whispercpp 1.3.7 → 1.3.8

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 653ebeab03cd2a1649a8875b6f57c3f2d8e0882a8598b309fa3120786854bde3
-  data.tar.gz: abab58ce120bfd629098f184859ce7ab10d021390f0e469adeb4f4ad38c7ba66
+  metadata.gz: 264b8094aa764edf57c6e11ded6d412f5f71e174031b7f6f3d331bcd42487920
+  data.tar.gz: c6f5b4eccab0ca11c4b5e87055e4149ef25c73b570cab57c70b54a03aba839e4
 SHA512:
-  metadata.gz: f46de7f44669b0edfe454ee1f934ce817c38e96459d8557db3f21fbdcebd31b7e2f77f7447ea398cc7a4e173e5433d2f3114005bf38e749f12529942b8934392
-  data.tar.gz: 4236987e5951e4317a84d61b50b0a312075d5714e473e2c726001a1a0b0f775f7e59743be63d3f19466a8d49847bbdc9ce74f1eb0c35c92ffd000cca7024034f
+  metadata.gz: b23056b8ab8bf49b5a72f8771dc91dba1b1bc128f1ddb7b1b6f0c09bc2545d4d65305576009bfc33fa92d9714acf116bb43b0e35b3c9750d7c107892d9a765c3
+  data.tar.gz: 675048319ec63d7c88b63e5919b4ad65808e8ff3d75b903e729b9385fae76759ca6014c57bedda988bd2e71d4a9027eb43cda5f439e9fdca0bc5b57fdfeecf7f

data/README.md

@@ -117,7 +117,7 @@ Whisper::Params.new(
     threshold: 1.0, # defaults to 0.5
     min_speech_duration_ms: 500, # defaults to 250
     min_silence_duration_ms: 200, # defaults to 100
-    max_speech_duration_s: 30000, # default is FLT_MAX,
+    max_speech_duration_s: 30000.0, # default is FLT_MAX,
     speech_pad_ms: 50, # defaults to 30
     samples_overlap: 0.5 # defaults to 0.1
   ),
@@ -176,7 +176,7 @@ See whisper.cpp's [README](https://github.com/ggml-org/whisper.cpp/blob/master/R
 Boolean options:
 
 * `-DGGML_BLAS=1` -> `--enable-ggml-blas`
-* `-DWHISER_COREML=OFF` -> `--disable-whisper-coreml`
+* `-DWHISPER_COREML=OFF` -> `--disable-whisper-coreml`
 
 Argument options:
 
@@ -184,7 +184,7 @@ Argument options:
 
 Combination:
 
-* `-DGGML_CUDA=1 -DCMAKE_CUDA_ARCHITECTURES="86"` -> `--enable-ggml-cuda --cmake_cuda-architectures="86"`
+* `-DGGML_CUDA=1 -DCMAKE_CUDA_ARCHITECTURES="86"` -> `--enable-ggml-cuda --cmake-cuda-architectures="86"`
 
 For boolean options like `GGML_CUDA`, the README says `-DGGML_CUDA=1`. You need strip `-D`, prepend `--enable-` for `1` or `ON` (`--disable-` for `0` or `OFF`) and make it kebab-case: `--enable-ggml-cuda`.  
 For options which require arguments like `CMAKE_CUDA_ARCHITECTURES`, the README says `-DCMAKE_CUDA_ARCHITECTURES="86"`. You need strip `-D`, prepend `--`, make it kebab-case, append `=` and append argument: `--cmake-cuda-architectures="86"`.
@@ -478,13 +478,14 @@ Development
     % cd whisper.cpp/bindings/ruby
     % rake test
 
-First call of `rake test` builds an extension and downloads a model for testing. After that, you add tests in `tests` directory and modify `ext/ruby_whisper.cpp`.
+First call of `rake test` builds an extension and downloads a model for testing. After that, you add tests in `test` directory and modify `ext/*.{h,c,cpp}`.
 
 If something seems wrong on build, running `rake clean` solves some cases.
 
