whispercpp 1.3.0 → 1.3.2

data/ext/{whisper.h → sources/include/whisper.h}

@@ -2,6 +2,7 @@
 #define WHISPER_H
 
 #include "ggml.h"
+#include "ggml-cpu.h"
 
 #include <stddef.h>
 #include <stdint.h>
@@ -99,6 +100,7 @@ extern "C" {
         WHISPER_AHEADS_LARGE_V1,
         WHISPER_AHEADS_LARGE_V2,
         WHISPER_AHEADS_LARGE_V3,
+        WHISPER_AHEADS_LARGE_V3_TURBO,
     };
 
     typedef struct whisper_ahead {
@@ -113,6 +115,7 @@ extern "C" {
 
     struct whisper_context_params {
         bool  use_gpu;
+        bool  flash_attn;
         int   gpu_device;  // CUDA device
 
         // [EXPERIMENTAL] Token-level timestamps with DTW
@@ -186,6 +189,15 @@ extern "C" {
         uint32_t             value; // Unicode code point or rule ID
     } whisper_grammar_element;
 
+    typedef struct whisper_vad_params {
+        float threshold;               // Probability threshold to consider as speech.
+        int   min_speech_duration_ms;  // Min duration for a valid speech segment.
+        int   min_silence_duration_ms; // Min silence duration to consider speech as ended.
+        float max_speech_duration_s;   // Max duration of a speech segment before forcing a new segment.
+        int   speech_pad_ms;           // Padding added before and after speech segments.
+        float samples_overlap;         // Overlap in seconds when copying audio samples from speech segment.
+    } whisper_vad_params;
+
     // Various functions for loading a ggml whisper model.
     // Allocate (almost) all memory needed for the model.
     // Return NULL on failure
@@ -237,6 +249,13 @@ extern "C" {
     //                     GPU, by caching compiled 'blobs' there.
     //                     Set to nullptr if not used.
     // Returns 0 on success. If OpenVINO is not enabled in build, this simply returns 1.
+    WHISPER_API int whisper_ctx_init_openvino_encoder_with_state(
+        struct whisper_context * ctx,
+          struct whisper_state * state,
+                    const char * model_path,
+                    const char * device,
+                    const char * cache_dir);
+
     WHISPER_API int whisper_ctx_init_openvino_encoder(
         struct whisper_context * ctx,
                     const char * model_path,
@@ -265,22 +284,6 @@ extern "C" {
                                int   n_samples,
                                int   n_threads);
 
-    // Convert RAW PCM audio to log mel spectrogram but applies a Phase Vocoder to speed up the audio x2.
-    // The resulting spectrogram is stored inside the default state of the provided whisper context.
-    // Returns 0 on success
-    WHISPER_API int whisper_pcm_to_mel_phase_vocoder(
-        struct whisper_context * ctx,
-                   const float * samples,
-                           int   n_samples,
-                           int   n_threads);
-
-    WHISPER_API int whisper_pcm_to_mel_phase_vocoder_with_state(
-        struct whisper_context * ctx,
-          struct whisper_state * state,
-                   const float * samples,
-                           int   n_samples,
-                           int   n_threads);
-
     // This can be used to set a custom log mel spectrogram inside the default state of the provided whisper context.
     // Use this instead of whisper_pcm_to_mel() if you want to provide your own log mel spectrogram.
     // n_mel must be 80
@@ -350,7 +353,7 @@ extern "C" {
     int whisper_token_count(struct whisper_context * ctx, const char * text);
 
     // Largest language id (i.e. number of available languages - 1)
-    WHISPER_API int whisper_lang_max_id();
+    WHISPER_API int whisper_lang_max_id(void);
 
     // Return the id of the specified language, returns -1 if not found
     // Examples:
@@ -430,6 +433,14 @@ extern "C" {
     WHISPER_API whisper_token whisper_token_transcribe(struct whisper_context * ctx);
 
     // Performance information from the default state.
+    struct whisper_timings {
+        float sample_ms;
+        float encode_ms;
+        float decode_ms;
+        float batchd_ms;
+        float prompt_ms;
+    };
+    WHISPER_API struct whisper_timings * whisper_get_timings(struct whisper_context * ctx);
     WHISPER_API void whisper_print_timings(struct whisper_context * ctx);
     WHISPER_API void whisper_reset_timings(struct whisper_context * ctx);
 
