whisper.cpp 0.2.0 → 0.3.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 9233f65df607fba80d30ff4ae4a86dcbfaa1a750c88de487a2f98ace0a15de75
-  data.tar.gz: 157dcee2a04c9b514c3b36a92500532521aa964eed35ef9ffefd6f2d6b5fd40b
+  metadata.gz: 7d83008a506279672a2d585d35f232940c6d81ee6175ea866baa9965684fa4a8
+  data.tar.gz: 6e79798baa580cbcd6cdfb886f197bb6a7c1533a1b9ad9d6fa1d1b15cd708028
 SHA512:
-  metadata.gz: 5a1b01fb0ae991764d318f054369f1740d9f7d5324aa8a0deedf7e5bad9f1c8ed84573061c3a68dfda1b73c60bc5761dd2f4f21931c22414098ebe3b1cc9b17d
-  data.tar.gz: bae0d8a234b23f88fca2a363438a43989ad01d6e732b95987f740b999b6fa9c74f300de1033e338a141d85e29d5dff2827763b408f599c34bbed50fe58f5d441
+  metadata.gz: f41063a2b14007e68f7084c5c1e19e513d7ff89648cb6cb2fe3dfdd8982c868d20b45639799b3d57d210a7cd6aabaf7d5cc23d2a03b821f5dffa2a827dadf4a8
+  data.tar.gz: 5c69521555fde8f6080a437a927af31741978fc66efca0e825f4be058739ebd988e7df218b7cd485773ff99466c57361e256e684a912042e959191764845f342

data/lib/whisper/model.rb

@@ -3,81 +3,122 @@ require_relative 'audio_processor'
 
 module Whisper
   class Model
+    TranscriptionResult = Struct.new(:language, :output)
 
-    TranscriptionResult = Struct.new :language, :output
-
-    def initialize model_path
-      params = Whisper.whisper_context_default_params
-      # Modify params as needed
-      params[:use_gpu] = true
-      params[:gpu_device] = 0
-
-      @ctx = Whisper.whisper_init_from_file_with_params model_path, params
-      raise 'Failed to initialize Whisper model' if @ctx.null?
+    def initialize(model_path)
+      @model_path = model_path
+      @ctx = nil
     end
 
-    def transcribe_from_file(audio_file_path, format: 'plaintext')
+    def transcribe_from_file(audio_file_path, format: 'plaintext', **params)
       # Load audio file and convert to float array
-      audio_data = Whisper::AudioProcessor.convert_to_float_array audio_file_path
+      audio_data = Whisper::AudioProcessor.convert_to_float_array(audio_file_path)
+      transcribe_from_audio_data(audio_data, format: format, **params)
+    end
+
+    def transcribe_from_audio_data(audio_data, format: 'plaintext', **params)
+      init_whisper_context(params)
 
       # Prepare full params
-      params = Whisper.whisper_full_default_params Whisper::WHISPER_SAMPLING_GREEDY
-      params[:n_threads] = 4
-      params[:translate] = false
-      params[:language] = FFI::Pointer::NULL # Auto-detect language
+      full_params = default_full_params(params)
 
       # Prepare audio data pointer
       n_samples = audio_data.size
       samples_ptr = FFI::MemoryPointer.new(:float, n_samples)
-      samples_ptr.write_array_of_float audio_data
+      samples_ptr.write_array_of_float(audio_data)
 
-      # Call the whisper_full function
-      result = Whisper.whisper_full @ctx, params, samples_ptr, n_samples
+      # Call the whisper_full_parallel function
+      n_processors = params.fetch :n_processors, ENV['WHISPER_N_PROCS']&.to_i || 1
+      result = Whisper.whisper_full_parallel(@ctx, full_params, samples_ptr, n_samples, n_processors)
       raise 'Transcription failed' if result != 0
 
       # Retrieve detected language
       lang_id = Whisper.whisper_full_lang_id(@ctx)
       language = Whisper.whisper_lang_str(lang_id)
 