 ### Need help ###
 
 * Windows support
+* Check of compilation with various hardware
 * Refinement of C/C++ code, especially memory management
 
 License

data/ext/options.rb

@@ -105,7 +105,7 @@ class Options
     FileUtils.mkpath File.dirname(cache_path)
     File.open cache_path, "w" do |file|
       @options.reject {|name, (type, value)| value.nil?}.each do |name, (type, value)|
-        line = "set(CACHE{%<name>s} TYPE %<type>s FORCE VALUE %<value>s)" % {
+        line = 'set(%<name>s %<value>s CACHE %<type>s "" FORCE)' %{
           name:,
           type:,
           value: value == true ? "ON" : value == false ? "OFF" : escape_cmake(value)

data/ext/ruby_whisper.c

@@ -133,7 +133,6 @@ void Init_whisper() {
   id_coreml_compiled_models = rb_intern("coreml_compiled_models");
   id_cache = rb_intern("cache");
   id_n_processors = rb_intern("n_processors");
-  id_extended = rb_intern("extended");
   id_start_log_callback_thread = rb_intern("start_log_callback_thread");
   id_log_callback_thread = rb_intern("@log_callback_thread");
   id_alive_p = rb_intern("alive?");

data/ext/ruby_whisper.h

@@ -110,6 +110,13 @@ typedef struct {
   int n_samples;
 } ruby_whisper_full_args;
 
+typedef struct segments_from_samples_args {
+  VALUE *context;
+  VALUE *params;
+  float *samples;
+  int n_samples;
+} segments_from_samples_args;
+
 typedef struct ruby_whisper_full_parallel_args {
   VALUE *context;
   VALUE *params;
@@ -149,7 +156,6 @@ typedef struct {
   VALUE context;
 } ruby_whisper_parakeet_model;
 
-extern ID id_extended;
 extern ID id_log_callback_thread;
 extern ID id_start_log_callback_thread;
 extern ID id_alive_p;

data/ext/ruby_whisper_context.c

@@ -349,7 +349,11 @@ fill_samples(VALUE rb_args)
 
   if (RB_TYPE_P(*args->src, T_ARRAY)) {
     for (int i = 0; i < args->n_samples; i++) {
-      args->dest[i] = RFLOAT_VALUE(rb_ary_entry(*args->src, i));
+      VALUE sample = rb_ary_entry(*args->src, i);
+      if (!RB_FLOAT_TYPE_P(sample)) {
+        sample = rb_to_float(sample);
+      }
+      args->dest[i] = RFLOAT_VALUE(sample);
     }
   } else {
     // TODO: use rb_block_call
@@ -357,6 +361,9 @@ fill_samples(VALUE rb_args)
     for (int i = 0; i < args->n_samples; i++) {
       // TODO: check if iter is exhausted and raise ArgumentError appropriately
       VALUE sample = rb_funcall(iter, id_next, 0);
+      if (!RB_FLOAT_TYPE_P(sample)) {
+        sample = rb_to_float(sample);
+      }
       args->dest[i] = RFLOAT_VALUE(sample);
     }
   }
@@ -745,6 +752,45 @@ ruby_whisper_full_get_segment_no_speech_prob(VALUE self, VALUE i_segment)
   return DBL2NUM(no_speech_prob);
 }
 
+static VALUE
+ruby_whisper_full_n_vad_segments(VALUE self)
+{
+  ruby_whisper *rw;
+  GetContext(self, rw);
+
+  return INT2NUM(whisper_full_n_vad_segments(rw->context));
+}
+
+static int
+ruby_whisper_full_check_vad_segment_index(const ruby_whisper *rw, const VALUE i_segment)
+{
+    const int c_i_segment = NUM2INT(i_segment);
+    if (c_i_segment < 0 || c_i_segment >= whisper_full_n_vad_segments(rw->context)) {
+      rb_raise(rb_eIndexError, "segment index %d out of range", c_i_segment);
+    }
+    return c_i_segment;
+}
+
+static VALUE
+ruby_whisper_full_get_vad_segment_t0(VALUE self, VALUE i_segment)
+{
+  ruby_whisper *rw;
+  GetContext(self, rw);
+  const int c_i_segment = ruby_whisper_full_check_vad_segment_index(rw, i_segment);
+
+  return LONG2NUM(whisper_full_get_vad_segment_t0(rw->context, c_i_segment));
+}
+
+static VALUE
+ruby_whisper_full_get_vad_segment_t1(VALUE self, VALUE i_segment)
+{
+  ruby_whisper *rw;
+  GetContext(self, rw);
+  const int c_i_segment = ruby_whisper_full_check_vad_segment_index(rw, i_segment);
+
+  return LONG2NUM(whisper_full_get_vad_segment_t1(rw->context, c_i_segment));
+}
+
 // High level API
 
 static VALUE
@@ -830,6 +876,9 @@ init_ruby_whisper_context(VALUE *mWhisper)
   rb_define_method(cContext, "full_get_segment_speaker_turn_next", ruby_whisper_full_get_segment_speaker_turn_next, 1);
   rb_define_method(cContext, "full_get_segment_text", ruby_whisper_full_get_segment_text, 1);
   rb_define_method(cContext, "full_get_segment_no_speech_prob", ruby_whisper_full_get_segment_no_speech_prob, 1);
+  rb_define_method(cContext, "full_n_vad_segments", ruby_whisper_full_n_vad_segments, 0);
+  rb_define_method(cContext, "full_get_vad_segment_t0", ruby_whisper_full_get_vad_segment_t0, 1);
+  rb_define_method(cContext, "full_get_vad_segment_t1", ruby_whisper_full_get_vad_segment_t1, 1);
   rb_define_method(cContext, "full", ruby_whisper_full, -1);
   rb_define_method(cContext, "full_parallel", ruby_whisper_full_parallel, -1);
 

data/ext/ruby_whisper_log_settable.h

@@ -41,7 +41,6 @@
   rb_define_singleton_method(mod, "drain_logs", ruby_whisper_##log_queue##_s_drain_logs, 0); \
   rb_define_singleton_method(mod, "log_set", ruby_whisper_##log_queue##_s_log_set, 2); \
   rb_set_end_proc(ruby_whisper_##log_queue##_end_proc, Qnil); \
-  rb_extend_object(mod, mLogSettable); \
-  rb_funcall(mLogSettable, id_extended, 1, mod);
+  rb_extend_object(mod, mLogSettable);
 
 #endif

data/ext/ruby_whisper_params.c

@@ -446,7 +446,7 @@ ruby_whisper_params_allocate(VALUE klass)
     rwp->params.vad_model_path = ruby_strdup(rwp->params.vad_model_path);
   }
   rwp->diarize = false;
-  rwp->vad_params = TypedData_Wrap_Struct(cVADParams, &ruby_whisper_vad_params_type, (void *)&rwp->params.vad_params);
+  rwp->vad_params = rb_class_new_instance(0, NULL, cVADParams);
   rwp->new_segment_callback_container = ruby_whisper_callback_container_allocate();
   rwp->progress_callback_container = ruby_whisper_callback_container_allocate();
   rwp->encoder_begin_callback_container = ruby_whisper_callback_container_allocate();
@@ -918,7 +918,7 @@ ruby_whisper_params_set_temperature(VALUE self, VALUE value)
 {
   ruby_whisper_params *rwp;
   TypedData_Get_Struct(self, ruby_whisper_params, &ruby_whisper_params_type, rwp);
-  rwp->params.temperature = RFLOAT_VALUE(value);
+  rwp->params.temperature = NUM2DBL(value);
   return value;
 }
 /*
@@ -943,7 +943,7 @@ ruby_whisper_params_set_max_initial_ts(VALUE self, VALUE value)
 {
   ruby_whisper_params *rwp;
   TypedData_Get_Struct(self, ruby_whisper_params, &ruby_whisper_params_type, rwp);
-  rwp->params.max_initial_ts = RFLOAT_VALUE(value);
+  rwp->params.max_initial_ts = NUM2DBL(value);
   return value;
 }
 /*
@@ -966,7 +966,7 @@ ruby_whisper_params_set_length_penalty(VALUE self, VALUE value)
 {
   ruby_whisper_params *rwp;
   TypedData_Get_Struct(self, ruby_whisper_params, &ruby_whisper_params_type, rwp);
-  rwp->params.length_penalty = RFLOAT_VALUE(value);
+  rwp->params.length_penalty = NUM2DBL(value);
   return value;
 }
 /*
@@ -989,7 +989,7 @@ ruby_whisper_params_set_temperature_inc(VALUE self, VALUE value)
 {
   ruby_whisper_params *rwp;
   TypedData_Get_Struct(self, ruby_whisper_params, &ruby_whisper_params_type, rwp);
-  rwp->params.temperature_inc = RFLOAT_VALUE(value);
+  rwp->params.temperature_inc = NUM2DBL(value);
   return value;
 }
 /*
@@ -1014,7 +1014,7 @@ ruby_whisper_params_set_entropy_thold(VALUE self, VALUE value)
 {
   ruby_whisper_params *rwp;
   TypedData_Get_Struct(self, ruby_whisper_params, &ruby_whisper_params_type, rwp);
-  rwp->params.entropy_thold = RFLOAT_VALUE(value);
+  rwp->params.entropy_thold = NUM2DBL(value);
   return value;
 }
 /*
@@ -1037,7 +1037,7 @@ ruby_whisper_params_set_logprob_thold(VALUE self, VALUE value)
 {
   ruby_whisper_params *rwp;
   TypedData_Get_Struct(self, ruby_whisper_params, &ruby_whisper_params_type, rwp);
-  rwp->params.logprob_thold = RFLOAT_VALUE(value);
+  rwp->params.logprob_thold = NUM2DBL(value);
   return value;
 }
 /*
@@ -1060,7 +1060,7 @@ ruby_whisper_params_set_no_speech_thold(VALUE self, VALUE value)
 {
   ruby_whisper_params *rwp;
   TypedData_Get_Struct(self, ruby_whisper_params, &ruby_whisper_params_type, rwp);
-  rwp->params.no_speech_thold = RFLOAT_VALUE(value);
+  rwp->params.no_speech_thold = NUM2DBL(value);
   return value;
 }
 static VALUE
@@ -1307,6 +1307,7 @@ static VALUE
 ruby_whisper_params_set_vad_params(VALUE self, VALUE value)
 {
   ruby_whisper_params *rwp;
+  rb_check_typeddata(value, &ruby_whisper_vad_params_type);
   TypedData_Get_Struct(self, ruby_whisper_params, &ruby_whisper_params_type, rwp);
   rwp->vad_params = value;
   return value;

data/ext/ruby_whisper_transcribe.cpp

@@ -15,28 +15,9 @@ 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 VALUE full_body(VALUE rb_args);
 extern VALUE full_parallel_body(VALUE rb_args);
 
-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;
-}
-
 /*
  * transcribe a single file
  * can emit to a block results

data/ext/ruby_whisper_vad_context.c

@@ -12,13 +12,6 @@ extern VALUE release_samples(VALUE parsed);
 
 extern VALUE ruby_whisper_vad_segments_s_init(struct whisper_vad_segments *segments);
 
-typedef struct segments_from_samples_args {
-  VALUE *context;
-  VALUE *params;
-  float *samples;
-  int n_samples;
-} segments_from_samples_args;
-
 static size_t
 ruby_whisper_vad_context_memsize(const void *p)
 {
@@ -77,7 +70,24 @@ ruby_whisper_vad_context_initialize(VALUE self, VALUE model_path)
   return Qnil;
 }
 
-static VALUE
+typedef struct {
+  struct whisper_vad_context *context;
+  struct whisper_vad_params *params;
+  float *samples;
+  int n_samples;
+  struct whisper_vad_segments *result;
+} segments_from_samples_without_gvl_args;
+
+static void*
+segments_from_samples_without_gvl(void *rb_args)
+{
+  segments_from_samples_without_gvl_args *args = (segments_from_samples_without_gvl_args *)rb_args;
+  args->result = whisper_vad_segments_from_samples(args->context, *args->params, args->samples, args->n_samples);
+
+  return NULL;
+}
+
+VALUE
 segments_from_samples_body(VALUE rb_args)
 {
   segments_from_samples_args *args = (segments_from_samples_args *)rb_args;
@@ -87,9 +97,19 @@ segments_from_samples_body(VALUE rb_args)
   GetVADContext(*args->context, rwvc);
   GetVADParams(*args->params, rwvp);
 
-  struct whisper_vad_segments *segments = whisper_vad_segments_from_samples(rwvc->context, rwvp->params, args->samples, args->n_samples);
+  segments_from_samples_without_gvl_args segments_from_samples_without_gvl_args = {
+    rwvc->context,
+    &rwvp->params,
+    args->samples,
+    args->n_samples,
+    NULL
+  };
+  rb_thread_call_without_gvl(segments_from_samples_without_gvl, (void *)&segments_from_samples_without_gvl_args, NULL, NULL);
+  if (!segments_from_samples_without_gvl_args.result) {
+    rb_raise(rb_eRuntimeError, "Failed to process audio");
+  }
 
-  return ruby_whisper_vad_segments_s_init(segments);
+  return ruby_whisper_vad_segments_s_init(segments_from_samples_without_gvl_args.result);
 }
 
 static VALUE

data/ext/ruby_whisper_vad_context_detect.cpp

@@ -16,18 +16,16 @@ extern const rb_data_type_t ruby_whisper_vad_params_type;
 extern const rb_data_type_t ruby_whisper_vad_segments_type;
 
 extern VALUE ruby_whisper_vad_segments_s_init(struct whisper_vad_segments *segments);
+extern VALUE segments_from_samples_body(VALUE rb_args);
 
 VALUE
 ruby_whisper_vad_detect(VALUE self, VALUE file_path, VALUE params) {
   ruby_whisper_vad_context *rwvc;
-  ruby_whisper_vad_params *rwvp;
   std::string cpp_file_path;
   std::vector<float> pcmf32;
   std::vector<std::vector<float>> pcmf32s;
-  whisper_vad_segments *segments;
 
   GetVADContext(self, rwvc);
-  TypedData_Get_Struct(params, ruby_whisper_vad_params, &ruby_whisper_vad_params_type, rwvp);
 
   if (rb_respond_to(file_path, id_to_path)) {
     file_path = rb_funcall(file_path, id_to_path, 0);
@@ -38,12 +36,13 @@ ruby_whisper_vad_detect(VALUE self, VALUE file_path, VALUE params) {
     rb_raise(rb_eRuntimeError, "Failed to open '%s' as WAV file\n", cpp_file_path.c_str());
   }
 
-  segments = whisper_vad_segments_from_samples(rwvc->context, rwvp->params, pcmf32.data(), pcmf32.size());
-  if (segments == nullptr) {
-    rb_raise(rb_eRuntimeError, "Failed to process audio\n");
-  }
-
-  return ruby_whisper_vad_segments_s_init(segments);
+  segments_from_samples_args args = {
+    &self,
+    &params,
+    pcmf32.data(),
+    (int)pcmf32.size(),
+  };
+  return segments_from_samples_body((VALUE)&args);
 }
 
 #ifdef __cplusplus

data/ext/ruby_whisper_vad_params.c

@@ -31,7 +31,7 @@ static ID id_samples_overlap;
 
 const rb_data_type_t ruby_whisper_vad_params_type = {
   "ruby_whisper_vad_params",
-  {0, 0, ruby_whisper_vad_params_memsize,},
+  {0, RUBY_DEFAULT_FREE, ruby_whisper_vad_params_memsize,},
   0, 0,
   0
 };
@@ -56,7 +56,7 @@ ruby_whisper_vad_params_set_threshold(VALUE self, VALUE value)
 {
   ruby_whisper_vad_params *rwvp;
   TypedData_Get_Struct(self, ruby_whisper_vad_params, &ruby_whisper_vad_params_type, rwvp);
-  rwvp->params.threshold = RFLOAT_VALUE(value);
+  rwvp->params.threshold = NUM2DBL(value);
   return value;
 }
 
@@ -125,7 +125,7 @@ ruby_whisper_vad_params_set_max_speech_duration_s(VALUE self, VALUE value)
 {
   ruby_whisper_vad_params *rwvp;
   TypedData_Get_Struct(self, ruby_whisper_vad_params, &ruby_whisper_vad_params_type, rwvp);
-  rwvp->params.max_speech_duration_s = RFLOAT_VALUE(value);
+  rwvp->params.max_speech_duration_s = NUM2DBL(value);
   return value;
 }
 
@@ -171,7 +171,7 @@ ruby_whisper_vad_params_set_samples_overlap(VALUE self, VALUE value)
 {
   ruby_whisper_vad_params *rwvp;
   TypedData_Get_Struct(self, ruby_whisper_vad_params, &ruby_whisper_vad_params_type, rwvp);
-  rwvp->params.samples_overlap = RFLOAT_VALUE(value);
+  rwvp->params.samples_overlap = NUM2DBL(value);
   return value;
 }
 

data/ext/ruby_whisper_vad_segment.c

@@ -68,7 +68,7 @@ ruby_whisper_vad_segment_get_start_time(VALUE self)
   TypedData_Get_Struct(self, ruby_whisper_vad_segment, &ruby_whisper_vad_segment_type, rwvs);
   TypedData_Get_Struct(rwvs->segments, ruby_whisper_vad_segments, &ruby_whisper_vad_segments_type, rwvss);
   t0 = whisper_vad_segments_get_segment_t0(rwvss->segments, rwvs->index);
-  return DBL2NUM(t0 * 10);
+  return LONG2NUM(t0 * 10);
 }
 
 static VALUE
@@ -81,7 +81,7 @@ ruby_whisper_vad_segment_get_end_time(VALUE self)
   TypedData_Get_Struct(self, ruby_whisper_vad_segment, &ruby_whisper_vad_segment_type, rwvs);
   TypedData_Get_Struct(rwvs->segments, ruby_whisper_vad_segments, &ruby_whisper_vad_segments_type, rwvss);
   t1 = whisper_vad_segments_get_segment_t1(rwvss->segments, rwvs->index);
-  return DBL2NUM(t1 * 10);
+  return LONG2NUM(t1 * 10);
 }
 
 static VALUE

data/ext/sources/CMakeLists.txt

@@ -1,6 +1,6 @@
 cmake_minimum_required(VERSION 3.5) # for add_link_options and implicit target directories.
 project("whisper.cpp" C CXX)
-project("whisper.cpp" VERSION 1.9.0)
+project("whisper.cpp" VERSION 1.9.1)
 include(CheckIncludeFileCXX)
 
 set(SOVERSION 1)
@@ -19,6 +19,7 @@ endif()
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake/")
 
 set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin)
+set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin)
 
 if (CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR)
     set(WHISPER_STANDALONE ON)

data/ext/sources/cmake/parakeet.pc.in

@@ -1,7 +1,7 @@
 prefix=@CMAKE_INSTALL_PREFIX@
 exec_prefix=${prefix}
-libdir=${prefix}/@CMAKE_INSTALL_LIBDIR@
-includedir=${prefix}/include
+libdir=@CMAKE_INSTALL_FULL_LIBDIR@
+includedir=@CMAKE_INSTALL_FULL_INCLUDEDIR@
 
 Name: parakeet
 Description: Port of NVIDIA's Parakeet model in C/C++

data/ext/sources/cmake/whisper.pc.in

@@ -1,7 +1,7 @@
 prefix=@CMAKE_INSTALL_PREFIX@
 exec_prefix=${prefix}
-libdir=${prefix}/@CMAKE_INSTALL_LIBDIR@
-includedir=${prefix}/include
+libdir=@CMAKE_INSTALL_FULL_LIBDIR@
+includedir=@CMAKE_INSTALL_FULL_INCLUDEDIR@
 
 Name: whisper
 Description: Port of OpenAI's Whisper model in C/C++

data/ext/sources/examples/cli/cli.cpp

@@ -465,7 +465,15 @@ static void output_txt(struct whisper_context * ctx, std::ofstream & fout, const
             speaker = estimate_diarization_speaker(pcmf32s, t0, t1);
         }
 
-        fout << speaker << text << "\n";
+        if (!speaker.empty()) {
+            fout << speaker << text;
+        } else {
+            while (*text == ' ' || *text == '\t') {
+                text++;
+            }
+            fout << text;
+        }
+        fout << "\n";
     }
 }
 

data/ext/sources/examples/common-ggml.cpp

@@ -75,6 +75,7 @@ bool ggml_common_quantize_0(
         case GGML_FTYPE_MOSTLY_MXFP4:
         case GGML_FTYPE_MOSTLY_NVFP4:
         case GGML_FTYPE_MOSTLY_Q1_0:
+        case GGML_FTYPE_MOSTLY_Q2_0:
                 {
                     fprintf(stderr, "%s: invalid model type %d\n", __func__, ftype);
                     return false;
@@ -217,6 +218,7 @@ bool ggml_common_quantize_0(
                 case GGML_TYPE_MXFP4:
                 case GGML_TYPE_NVFP4:
                 case GGML_TYPE_Q1_0:
+                case GGML_TYPE_Q2_0:
                 case GGML_TYPE_COUNT:
                     {
                         fprintf(stderr, "%s: unsupported quantization type %d (%s)\n", __func__, ttype, ggml_type_name((ggml_type) ttype));

data/ext/sources/examples/vad-speech-segments/speech.cpp

@@ -26,6 +26,7 @@ static void vad_print_usage(int /*argc*/, char ** argv, const cli_params & param
     fprintf(stderr, "\n");
     fprintf(stderr, "usage: %s [options] file\n", argv[0]);
     fprintf(stderr, "supported audio formats: flac, mp3, ogg, wav\n");
+    fprintf(stderr, "Note: Output timestamps are in centiseconds (1/100th of a second).\n");
     fprintf(stderr, "\n");
     fprintf(stderr, "options:\n");
     fprintf(stderr, "  -h,        --help                          [default] show this help message and exit\n");
@@ -64,7 +65,7 @@ static bool vad_params_parse(int argc, char ** argv, cli_params & params) {
         else if (arg == "-ug"   || arg == "--use-gpu")                     { params.use_gpu                     = true; }
         else if (arg == "-vm"   || arg == "--vad-model")                   { params.vad_model                   = ARGV_NEXT; }
         else if (arg == "-vt"   || arg == "--vad-threshold")               { params.vad_threshold               = std::stof(ARGV_NEXT); }
-        else if (arg == "-vsd"  || arg == "--vad-min-speech-duration-ms")  { params.vad_min_speech_duration_ms  = std::stoi(ARGV_NEXT); }
+        else if (arg == "-vspd" || arg == "--vad-min-speech-duration-ms")  { params.vad_min_speech_duration_ms  = std::stoi(ARGV_NEXT); }
         else if (arg == "-vsd"  || arg == "--vad-min-silence-duration-ms") { params.vad_min_speech_duration_ms  = std::stoi(ARGV_NEXT); }
         else if (arg == "-vmsd" || arg == "--vad-max-speech-duration-s")   { params.vad_max_speech_duration_s   = std::stof(ARGV_NEXT); }
         else if (arg == "-vp"   || arg == "--vad-speech-pad-ms")           { params.vad_speech_pad_ms           = std::stoi(ARGV_NEXT); }
@@ -122,7 +123,7 @@ int main(int argc, char ** argv) {
         return 3;
     }
 
-    // Get the the vad segements using the probabilities that have been computed
+    // Get the the vad segments using the probabilities that have been computed
     // previously and stored in the whisper_vad_context.
     struct whisper_vad_params params = whisper_vad_default_params();
     params.threshold = cli_params.vad_threshold;

data/ext/sources/ggml/CMakeLists.txt

@@ -4,8 +4,8 @@ project("ggml" C CXX ASM)
 
 ### GGML Version
 set(GGML_VERSION_MAJOR 0)
-set(GGML_VERSION_MINOR 15)
-set(GGML_VERSION_PATCH 1)
+set(GGML_VERSION_MINOR 16)
+set(GGML_VERSION_PATCH 0)
 set(GGML_VERSION_BASE "${GGML_VERSION_MAJOR}.${GGML_VERSION_MINOR}.${GGML_VERSION_PATCH}")
 
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake/")
@@ -249,7 +249,7 @@ option(GGML_SYCL                            "ggml: use SYCL"
 option(GGML_SYCL_F16                        "ggml: use 16 bit floats for sycl calculations"   OFF)
 option(GGML_SYCL_GRAPH                      "ggml: enable graphs in the SYCL backend"         ON)
 option(GGML_SYCL_HOST_MEM_FALLBACK          "ggml: allow host memory fallback in SYCL reorder (requires kernel 6.8+)" ON)
-option(GGML_SYCL_SUPPORT_LEVEL_ZERO         "ggml: use Level Zero API in SYCL backend"  ON)
+option(GGML_SYCL_SUPPORT_LEVEL_ZERO_API     "ggml: use Level Zero API in SYCL backend"  ON)
 option(GGML_SYCL_DNN                        "ggml: enable oneDNN in the SYCL backend"         ON)
 set   (GGML_SYCL_TARGET "INTEL" CACHE STRING
                                             "ggml: sycl target device")
@@ -266,7 +266,6 @@ set   (GGML_OPENCL_TARGET_VERSION "300" CACHE STRING
                                             "ggml: OpenCL API version to target")
 
 option(GGML_HEXAGON                         "ggml: enable Hexagon backend"                    OFF)
-set(GGML_HEXAGON_FP32_QUANTIZE_GROUP_SIZE 128 CACHE STRING "ggml: quantize group size (32, 64, or 128)")
 
 # toolchain for vulkan-shaders-gen
 set   (GGML_VULKAN_SHADERS_GEN_TOOLCHAIN "" CACHE FILEPATH "ggml: toolchain file for vulkan-shaders-gen")

data/ext/sources/ggml/include/ggml-cuda.h

@@ -30,9 +30,6 @@ GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_cuda_buffer_type(int de
 // conduct allreduce operation between devices
 GGML_BACKEND_API bool ggml_backend_cuda_allreduce_tensor(ggml_backend_t * backends, struct ggml_tensor ** tensors, size_t n_backends);
 
-// split tensor buffer that splits matrices by rows across multiple devices
-GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_cuda_split_buffer_type(int main_device, const float * tensor_split);
-
 // pinned host buffer for use with the CPU backend for faster copies between CPU and GPU
 GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_cuda_host_buffer_type(void);
 

data/ext/sources/ggml/include/ggml-sycl.h

@@ -27,6 +27,14 @@ GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_sycl_buffer_type(int de
 // split tensor buffer that splits matrices by rows across multiple devices
 GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_sycl_split_buffer_type(const float * tensor_split);
 
+// Tensor parallelism (--split-mode tensor): comm_init/free/allreduce_tensor
+// trio queried by the meta-backend via ggml_backend_reg_get_proc_address.
+// See typedefs in ggml/include/ggml-backend.h. Mirrors the CUDA backend's
+// pattern (ggml_backend_cuda_comm_*).
+GGML_BACKEND_API void * ggml_backend_sycl_comm_init(ggml_backend_t * backends, size_t n_backends);
+GGML_BACKEND_API void   ggml_backend_sycl_comm_free(void * comm_ctx);
+GGML_BACKEND_API bool   ggml_backend_sycl_comm_allreduce_tensor(void * comm_ctx, struct ggml_tensor ** tensors);
+
 // pinned host buffer for use with the CPU backend for faster copies between CPU and GPU
 GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_sycl_host_buffer_type(void);
 

data/ext/sources/ggml/include/ggml.h

@@ -429,7 +429,8 @@ extern "C" {
         GGML_TYPE_MXFP4   = 39, // MXFP4 (1 block)
         GGML_TYPE_NVFP4   = 40, // NVFP4 (4 blocks, E4M3 scale)
         GGML_TYPE_Q1_0    = 41,
-        GGML_TYPE_COUNT   = 42,
+        GGML_TYPE_Q2_0    = 42,
+        GGML_TYPE_COUNT   = 43,
     };
 
     // precision
@@ -473,6 +474,7 @@ extern "C" {
         GGML_FTYPE_MOSTLY_MXFP4   = 25, // except 1d tensors
         GGML_FTYPE_MOSTLY_NVFP4   = 26, // except 1d tensors
         GGML_FTYPE_MOSTLY_Q1_0    = 27, // except 1d tensors
+        GGML_FTYPE_MOSTLY_Q2_0    = 28, // except 1d tensors
     };
 
     // available tensor operations:

data/ext/sources/ggml/src/CMakeLists.txt

@@ -438,7 +438,14 @@ if (GGML_CPU_ALL_VARIANTS)
             ggml_add_cpu_backend_variant(power8_2       POWER8  VSX)
             ggml_add_cpu_backend_variant(power9         POWER9  VSX)
             ggml_add_cpu_backend_variant(power10        POWER10 VSX)
-            ggml_add_cpu_backend_variant(power11        POWER11 VSX)
+            # POWER11 backend: only if compiler supports -mcpu=power11
+            check_cxx_compiler_flag("-mcpu=power11" GGML_CXX_SUPPORTS_POWER11)
+            if (GGML_CXX_SUPPORTS_POWER11)
+                message(STATUS "Compiler supports -mcpu=power11, enabling POWER11 backend")
+                ggml_add_cpu_backend_variant(power11 POWER11 VSX)
+            else()
+                message(STATUS "Skipping POWER11 backend: compiler does not support -mcpu=power11")
+            endif()
         else()
             message(FATAL_ERROR "Unsupported PowerPC target OS: ${CMAKE_SYSTEM_NAME}")
         endif()