@@ -498,7 +509,6 @@ extern "C" {
 
         // [EXPERIMENTAL] speed-up techniques
         // note: these can significantly reduce the quality of the output
-        bool speed_up;          // speed-up the audio by 2x using Phase Vocoder
         bool debug_mode;        // enable debug_mode provides extra info (eg. Dump log_mel)
         int  audio_ctx;         // overwrite the audio context size (0 = use default)
 
@@ -521,8 +531,8 @@ extern "C" {
         bool detect_language;
 
         // common decoding parameters:
-        bool suppress_blank;    // ref: https://github.com/openai/whisper/blob/f82bc59f5ea234d4b97fb2860842ed38519f7e65/whisper/decoding.py#L89
-        bool suppress_non_speech_tokens; // ref: https://github.com/openai/whisper/blob/7858aa9c08d98f75575035ecd6481f462d66ca27/whisper/tokenizer.py#L224-L253
+        bool suppress_blank; // ref: https://github.com/openai/whisper/blob/f82bc59f5ea234d4b97fb2860842ed38519f7e65/whisper/decoding.py#L89
+        bool suppress_nst;   // non-speech tokens, ref: https://github.com/openai/whisper/blob/7858aa9c08d98f75575035ecd6481f462d66ca27/whisper/tokenizer.py#L224-L253
 
         float temperature;      // initial decoding temperature, ref: https://ai.stackexchange.com/a/32478
         float max_initial_ts;   // ref: https://github.com/openai/whisper/blob/f82bc59f5ea234d4b97fb2860842ed38519f7e65/whisper/decoding.py#L97
@@ -533,7 +543,7 @@ extern "C" {
         float temperature_inc;
         float entropy_thold;    // similar to OpenAI's "compression_ratio_threshold"
         float logprob_thold;
-        float no_speech_thold;  // TODO: not implemented
+        float no_speech_thold;
 
         struct {
             int best_of;    // ref: https://github.com/openai/whisper/blob/f82bc59f5ea234d4b97fb2860842ed38519f7e65/whisper/transcribe.py#L264
@@ -569,13 +579,20 @@ extern "C" {
         size_t                           n_grammar_rules;
         size_t                           i_start_rule;
         float                            grammar_penalty;
+
+        // Voice Activity Detection (VAD) params
+        bool         vad;                         // Enable VAD
+        const char * vad_model_path;              // Path to VAD model
+
+        whisper_vad_params vad_params;
     };
 
     // NOTE: this function allocates memory, and it is the responsibility of the caller to free the pointer - see whisper_free_context_params & whisper_free_params()
-    WHISPER_API struct whisper_context_params * whisper_context_default_params_by_ref();
-    WHISPER_API struct whisper_context_params whisper_context_default_params(void);
+    WHISPER_API struct whisper_context_params * whisper_context_default_params_by_ref(void);
+    WHISPER_API struct whisper_context_params   whisper_context_default_params       (void);
+
     WHISPER_API struct whisper_full_params * whisper_full_default_params_by_ref(enum whisper_sampling_strategy strategy);
-    WHISPER_API struct whisper_full_params whisper_full_default_params(enum whisper_sampling_strategy strategy);
+    WHISPER_API struct whisper_full_params   whisper_full_default_params       (enum whisper_sampling_strategy strategy);
 
     // Run the entire model: PCM -> log mel spectrogram -> encoder -> decoder -> text
     // Not thread safe for same context
@@ -651,6 +668,53 @@ extern "C" {
     WHISPER_API float whisper_full_get_token_p           (struct whisper_context * ctx, int i_segment, int i_token);
     WHISPER_API float whisper_full_get_token_p_from_state(struct whisper_state * state, int i_segment, int i_token);
 
+    //
+    // Voice Activity Detection (VAD)
+    //
+
+    struct whisper_vad_context;
+
+    WHISPER_API struct whisper_vad_params whisper_vad_default_params(void);
+
+    struct whisper_vad_context_params {
+        int   n_threads;  // The number of threads to use for processing.
+        bool  use_gpu;
+        int   gpu_device; // CUDA device
+    };
+
+    WHISPER_API struct whisper_vad_context_params whisper_vad_default_context_params(void);
+
+    WHISPER_API struct whisper_vad_context * whisper_vad_init_from_file_with_params(const char * path_model,              struct whisper_vad_context_params params);
+    WHISPER_API struct whisper_vad_context * whisper_vad_init_with_params          (struct whisper_model_loader * loader, struct whisper_vad_context_params params);
+
+    WHISPER_API bool whisper_vad_detect_speech(
+            struct whisper_vad_context * vctx,
+                           const float * samples,
+                                   int   n_samples);
+
+    WHISPER_API int     whisper_vad_n_probs(struct whisper_vad_context * vctx);
+    WHISPER_API float * whisper_vad_probs  (struct whisper_vad_context * vctx);
+
+    struct whisper_vad_segments;
+
+    WHISPER_API struct whisper_vad_segments * whisper_vad_segments_from_probs(
+            struct whisper_vad_context * vctx,
+            struct whisper_vad_params    params);
+
+    WHISPER_API struct whisper_vad_segments * whisper_vad_segments_from_samples(
+            struct whisper_vad_context * vctx,
+            struct whisper_vad_params    params,
+                           const float * samples,
+                                   int   n_samples);
+
+    WHISPER_API int whisper_vad_segments_n_segments(struct whisper_vad_segments * segments);
+
+    WHISPER_API float whisper_vad_segments_get_segment_t0(struct whisper_vad_segments * segments, int i_segment);
+    WHISPER_API float whisper_vad_segments_get_segment_t1(struct whisper_vad_segments * segments, int i_segment);
+
+    WHISPER_API void whisper_vad_free_segments(struct whisper_vad_segments * segments);
+    WHISPER_API void whisper_vad_free         (struct whisper_vad_context  * ctx);
+
     ////////////////////////////////////////////////////////////////////////////
 
     // Temporary helpers needed for exposing ggml interface
@@ -664,6 +728,9 @@ extern "C" {
 
     WHISPER_API void whisper_log_set(ggml_log_callback log_callback, void * user_data);
 
+    // Get the no_speech probability for the specified segment
+    WHISPER_API float whisper_full_get_segment_no_speech_prob           (struct whisper_context * ctx, int i_segment);
+    WHISPER_API float whisper_full_get_segment_no_speech_prob_from_state(struct whisper_state * state, int i_segment);
 #ifdef __cplusplus
 }
 #endif

data/ext/sources/src/CMakeLists.txt

@@ -0,0 +1,143 @@
+# TODO: should not use this
+if (WIN32)
+    add_compile_definitions(_CRT_SECURE_NO_WARNINGS)
+
+    if (BUILD_SHARED_LIBS)
+        set(CMAKE_WINDOWS_EXPORT_ALL_SYMBOLS ON)
+    endif()
+endif()
+
+
+if (WHISPER_ALL_WARNINGS)
+    if (NOT MSVC)
+        list(APPEND WARNING_FLAGS -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function)
+        list(APPEND C_FLAGS       -Wshadow -Wstrict-prototypes -Wpointer-arith -Wmissing-prototypes
+                                  -Werror=implicit-int -Werror=implicit-function-declaration)
+        list(APPEND CXX_FLAGS     -Wmissing-declarations -Wmissing-noreturn)
+
+        list(APPEND C_FLAGS   ${WARNING_FLAGS})
+        list(APPEND CXX_FLAGS ${WARNING_FLAGS})
+
+        add_compile_options("$<$<COMPILE_LANGUAGE:C>:${C_FLAGS}>"
+                            "$<$<COMPILE_LANGUAGE:CXX>:${CXX_FLAGS}>")
+    else()
+        # todo : msvc
+        set(C_FLAGS   "")
+        set(CXX_FLAGS "")
+    endif()
+endif()
+
+if (WHISPER_COREML)
+    find_library(FOUNDATION_FRAMEWORK Foundation)
+    find_library(COREML_FRAMEWORK CoreML)
+
+    if (COREML_FRAMEWORK)
+        message(STATUS "CoreML framework found")
+
+        set(WHISPER_EXTRA_FLAGS ${WHISPER_EXTRA_FLAGS} -DWHISPER_USE_COREML)
+    else()
+        message(FATAL_ERROR "CoreML framework not found")
+    endif()
+
+    if (WHISPER_COREML_ALLOW_FALLBACK)
+        set(WHISPER_EXTRA_FLAGS ${WHISPER_EXTRA_FLAGS} -DWHISPER_COREML_ALLOW_FALLBACK)
+    endif()
+endif()
+
+if (WHISPER_OPENVINO)
+    find_package(OpenVINO REQUIRED COMPONENTS Runtime)
+endif()
+
+#
+# libraries
+#
+
+# whisper.coreml
+
+if (WHISPER_COREML)
+    set(TARGET whisper.coreml)
+
+    add_library(${TARGET}
+        coreml/whisper-encoder.h
+        coreml/whisper-encoder.mm
+        coreml/whisper-encoder-impl.h
+        coreml/whisper-encoder-impl.m
+        )
+
+    include(DefaultTargetOptions)
+
+    target_include_directories(${TARGET} PUBLIC
+        .
+        )
+
+    target_link_libraries(${TARGET} PRIVATE ${FOUNDATION_FRAMEWORK} ${COREML_FRAMEWORK})
+
+    set_target_properties(${TARGET} PROPERTIES
+        COMPILE_FLAGS "-fobjc-arc"
+        XCODE_ATTRIBUTE_CLANG_ENABLE_OBJC_ARC YES
+        )
+    set_target_properties(${TARGET} PROPERTIES FOLDER "libs")
+endif()
+
+if (WHISPER_OPENVINO)
+    set(TARGET whisper.openvino)
+
+    add_library(${TARGET} OBJECT
+        openvino/whisper-openvino-encoder.h
+        openvino/whisper-openvino-encoder.cpp
+        )
+
+    target_include_directories(${TARGET} PUBLIC
+        .
+        )
+
+    set_property(TARGET ${TARGET} PROPERTY POSITION_INDEPENDENT_CODE ON)
+    set(WHISPER_EXTRA_FLAGS ${WHISPER_EXTRA_FLAGS} -DWHISPER_USE_OPENVINO)
+
+    target_link_libraries(${TARGET} PRIVATE ggml openvino::runtime)
+    set_target_properties(${TARGET} PROPERTIES FOLDER "libs")
+endif()
+
+# whisper
+
+add_library(whisper
+            ../include/whisper.h
+            whisper-arch.h
+            whisper.cpp
+            )
+
+# Set the version numbers
+set_target_properties(whisper PROPERTIES
+    VERSION ${PROJECT_VERSION}
+    SOVERSION ${SOVERSION}
+)
+
+target_include_directories(whisper PUBLIC . ../include)
+target_compile_features   (whisper PUBLIC cxx_std_11) # don't bump
+
+if (CMAKE_CXX_BYTE_ORDER STREQUAL "BIG_ENDIAN")
+    set(WHISPER_EXTRA_FLAGS ${WHISPER_EXTRA_FLAGS} -DWHISPER_BIG_ENDIAN)
+endif()
+
+if (WHISPER_EXTRA_FLAGS)
+    target_compile_options(whisper PRIVATE ${WHISPER_EXTRA_FLAGS})
+endif()
+
+target_link_libraries(whisper PUBLIC ggml)
+
+if (WHISPER_COREML)
+    target_link_libraries(whisper PRIVATE whisper.coreml)
+endif()
+
+if (WHISPER_OPENVINO)
+    target_link_libraries(whisper PRIVATE whisper.openvino)
+endif()
+
+if (WHISPER_MKL)
+    target_link_libraries(whisper PRIVATE MKL::MKL)
+endif()
+
+if (BUILD_SHARED_LIBS)
+    set_target_properties(whisper PROPERTIES POSITION_INDEPENDENT_CODE ON)
+    target_compile_definitions(whisper PRIVATE WHISPER_SHARED WHISPER_BUILD)
+endif()