-      n_segments = Whisper.whisper_full_n_segments @ctx
+      # Retrieve the transcription output
+      n_segments = Whisper.whisper_full_n_segments(@ctx)
+      output = format_transcription(format, n_segments: n_segments)
+
+      TranscriptionResult.new(language, output)
+    end
+
+    def close
+      Whisper.whisper_free(@ctx) unless @ctx.nil?
+    end
+
+    private
+
+    def init_whisper_context(params)
+      return unless @ctx.nil?
+
+      ctx_params = Whisper.whisper_context_default_params
+
+      # Set user-provided context params
+      user_ctx_params = params.fetch(:context_params, {})
+      user_ctx_params.each do |key, value|
+        if ctx_params.members.include?(field)
+          ctx_params[key] = value
+        else
+          warn "Unknown context_param field: #{field}"
+        end
+      end
+      ctx_params[:gpu_device] = ENV['WHISPER_GPU']&.to_i || 0
+
+      # Initialize context
+      @ctx = Whisper.whisper_init_from_file_with_params(@model_path, ctx_params)
+      raise 'Failed to initialize Whisper model' if @ctx.null?
+    end
+
+    def default_full_params(params)
+      # Get default full params
+      strategy = params.fetch(:sampling_strategy, Whisper::WHISPER_SAMPLING_GREEDY)
+      full_params = Whisper.whisper_full_default_params(strategy)
+
+      # Set user-provided full params
+      user_full_params = params.fetch(:full_params, {})
+      user_full_params.each do |key, value|
+        if full_params.members.include?(field)
+          full_params[key] = value
+        else
+          warn "Unknown full_param field: #{field}"
+        end
+      end
+
+      full_params
+    end
+
+    def format_transcription(format, n_segments:)
       output = ''
       case format.downcase
       when 'plaintext'
         n_segments.times do |i|
-          segment_text = Whisper.whisper_full_get_segment_text @ctx, i
+          segment_text = Whisper.whisper_full_get_segment_text(@ctx, i)
           output += segment_text
         end
       when 'srt'
         n_segments.times do |i|
-          start_time   = Whisper.whisper_full_get_segment_t0(@ctx, i) / 100.0
-          end_time     = Whisper.whisper_full_get_segment_t1(@ctx, i) / 100.0
-          segment_text = Whisper.whisper_full_get_segment_text @ctx, i
+          start_time = Whisper.whisper_full_get_segment_t0(@ctx, i) / 100.0
+          end_time = Whisper.whisper_full_get_segment_t1(@ctx, i) / 100.0
+          segment_text = Whisper.whisper_full_get_segment_text(@ctx, i)
 
           output += "#{i + 1}\n"
-          output += "#{format_time_srt start_time} --> #{format_time_srt end_time}\n"
+          output += "#{format_time_srt(start_time)} --> #{format_time_srt(end_time)}\n"
           output += "#{segment_text.strip}\n\n"
         end
       else
         raise "Unsupported format: #{format}"
       end
-
-      TranscriptionResult.new language, output
-    end
-
-    def close
-      Whisper.whisper_free @ctx
+      output
     end
 
-    private
-
-    def format_time_srt seconds
-      hours   = (seconds / 3600).to_i
+    def format_time_srt(seconds)
+      hours = (seconds / 3600).to_i
       minutes = ((seconds % 3600) / 60).to_i
-      secs    = (seconds % 60).to_i
-      millis  = ((seconds - seconds.to_i) * 1000).to_i
-      format '%02d:%02d:%02d,%03d', hours, minutes, secs, millis
+      secs = (seconds % 60).to_i
+      millis = ((seconds - seconds.to_i) * 1000).to_i
+      format('%02d:%02d:%02d,%03d', hours, minutes, secs, millis)
     end
-
   end
 end
 

data/lib/whisper/version.rb

@@ -0,0 +1,5 @@
+module Whisper
+
+  VERSION = '0.3.0'
+
+end

data/lib/whisper.rb

@@ -29,6 +29,20 @@ module Whisper
     :WHISPER_AHEADS_LARGE_V3
   ]
 
+  # Enums for sampling strategy
+  enum :whisper_sampling_strategy, [
+    :WHISPER_SAMPLING_GREEDY,
+    :WHISPER_SAMPLING_BEAM_SEARCH
+  ]
+
+  # Callbacks
+  callback :whisper_new_segment_callback, [:pointer, :pointer, :int, :pointer], :void
+  callback :whisper_progress_callback, [:pointer, :pointer, :int, :pointer], :void
+  callback :whisper_encoder_begin_callback, [:pointer, :pointer, :pointer], :bool
+  callback :whisper_logits_filter_callback, [:pointer, :pointer, :pointer, :int, :pointer, :pointer], :void
+  callback :ggml_abort_callback, [:pointer], :bool
+  callback :ggml_log_callback, [:int, :string, :pointer], :void
+
   # Structs Definitions
 
