kernel.` for Windows compatibility

* Use cmake's -C option instead of multiple -D options

* Fix memsize calculation

* Remove unused argument

* Add is_interrupted field to abort callback container

* Fix RBS syntax

* Address document comment for RDoc

* Add .document for RDoc

* Add .rdoc_options

* Run #full without GVL

* Initialize callbacks with nil

* Specify implicity Whisper::Params to distinguish from Whisper::Context::Params

* Run callbacks without GVL

* Call log callback with GVL

* Run full_parallel without GVL

* Run transcribe without GVL

* Fix ruby_whisper_lock_gvl and ruby_whisper_unlock_gvl

* Fix return value of encoder_begin_callback

* Report GVL unlocking from transcribe

* Remove unused interface

* Restore overload of full_parallel

* Close process

* Fix struct name

* Make is_without_gvl thread local

* Use rb_thread_call_with_gvl instead of global variable

* Retrieve instance variable in GVL

* Narrow acceptable MemoryView format

* Fix option cache path

* Reduce files in package

* Use append_cflags

* Add ext/*.rb to task dependencies

* Use copy instead of cp

* Make TestPackage more portable

* Patch for lower version Ruby

* Make build scripts more portable

* Add Windows support

* Don't raise exceptions
This commit is contained in:
KITAITI Makoto
2026-05-07 06:28:18 +02:00
committed by GitHub
parent 4bf733672b
commit c81b2dabbc
17 changed files with 653 additions and 166 deletions
+44 -3
View File
@@ -15,8 +15,37 @@ extern ID id_call;
extern ID id_to_path;
extern ID transcribe_option_names[1];
extern void
prepare_transcription(ruby_whisper_params * rwp, VALUE * self, int n_processors);
extern void prepare_transcription(ruby_whisper_params * rwp, VALUE * self, int n_processors);
typedef struct{
struct whisper_context *context;
struct whisper_full_params *params;
float *samples;
size_t n_samples;
int n_processors;
int result;
} transcribe_without_gvl_args;
static void*
transcribe_without_gvl(void *rb_args)
{
transcribe_without_gvl_args *args = (transcribe_without_gvl_args *)rb_args;
args->result = whisper_full_parallel(args->context, *args->params, args->samples, args->n_samples, args->n_processors);
return NULL;
}
typedef struct {
ruby_whisper_abort_callback_container *abort_callback_container;
} transcribe_ubf_args;
static void
transcribe_ubf(void *rb_args)
{
transcribe_ubf_args *args = (transcribe_ubf_args *)rb_args;
args->abort_callback_container->is_interrupted = true;
}
/*
* transcribe a single file
@@ -75,7 +104,19 @@ ruby_whisper_transcribe(int argc, VALUE *argv, VALUE self) {
prepare_transcription(rwp, &self, n_processors);
if (whisper_full_parallel(rw->context, rwp->params, pcmf32.data(), pcmf32.size(), n_processors) != 0) {
transcribe_without_gvl_args args = {
rw->context,
&rwp->params,
pcmf32.data(),
pcmf32.size(),
n_processors,
0,
};
transcribe_ubf_args ubf_args = {
rwp->abort_callback_container,
};
rb_thread_call_without_gvl(transcribe_without_gvl, (void *)&args, transcribe_ubf, (void *)&ubf_args);
if (args.result != 0) {
fprintf(stderr, "failed to process audio\n");
return self;
}