data/ext/sources/src/coreml/whisper-decoder-impl.h

@@ -0,0 +1,158 @@
+//
+// whisper-decoder-impl.h
+//
+// This file was automatically generated and should not be edited.
+//
+
+#import <Foundation/Foundation.h>
+#import <CoreML/CoreML.h>
+#include <stdint.h>
+#include <os/log.h>
+
+NS_ASSUME_NONNULL_BEGIN
+
+/// Model Prediction Input Type
+API_AVAILABLE(macos(10.15), ios(13.0), watchos(6.0), tvos(13.0)) __attribute__((visibility("hidden")))
+@interface whisper_decoder_implInput : NSObject<MLFeatureProvider>
+
+/// token_data as 1 by 1 matrix of floats
+@property (readwrite, nonatomic, strong) MLMultiArray * token_data;
+
+/// audio_data as 1 × 1500 × 384 3-dimensional array of floats
+@property (readwrite, nonatomic, strong) MLMultiArray * audio_data;
+- (instancetype)init NS_UNAVAILABLE;
+- (instancetype)initWithToken_data:(MLMultiArray *)token_data audio_data:(MLMultiArray *)audio_data NS_DESIGNATED_INITIALIZER;
+
+@end
+
+/// Model Prediction Output Type
+API_AVAILABLE(macos(10.15), ios(13.0), watchos(6.0), tvos(13.0)) __attribute__((visibility("hidden")))
+@interface whisper_decoder_implOutput : NSObject<MLFeatureProvider>
+
+/// cast_76 as multidimensional array of floats
+@property (readwrite, nonatomic, strong) MLMultiArray * cast_76;
+- (instancetype)init NS_UNAVAILABLE;
+- (instancetype)initWithCast_76:(MLMultiArray *)cast_76 NS_DESIGNATED_INITIALIZER;
+
+@end
+
+/// Class for model loading and prediction
+API_AVAILABLE(macos(10.15), ios(13.0), watchos(6.0), tvos(13.0)) __attribute__((visibility("hidden")))
+@interface whisper_decoder_impl : NSObject
+@property (readonly, nonatomic, nullable) MLModel * model;
+
+/**
+    URL of the underlying .mlmodelc directory.
+*/
++ (nullable NSURL *)URLOfModelInThisBundle;
+
+/**
+    Initialize whisper_decoder_impl instance from an existing MLModel object.
+
+    Usually the application does not use this initializer unless it makes a subclass of whisper_decoder_impl.
+    Such application may want to use `-[MLModel initWithContentsOfURL:configuration:error:]` and `+URLOfModelInThisBundle` to create a MLModel object to pass-in.
+*/
+- (instancetype)initWithMLModel:(MLModel *)model NS_DESIGNATED_INITIALIZER;
+
+/**
+    Initialize whisper_decoder_impl instance with the model in this bundle.
+*/
+- (nullable instancetype)init;
+
+/**
+    Initialize whisper_decoder_impl instance with the model in this bundle.
+
+    @param configuration The model configuration object
+    @param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
+*/
+- (nullable instancetype)initWithConfiguration:(MLModelConfiguration *)configuration error:(NSError * _Nullable __autoreleasing * _Nullable)error;
+
+/**
+    Initialize whisper_decoder_impl instance from the model URL.
+
+    @param modelURL URL to the .mlmodelc directory for whisper_decoder_impl.
+    @param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
+*/
+- (nullable instancetype)initWithContentsOfURL:(NSURL *)modelURL error:(NSError * _Nullable __autoreleasing * _Nullable)error;
+
+/**
+    Initialize whisper_decoder_impl instance from the model URL.
+
+    @param modelURL URL to the .mlmodelc directory for whisper_decoder_impl.
+    @param configuration The model configuration object
+    @param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
+*/
+- (nullable instancetype)initWithContentsOfURL:(NSURL *)modelURL configuration:(MLModelConfiguration *)configuration error:(NSError * _Nullable __autoreleasing * _Nullable)error;
+
+/**
+    Construct whisper_decoder_impl instance asynchronously with configuration.
+    Model loading may take time when the model content is not immediately available (e.g. encrypted model). Use this factory method especially when the caller is on the main thread.
+
+    @param configuration The model configuration
+    @param handler When the model load completes successfully or unsuccessfully, the completion handler is invoked with a valid whisper_decoder_impl instance or NSError object.
+*/
++ (void)loadWithConfiguration:(MLModelConfiguration *)configuration completionHandler:(void (^)(whisper_decoder_impl * _Nullable model, NSError * _Nullable error))handler API_AVAILABLE(macos(11.0), ios(14.0), watchos(7.0), tvos(14.0)) __attribute__((visibility("hidden")));
+
+/**
+    Construct whisper_decoder_impl instance asynchronously with URL of .mlmodelc directory and optional configuration.
+
+    Model loading may take time when the model content is not immediately available (e.g. encrypted model). Use this factory method especially when the caller is on the main thread.
+
+    @param modelURL The model URL.
+    @param configuration The model configuration
+    @param handler When the model load completes successfully or unsuccessfully, the completion handler is invoked with a valid whisper_decoder_impl instance or NSError object.
+*/
++ (void)loadContentsOfURL:(NSURL *)modelURL configuration:(MLModelConfiguration *)configuration completionHandler:(void (^)(whisper_decoder_impl * _Nullable model, NSError * _Nullable error))handler API_AVAILABLE(macos(11.0), ios(14.0), watchos(7.0), tvos(14.0)) __attribute__((visibility("hidden")));
+
+/**
+    Make a prediction using the standard interface
+    @param input an instance of whisper_decoder_implInput to predict from
+    @param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
+    @return the prediction as whisper_decoder_implOutput
+*/
+- (nullable whisper_decoder_implOutput *)predictionFromFeatures:(whisper_decoder_implInput *)input error:(NSError * _Nullable __autoreleasing * _Nullable)error;
+
+/**
+    Make a prediction using the standard interface
+    @param input an instance of whisper_decoder_implInput to predict from
+    @param options prediction options
+    @param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
+    @return the prediction as whisper_decoder_implOutput
+*/
+- (nullable whisper_decoder_implOutput *)predictionFromFeatures:(whisper_decoder_implInput *)input options:(MLPredictionOptions *)options error:(NSError * _Nullable __autoreleasing * _Nullable)error;
+
+/**
+    Make an asynchronous prediction using the standard interface
+    @param input an instance of whisper_decoder_implInput to predict from
+    @param completionHandler a block that will be called upon completion of the prediction. error will be nil if no error occurred.
+*/
+- (void)predictionFromFeatures:(whisper_decoder_implInput *)input completionHandler:(void (^)(whisper_decoder_implOutput * _Nullable output, NSError * _Nullable error))completionHandler API_AVAILABLE(macos(14.0), ios(17.0), watchos(10.0), tvos(17.0)) __attribute__((visibility("hidden")));
+
+/**
+    Make an asynchronous prediction using the standard interface
+    @param input an instance of whisper_decoder_implInput to predict from
+    @param options prediction options
+    @param completionHandler a block that will be called upon completion of the prediction. error will be nil if no error occurred.
+*/
+- (void)predictionFromFeatures:(whisper_decoder_implInput *)input options:(MLPredictionOptions *)options completionHandler:(void (^)(whisper_decoder_implOutput * _Nullable output, NSError * _Nullable error))completionHandler API_AVAILABLE(macos(14.0), ios(17.0), watchos(10.0), tvos(17.0)) __attribute__((visibility("hidden")));
+
+/**
+    Make a prediction using the convenience interface
+    @param token_data 1 by 1 matrix of floats
+    @param audio_data 1 × 1500 × 384 3-dimensional array of floats
+    @param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
+    @return the prediction as whisper_decoder_implOutput
+*/
+- (nullable whisper_decoder_implOutput *)predictionFromToken_data:(MLMultiArray *)token_data audio_data:(MLMultiArray *)audio_data error:(NSError * _Nullable __autoreleasing * _Nullable)error;
+
+/**
+    Batch prediction
+    @param inputArray array of whisper_decoder_implInput instances to obtain predictions from
+    @param options prediction options
+    @param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
+    @return the predictions as NSArray<whisper_decoder_implOutput *>
+*/
+- (nullable NSArray<whisper_decoder_implOutput *> *)predictionsFromInputs:(NSArray<whisper_decoder_implInput*> *)inputArray options:(MLPredictionOptions *)options error:(NSError * _Nullable __autoreleasing * _Nullable)error;
+@end
+
+NS_ASSUME_NONNULL_END