   # whisper_ahead struct
@@ -77,6 +91,28 @@ module Whisper
     )
   end
 
+  # whisper_model_loader struct
+  #class WhisperModelLoader < FFI::Struct
+  #  callback :read_callback, [:pointer, :pointer, :size_t], :size_t
+  #  callback :eof_callback, [:pointer], :bool
+  #  callback :close_callback, [:pointer], :void
+
+  #  layout(
+  #    :context, :pointer,
+  #    :read, :read_callback,
+  #    :eof, :eof_callback,
+  #    :close, :close_callback
+  #  )
+  #end
+
+  # whisper_grammar_element struct
+  class WhisperGrammarElement < FFI::Struct
+    layout(
+      :type, :int,
+      :value, :uint32
+    )
+  end
+
   # greedy sampling parameters
   class WhisperGreedyParams < FFI::Struct
     layout(
@@ -134,15 +170,15 @@ module Whisper
       :no_speech_thold, :float,
       :greedy, WhisperGreedyParams,
       :beam_search, WhisperBeamSearchParams,
-      :new_segment_callback, :pointer,
+      :new_segment_callback, :whisper_new_segment_callback,
       :new_segment_callback_user_data, :pointer,
-      :progress_callback, :pointer,
+      :progress_callback, :whisper_progress_callback,
       :progress_callback_user_data, :pointer,
-      :encoder_begin_callback, :pointer,
+      :encoder_begin_callback, :whisper_encoder_begin_callback,
       :encoder_begin_callback_user_data, :pointer,
-      :abort_callback, :pointer,
+      :abort_callback, :ggml_abort_callback,
       :abort_callback_user_data, :pointer,
-      :logits_filter_callback, :pointer,
+      :logits_filter_callback, :whisper_logits_filter_callback,
       :logits_filter_callback_user_data, :pointer,
       :grammar_rules, :pointer,
       :n_grammar_rules, :size_t,
@@ -151,41 +187,162 @@ module Whisper
     )
   end
 
+  # Get default context params
+  attach_function :whisper_context_default_params, [], WhisperContextParams.by_value
+  # Get default full params
+  attach_function :whisper_full_default_params, [:int], WhisperFullParams.by_value
+
   # Function Bindings
 
   # Initialize context with params
   attach_function :whisper_init_from_file_with_params, [:string, WhisperContextParams.by_value], :pointer
+  attach_function :whisper_init_from_buffer_with_params, [:pointer, :size_t, WhisperContextParams.by_value], :pointer
+  #attach_function :whisper_init_with_params, [WhisperModelLoader.by_ref, WhisperContextParams.by_value], :pointer
 
-  # Get default context params
-  attach_function :whisper_context_default_params, [], WhisperContextParams.by_value
+  # Initialize context without state
+  attach_function :whisper_init_from_file_with_params_no_state, [:string, WhisperContextParams.by_value], :pointer
+  attach_function :whisper_init_from_buffer_with_params_no_state, [:pointer, :size_t, WhisperContextParams.by_value], :pointer
+  #attach_function :whisper_init_with_params_no_state, [WhisperModelLoader.by_ref, WhisperContextParams.by_value], :pointer
 
-  # Get default full params
-  attach_function :whisper_full_default_params, [:int], WhisperFullParams.by_value
+  # Initialize state
+  attach_function :whisper_init_state, [:pointer], :pointer
+
+  # OpenVINO functions
+  #attach_function :whisper_ctx_init_openvino_encoder_with_state, [:pointer, :pointer, :string, :string, :string], :int
+  #attach_function :whisper_ctx_init_openvino_encoder, [:pointer, :string, :string, :string], :int
 
   # Free functions
   attach_function :whisper_free, [:pointer], :void
+  attach_function :whisper_free_state, [:pointer], :void
+  attach_function :whisper_free_params, [:pointer], :void
+  attach_function :whisper_free_context_params, [:pointer], :void
+
+  # PCM to Mel spectrogram
+  attach_function :whisper_pcm_to_mel, [:pointer, :pointer, :int, :int], :int
+  attach_function :whisper_pcm_to_mel_with_state, [:pointer, :pointer, :pointer, :int, :int], :int
+
+  # Set custom Mel spectrogram
+  attach_function :whisper_set_mel, [:pointer, :pointer, :int, :int], :int
+  attach_function :whisper_set_mel_with_state, [:pointer, :pointer, :pointer, :int, :int], :int
+
+  # Encode
+  attach_function :whisper_encode, [:pointer, :int, :int], :int
+  attach_function :whisper_encode_with_state, [:pointer, :pointer, :int, :int], :int
+
+  # Decode
+  attach_function :whisper_decode, [:pointer, :pointer, :int, :int, :int], :int
+  attach_function :whisper_decode_with_state, [:pointer, :pointer, :pointer, :int, :int, :int], :int
+
+  # Tokenize
+  attach_function :whisper_tokenize, [:pointer, :string, :pointer, :int], :int
+  attach_function :whisper_token_count, [:pointer, :string], :int
+
+  # Language functions
+  attach_function :whisper_lang_max_id, [], :int
+  attach_function :whisper_lang_id, [:string], :int
+  attach_function :whisper_lang_str, [:int], :string
+  attach_function :whisper_lang_str_full, [:int], :string
+
+  # Auto-detect language
+  attach_function :whisper_lang_auto_detect, [:pointer, :int, :int, :pointer], :int
+  attach_function :whisper_lang_auto_detect_with_state, [:pointer, :pointer, :int, :int, :pointer], :int
+
+  # Model info
+  attach_function :whisper_n_len, [:pointer], :int
+  attach_function :whisper_n_len_from_state, [:pointer], :int
+  attach_function :whisper_n_vocab, [:pointer], :int
+  attach_function :whisper_n_text_ctx, [:pointer], :int
+  attach_function :whisper_n_audio_ctx, [:pointer], :int
+  attach_function :whisper_is_multilingual, [:pointer], :int
+
+  attach_function :whisper_model_n_vocab, [:pointer], :int
+  attach_function :whisper_model_n_audio_ctx, [:pointer], :int
+  attach_function :whisper_model_n_audio_state, [:pointer], :int
+  attach_function :whisper_model_n_audio_head, [:pointer], :int
+  attach_function :whisper_model_n_audio_layer, [:pointer], :int
+  attach_function :whisper_model_n_text_ctx, [:pointer], :int
+  attach_function :whisper_model_n_text_state, [:pointer], :int
+  attach_function :whisper_model_n_text_head, [:pointer], :int
+  attach_function :whisper_model_n_text_layer, [:pointer], :int
+  attach_function :whisper_model_n_mels, [:pointer], :int
+  attach_function :whisper_model_ftype, [:pointer], :int
+  attach_function :whisper_model_type, [:pointer], :int
+
+  # Get logits
+  attach_function :whisper_get_logits, [:pointer], :pointer
+  attach_function :whisper_get_logits_from_state, [:pointer], :pointer
+
+  # Token functions
+  attach_function :whisper_token_to_str, [:pointer, :int32], :string
+  attach_function :whisper_model_type_readable, [:pointer], :string
+
+  # Special tokens
+  attach_function :whisper_token_eot, [:pointer], :int32
+  attach_function :whisper_token_sot, [:pointer], :int32
+  attach_function :whisper_token_solm, [:pointer], :int32
+  attach_function :whisper_token_prev, [:pointer], :int32
+  attach_function :whisper_token_nosp, [:pointer], :int32
+  attach_function :whisper_token_not, [:pointer], :int32
+  attach_function :whisper_token_beg, [:pointer], :int32
+  attach_function :whisper_token_lang, [:pointer, :int], :int32
+
+  # Task tokens
+  attach_function :whisper_token_translate, [:pointer], :int32
+  attach_function :whisper_token_transcribe, [:pointer], :int32
+
+  # Timings
+  attach_function :whisper_print_timings, [:pointer], :void
+  attach_function :whisper_reset_timings, [:pointer], :void
+  attach_function :whisper_print_system_info, [], :string
 
   # Full transcription function
   attach_function :whisper_full, [:pointer, WhisperFullParams.by_value, :pointer, :int], :int
+  attach_function :whisper_full_with_state, [:pointer, :pointer, WhisperFullParams.by_value, :pointer, :int], :int
+
+  # Parallel processing
+  attach_function :whisper_full_parallel, [:pointer, WhisperFullParams.by_value, :pointer, :int, :int], :int
 
   # Number of segments
   attach_function :whisper_full_n_segments, [:pointer], :int
+  attach_function :whisper_full_n_segments_from_state, [:pointer], :int
 
-  # Get segment text
-  attach_function :whisper_full_get_segment_text, [:pointer, :int], :string
-
-  # Get segment start and end times
+  # Get segment info
   attach_function :whisper_full_get_segment_t0, [:pointer, :int], :int64
+  attach_function :whisper_full_get_segment_t0_from_state, [:pointer, :int], :int64
+
   attach_function :whisper_full_get_segment_t1, [:pointer, :int], :int64
+  attach_function :whisper_full_get_segment_t1_from_state, [:pointer, :int], :int64
 
-    # Get detected language ID
-  attach_function :whisper_full_lang_id, [:pointer], :int
+  attach_function :whisper_full_get_segment_speaker_turn_next, [:pointer, :int], :bool
+  attach_function :whisper_full_get_segment_speaker_turn_next_from_state, [:pointer, :int], :bool
 
-  # Convert language ID to string
-  attach_function :whisper_lang_str, [:int], :string
+  attach_function :whisper_full_get_segment_text, [:pointer, :int], :string
+  attach_function :whisper_full_get_segment_text_from_state, [:pointer, :int], :string
 
-  # void log_callback(int level, const char *msg, void *user_data);
-  callback :ggml_log_callback, [:int, :string, :pointer], :void
+  attach_function :whisper_full_n_tokens, [:pointer, :int], :int
+  attach_function :whisper_full_n_tokens_from_state, [:pointer, :int], :int
+
+  attach_function :whisper_full_get_token_text, [:pointer, :int, :int], :string
+  attach_function :whisper_full_get_token_text_from_state, [:pointer, :pointer, :int, :int], :string
+
+  attach_function :whisper_full_get_token_id, [:pointer, :int, :int], :int32
+  attach_function :whisper_full_get_token_id_from_state, [:pointer, :int, :int], :int32
+
+  attach_function :whisper_full_get_token_data, [:pointer, :int, :int], WhisperTokenData.by_value
+  attach_function :whisper_full_get_token_data_from_state, [:pointer, :int, :int], WhisperTokenData.by_value
+
+  attach_function :whisper_full_get_token_p, [:pointer, :int, :int], :float
+  attach_function :whisper_full_get_token_p_from_state, [:pointer, :int, :int], :float
+
+  # Language ID
+  attach_function :whisper_full_lang_id, [:pointer], :int
+  attach_function :whisper_full_lang_id_from_state, [:pointer], :int
+
+  # Benchmarks
+  attach_function :whisper_bench_memcpy, [:int], :int
+  attach_function :whisper_bench_memcpy_str, [:int], :string
+  attach_function :whisper_bench_ggml_mul_mat, [:int], :int
+  attach_function :whisper_bench_ggml_mul_mat_str, [:int], :string
 
   # Set the log callback
   attach_function :whisper_log_set, [:ggml_log_callback, :pointer], :void
@@ -196,6 +353,5 @@ module Whisper
   end
   # Set the no-op log callback to suppress logging
   Whisper.whisper_log_set NOOP_LOG_CALLBACK, FFI::Pointer::NULL unless ENV['WHISPER_DEBUG']
-
 end
 

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: whisper.cpp
 version: !ruby/object:Gem::Version
-  version: 0.2.0
+  version: 0.3.0
 platform: ruby
 authors:
 - Braulio Oliveira
 autorequire:
 bindir: bin
 cert_chain: []
-date: 2024-09-30 00:00:00.000000000 Z
+date: 2024-10-12 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: ffi
@@ -73,6 +73,7 @@ files:
 - lib/whisper.rb
 - lib/whisper/audio_processor.rb
 - lib/whisper/model.rb
+- lib/whisper/version.rb
 - script/console
 - whisper.cpp.gemspec
 homepage: