whispercpp 1.3.6 → 1.3.8

data/sig/whisper.rbs

@@ -5,10 +5,10 @@ module Whisper
   end
 
   type log_callback = ^(Integer level, String message, Object user_data) -> void
-  type new_segment_callback = ^(Whisper::Context, void, Integer n_new, Object user_data) -> void
-  type progress_callback = ^(Whisper::Context, void, Integer progress, Object user_data) -> void
-  type encoder_begin_callback = ^(Whisper::Context, void, Object user_data) -> void
-  type abort_callback = ^(Whisper::Context, void, Object user_data) -> boolish
+  type new_segment_callback = ^(Whisper::Context, untyped, Integer n_new, Object user_data) -> void
+  type progress_callback = ^(Whisper::Context, untyped, Integer progress, Object user_data) -> void
+  type encoder_begin_callback = ^(Whisper::Context, untyped, Object user_data) -> void
+  type abort_callback = ^(Whisper::Context, untyped, Object user_data) -> boolish
 
   VERSION: String
   LOG_LEVEL_NONE: Integer
@@ -37,10 +37,24 @@ module Whisper
   def self.lang_id: (string name) -> Integer
   def self.lang_str: (Integer id) -> String
   def self.lang_str_full: (Integer id) -> String
-  def self.log_set: (log_callback, Object? user_data) -> log_callback
+  def self.log_set: (log_callback?, Object? user_data) -> log_callback
   def self.system_info_str: () -> String
 
+  module Output
+    module Context
+      def to_srt: () -> String
+      def to_webvtt: () -> String
+    end
+
+    module Segment
+      def to_srt_cue: () -> String
+      def to_webvtt_cue: () -> String
+    end
+  end
+
   class Context
+    include Output::Context
+
     def self.new: (String | path | ::URI::HTTP) -> instance
 
     # transcribe a single file
@@ -52,8 +66,11 @@ module Whisper
     #       puts text
     #     end
     #
-    def transcribe: (path, Params, ?n_processors: Integer) -> self
-                  | (path, Params, ?n_processors: Integer) { (String) -> void } -> self
+    # If `n_processors` is greater than 1, you cannot set any callbacks including
+    # new_segment_callback, progress_callback, encoder_begin_callback, abort_callback,
+    # and log_callback set by Whisper.log_set
+    def transcribe: (path, Whisper::Params, ?n_processors: Integer) -> self
+                  | (path, Whisper::Params, ?n_processors: Integer) { (String) -> void } -> self
 
     def model_n_vocab: () -> Integer
     def model_n_audio_ctx: () -> Integer
@@ -71,7 +88,7 @@ module Whisper
     #       puts segment.text
     #     end
     #
-    # Returns an Enumerator if no block given:
+    # Returns an `Enumerator` if no block given:
     #
     #     whisper.transcribe("path/to/audio.wav", params)
     #     enum = whisper.each_segment
@@ -88,25 +105,25 @@ module Whisper
     #
     def full_lang_id: () -> Integer
 
-    # Start time of a segment indexed by +segment_index+ in centiseconds (10 times milliseconds).
+    # Start time of a segment indexed by `segment_index` in centiseconds (10 times milliseconds).
     #
     #     full_get_segment_t0(3) # => 1668 (16680 ms)
     #
     def full_get_segment_t0: (Integer) -> Integer
 
-    # End time of a segment indexed by +segment_index+ in centiseconds (10 times milliseconds).
+    # End time of a segment indexed by `segment_index` in centiseconds (10 times milliseconds).
     #
     #     full_get_segment_t1(3) # => 1668 (16680 ms)
     #
     def full_get_segment_t1: (Integer) -> Integer
 
-    # Whether the next segment indexed by +segment_index+ is predicated as a speaker turn.
+    # Whether the next segment indexed by `segment_index` is predicated as a speaker turn.
     #
     #     full_get_segment_speacker_turn_next(3) # => true
     #
     def full_get_segment_speaker_turn_next: (Integer) -> (true | false)
 
-    # Text of a segment indexed by +segment_index+.
+    # Text of a segment indexed by `segment_index`.
     #
     #     full_get_segment_text(3) # => "ask not what your country can do for you, ..."
     #
@@ -114,36 +131,42 @@ module Whisper
 
     def full_get_segment_no_speech_prob: (Integer) -> Float
 
-    # Run the entire model: PCM -> log mel spectrogram -> encoder -> decoder -> text
-    # Not thread safe for same context
+    def full_n_vad_segments: () -> Integer
+
+    def full_get_vad_segment_t0: (Integer) -> Integer
+
+    def full_get_vad_segment_t1: (Integer) -> Integer
+
+    # Run the entire model: PCM -> log mel spectrogram -> encoder -> decoder -> text  
+    # Not thread safe for same context  
     # Uses the specified decoding strategy to obtain the text.
     #
-    # The second argument +samples+ must be an array of samples, respond to :length, or be a MemoryView of an array of float. It must be 32 bit float PCM audio data.
+    # The second argument `samples` must be an array of samples, respond to `:length`, or be a MemoryView of an array of float. It must be 32 bit float PCM audio data.
     #
-    def full: (Params, Array[Float] samples, ?Integer n_samples) -> self
-            | (Params, _Samples, ?Integer n_samples) -> self
+    def full: (Whisper::Params, Array[Float] samples, ?Integer n_samples) -> self
+            | (Whisper::Params, _Samples, ?Integer n_samples) -> self
 
-    # Split the input audio in chunks and process each chunk separately using whisper_full_with_state()
-    # Result is stored in the default state of the context
-    # Not thread safe if executed in parallel on the same context.
-    # It seems this approach can offer some speedup in some cases.
+    # Split the input audio in chunks and process each chunk separately using `whisper_full_with_state()`  
+    # Result is stored in the default state of the context  
+    # Not thread safe if executed in parallel on the same context.  
+    # It seems this approach can offer some speedup in some cases.  
     # However, the transcription accuracy can be worse at the beginning and end of each chunk.
     #
-    def full_parallel: (Params, Array[Float], ?Integer n_samples) -> self
-                     | (Params, _Samples, ?Integer n_samples) -> self
-                     | (Params, _Samples, ?Integer? n_samples, Integer n_processors) -> self
-
-    def to_srt: () -> String
-    def to_webvtt: () -> String
+    # If `n_processors` is greater than 1, you cannot set any callbacks including
+    # new_segment_callback, progress_callback, encoder_begin_callback, abort_callback,
+    # and log_callback set by Whisper.log_set
+    def full_parallel: (Whisper::Params, Array[Float], ?Integer n_samples) -> self
+                     | (Whisper::Params, _Samples, ?Integer n_samples) -> self
+                     | (Whisper::Params, _Samples, ?Integer? n_samples, Integer n_processors) -> self
 
     class Params
       def self.new: (
-        use_gpu: boolish,
-        flash_attn: boolish,
-        gpu_device: Integer,
-        dtw_token_timestamps: boolish,
-        dtw_aheads_preset: Integer,
-        dtw_n_top: Integer | nil,
+        ?use_gpu: boolish,
+        ?flash_attn: boolish,
+        ?gpu_device: Integer,
+        ?dtw_token_timestamps: boolish,
+        ?dtw_aheads_preset: Integer,
+        ?dtw_n_top: Integer | nil,
       ) -> instance
 
       def use_gpu=: (boolish) -> boolish
@@ -211,35 +234,35 @@ module Whisper
     def translate: () -> (true | false)
     def no_context=: (boolish) -> boolish
 
-    # If true, does not use past transcription (if any) as initial prompt for the decoder.
+    # If `true`, does not use past transcription (if any) as initial prompt for the decoder.
     #
     def no_context: () -> (true | false)
 
     def single_segment=: (boolish) -> boolish
 
-    # If true, forces single segment output (useful for streaming).
+    # If `true`, forces single segment output (useful for streaming).
     #
     def single_segment: () -> (true | false)
 
     def print_special=: (boolish) -> boolish
 
-    # If true, prints special tokens (e.g. <SOT>, <EOT>, <BEG>, etc.).
+    # If `true`, prints special tokens (e.g. <SOT>, <EOT>, <BEG>, etc.).
     #
     def print_special: () -> (true | false)
 
     def print_progress=: (boolish) -> boolish
 
-    # If true, prints progress information.
+    # If `true`, prints progress information.
     #
     def print_progress: () -> (true | false)
 
     def print_realtime=: (boolish) -> boolish
 
-    # If true, prints results from within whisper.cpp. (avoid it, use callback instead)
+    # If `true`, prints results from within whisper.cpp. (avoid it, use callback instead)
     #
     def print_realtime: () -> (true | false)
 
-    # If true, prints timestamps for each text segment when printing realtime.
+    # If `true`, prints timestamps for each text segment when printing realtime.
     #
     def print_timestamps=: (boolish) -> boolish
 
@@ -247,19 +270,19 @@ module Whisper
 
     def suppress_blank=: (boolish) -> boolish
 
-    # If true, suppresses blank outputs.
+    # If `true`, suppresses blank outputs.
     #
     def suppress_blank: () -> (true | false)
 
     def suppress_nst=: (boolish) -> boolish
 
-    # If true, suppresses non-speech-tokens.
+    # If `true`, suppresses non-speech-tokens.
     #
     def suppress_nst: () -> (true | false)
 
     def token_timestamps=: (boolish) -> boolish
 
-    # If true, enables token-level timestamps.
+    # If `true`, enables token-level timestamps.
     #
     def token_timestamps: () -> (true | false)
 
@@ -271,16 +294,16 @@ module Whisper
 
     def split_on_word=: (boolish) -> boolish
 
-    # If true, split on word rather than on token (when used with max_len).
+    # If `true`, split on word rather than on token (when used with max_len).
     #
     def split_on_word: () -> (true | false)
 
     def initial_prompt=: (_ToS) -> _ToS
     def carry_initial_prompt=: (boolish) -> boolish
 
-    # Tokens to provide to the whisper decoder as initial prompt
-    # these are prepended to any existing text context from a previous call
-    # use whisper_tokenize() to convert text to tokens.
+    # Tokens to provide to the whisper decoder as initial prompt  
+    # these are prepended to any existing text context from a previous call  
+    # use whisper_tokenize() to convert text to tokens.  
     # Maximum of whisper_n_text_ctx()/2 tokens are used (typically 224).
     #
     def initial_prompt: () -> (String | nil)
@@ -288,7 +311,7 @@ module Whisper
 
     def diarize=: (boolish) -> boolish
 
-    # If true, enables diarization.
+    # If `true`, enables diarization.
     #
     def diarize: () -> (true | false)
 
@@ -417,7 +440,7 @@ module Whisper
     #
     def on_new_segment: { (Segment) -> void } -> void
 
-    # Hook called on progress update. Yields each progress Integer between 0 and 100.
+    # Hook called on progress update. Yields each progress `Integer` between 0 and 100.
     #
     def on_progress: { (Integer progress) -> void } -> void
 
@@ -425,7 +448,7 @@ module Whisper
     #
     def on_encoder_begin: { () -> void } -> void
 
-    # Call block to determine whether abort or not. Return +true+ when you want to abort.
+    # Call block to determine whether abort or not. Return `true` when you want to abort.
     #
     #     params.abort_on do
     #       if some_condition
@@ -438,6 +461,9 @@ module Whisper
     def abort_on: { (Object user_data) -> boolish } -> void
   end
 
+  module LogSettable
+  end
+
   class Model
     def self.pre_converted_models: () -> Hash[String, Model::URI]
     def self.coreml_compiled_models: () -> Hash[Model::URI, Model::ZipURI]
@@ -468,6 +494,8 @@ module Whisper
   end
 
   class Segment
+    include Output::Segment
+
     type deconstructed_keys = {
       start_time: (Integer | nil),
       end_time: (Integer | nil),
@@ -498,21 +526,18 @@ module Whisper
 
     # Yields each Whisper::Token:
     #
-    #   whisper.each_segment.first.each_token do |token|
-    #     p token
-    #   end
+    #     whisper.each_segment.first.each_token do |token|
+    #       p token
+    #     end
     #
-    # Returns an Enumerator if no block is given:
+    # Returns an `Enumerator` if no block is given:
     #
-    #   whisper.each_segment.first.each_token.to_a # => [#<Whisper::Token>, ...]
+    #     whisper.each_segment.first.each_token.to_a # => [#<Whisper::Token>, ...]
     #
     def each_token: { (Token) -> void } -> void
                   | () -> Enumerator[Token]
-    def to_srt_cue: () -> String
-    def to_webvtt_cue: () -> String
-
 
-    #  Possible keys: :start_time, :end_time, :text, :no_speech_prob, :speaker_turn_next
+    #  Possible keys: `:start_time`, `:end_time`, `:text`, `:no_speech_prob`, `:speaker_turn_next`
     #
     #      whisper.each_segment do |segment|
     #        segment => {start_time:, end_time:, text:, no_speech_prob:, speaker_turn_next:}
@@ -522,7 +547,7 @@ module Whisper
     def deconstruct_keys: (Array[:start_time | :end_time | :text | :no_speech_prob | :speaker_turn_next | :n_tokens] | nil) -> deconstructed_keys
   end
 
-  module Token
+  class Token
     type deconstructed_keys = {
       id: (Integer | nil),
       tid: (Integer | nil),
@@ -563,7 +588,7 @@ module Whisper
 
     # [EXPERIMENTAL] Token-level timestamps with DTW
     #
-    # Do not use if you haven't computed token-level timestamps with dtw.
+    # Do not use if you haven't computed token-level timestamps with dtw.  
     # Roughly corresponds to the moment in audio in which the token was output.
     #
     def t_dtw: () -> Integer
@@ -574,14 +599,14 @@ module Whisper
 
     # Start time of the token.
     #
-    # Token-level timestamp data.
+    # Token-level timestamp data.  
     # Do not use if you haven't computed token-level timestamps.
     #
     def start_time: () -> Integer
 
     # End time of the token.
     #
-    # Token-level timestamp data.
+    # Token-level timestamp data.  
     # Do not use if you haven't computed token-level timestamps.
     #
     def end_time: () -> Integer
@@ -592,6 +617,336 @@ module Whisper
     def deconstruct_keys: (Array[:id | :tid | :probability | :log_probability | :pt | :ptsum | :t_dtw | :voice_length | :start_time | :end_time | :text] | nil) -> deconstructed_keys
   end
 
+  module Parakeet
+    extend LogSettable
+
+    VERSION: String
+
+    # Control logging output. The default behavior is to print to stderr.
+    #
+    def self.log_set: (nil, Object? user_data) -> nil
+                    | (^(Integer level, String message, Object user_data) -> void, Object? user_data) -> nil
+    def self.system_info_str: () -> String
+
+    class Context
+      include Output::Context
+
+      # Load a Parakeet model from the given file path.
+      #
+      def self.new: (String | path | ::URI::HTTP, ?Params) -> instance
+
+      # Transcribe a single audio file.
+      #
+      def transcribe: (path audio_file_path, Whisper::Parakeet::Params) -> self
+
+      # Run the entire model: PCM -> log mel spectrogram -> encoder -> decoder -> text.
+      # Not thread safe for the same context.
+      #
+      # The second argument `samples` must be an array of samples, respond to `:length`,
+      # or be a MemoryView of an array of float. It must be 32 bit float PCM audio data.
+      #
+      def full: (Whisper::Parakeet::Params, Array[Float] samples, ?Integer n_samples) -> self
+              | (Whisper::Parakeet::Params, _Samples, ?Integer n_samples) -> self
+
+      # Number of generated text segments.
+      #
+      def full_n_segments: () -> Integer
+
+      # Start time of a segment indexed by `segment_index` in centiseconds (10 times milliseconds).
+      #
+      #     full_get_segment_t0(3) # => 1668 (16680 ms)
+      #
+      def full_get_segment_t0: (Integer segment_index) -> Integer
+
+      # End time of a segment indexed by `segment_index` in centiseconds (10 times milliseconds).
+      #
+      #     full_get_segment_t1(3) # => 1668 (16680 ms)
+      #
+      def full_get_segment_t1: (Integer segment_index) -> Integer
+
+      # Text of a segment indexed by `segment_index`.
+      #
+      #     full_get_segment_text(3) # => "ask not what your country can do for you, ..."
+      #
+      def full_get_segment_text: (Integer segment_index) -> String
+
+      # Number of tokens in the segment indexed by `segment_index`.
+      #
+      def full_n_tokens: (Integer segment_index) -> Integer
+
+      # Text of the token indexed by `token_index` in the segment indexed by `segment_index`.
+      #
+      def full_get_token_text: (Integer segment_index, Integer token_index) -> String
+
+      # Token id of the token indexed by `token_index` in the segment indexed by `segment_index`.
+      #
+      def full_get_token_id: (Integer segment_index, Integer token_index) -> Integer
+
+      # Probability of the token indexed by `token_index` in the segment indexed by `segment_index`.
+      #
+      def full_get_token_p: (Integer segment_index, Integer token_index) -> Float
+
+      # Token data of the token indexed by `token_index` in the segment indexed by `segment_index`.
+      #
+      def full_get_token_data: (Integer segment_index, Integer token_index) -> Token
+
+      def model: () -> Model
+
+      # Yields each Whisper::Parakeet::Segment:
+      #
+      #     parakeet.transcribe("path/to/audio.wav", params)
+      #     parakeet.each_segment do |segment|
+      #       puts segment.text
+      #     end
+      #
+      # Returns an `Enumerator` if no block given:
+      #
+      #     parakeet.transcribe("path/to/audio.wav", params)
+      #     enum = parakeet.each_segment
+      #     enum.to_a # => [#<Whisper::Parakeet::Segment>, ...]
+      #
+      def each_segment: { (Segment) -> void } -> void
+                      | () -> Enumerator[Segment]
+
+      class Params
+        def self.new: (?use_gpu: boolish, ?gpu_device: Integer) -> instance
+        def use_gpu: () -> boolish
+        def use_gpu=: (boolish) -> boolish
+        def gpu_device: () -> Integer
+        def gpu_device=: (Integer) -> Integer
+      end
+    end
+
+    class Params
+      def self.new: (
+        ?n_threads: Integer,
+        ?offset_ms: Integer,
+        ?duration_ms: Integer,
+        ?no_context: boolish,
+        ?audio_ctx: Integer,
+        ?new_segment_callback: ^(Whisper::Parakeet::Context, untyped, Integer n_new, Object user_data) -> void,
+        ?new_segment_callback_user_data: Object,
+        ?new_token_callback: ^(Whisper::Parakeet::Context, untyped, Whisper::Parakeet::Token, Object user_data) -> void,
+        ?new_token_callback_user_data: Object,
+        ?progress_callback: ^(Whisper::Parakeet::Context, untyped, Integer progress, Object user_data) -> void,
+        ?progress_callback_user_data: Object,
+        ?encoder_begin_callback: ^(Whisper::Parakeet::Context, untyped, Object user_data) -> boolish,
+        ?encoder_begin_callback_user_data: Object,
+        ?abort_callback: ^(Object user_data) -> boolish,
+        ?abort_callback_user_data: Object
+      ) -> instance
+
+      # Number of threads to use.
+      #
+      def n_threads=: (Integer) -> Integer
+      def n_threads: () -> Integer
+
+      # Start offset in ms.
+      #
+      def offset_ms=: (Integer) -> Integer
+      def offset_ms: () -> Integer
+
+      # Audio duration to process in ms.
+      #
+      def duration_ms=: (Integer) -> Integer
+      def duration_ms: () -> Integer
+
+      # If `true`, does not use past transcription (if any) as context.
+      #
+      def no_context=: (boolish) -> boolish
+      def no_context: () -> (true | false)
+
+      # Overwrite the audio context size. `0` uses the default value.
+      #
+      def audio_ctx=: (Integer) -> Integer
+      def audio_ctx: () -> Integer
+
+      # Sets new segment callback, called for every newly generated text segment.
+      #
+      #     params.new_segment_callback = ->(context, _, n_new, user_data) {
+      #       # ...
+      #     }
+      #
+      def new_segment_callback=: (^(Whisper::Parakeet::Context, untyped, Integer n_new, Object user_data) -> void) -> (^(Whisper::Parakeet::Context, untyped, Integer n_new, Object user_data) -> void)
+      def new_segment_callback: () -> ((^(Whisper::Parakeet::Context, untyped, Integer n_new, Object user_data) -> void) | nil)
+
+      # Sets user data passed to the last argument of new segment callback.
+      #
+      def new_segment_callback_user_data=: (Object?) -> Object?
+      def new_segment_callback_user_data: () -> Object?
+
+      # Sets token callback, called for every newly predicted token.
+      #
+      def new_token_callback=: (^(Whisper::Parakeet::Context, untyped, Whisper::Parakeet::Token, Object user_data) -> void) -> (^(Whisper::Parakeet::Context, untyped, Whisper::Parakeet::Token, Object user_data) -> void)
+      def new_token_callback: () -> ((^(Whisper::Parakeet::Context, untyped, Whisper::Parakeet::Token, Object user_data) -> void) | nil)
+
+      # Sets user data passed to the last argument of token callback.
+      #
+      def new_token_callback_user_data=: (Object?) -> Object?
+      def new_token_callback_user_data: () -> Object?
+
+      # Sets progress callback, called on each progress update.
+      #
+      # +progress+ is an Integer between 0 and 100.
+      #
+      def progress_callback=: (^(Whisper::Parakeet::Context, untyped, Integer progress, Object user_data) -> void) -> (^(Whisper::Parakeet::Context, untyped, Integer progress, Object user_data) -> void)
+      def progress_callback: () -> ((^(Whisper::Parakeet::Context, untyped, Integer progress, Object user_data) -> void) | nil)
+
+      # Sets user data passed to the last argument of progress callback.
+      #
+      def progress_callback_user_data=: (Object?) -> Object?
+      def progress_callback_user_data: () -> Object?
+
+      # Sets encoder begin callback, called each time before the encoder starts.
+      #
+      # If it returns `false`, the computation is aborted.
+      #
+      def encoder_begin_callback=: (^(Whisper::Parakeet::Context, untyped, Object user_data) -> boolish) -> (^(Whisper::Parakeet::Context, untyped, Object user_data) -> boolish)
+      def encoder_begin_callback: () -> ((^(Whisper::Parakeet::Context, untyped, Object user_data) -> boolish) | nil)
+
+      # Sets user data passed to the last argument of encoder begin callback.
+      #
+      def encoder_begin_callback_user_data=: (Object?) -> Object?
+      def encoder_begin_callback_user_data: () -> Object?
+
+      # Sets abort callback, called each time before ggml computation starts.
+      #
+      def abort_callback=: (^(Object user_data) -> boolish) -> (^(Object user_data) -> boolish)
+      def abort_callback: () -> ((^(Object user_data) -> boolish) | nil)
+
+      # Sets user data passed to the last argument of abort callback.
+      #
+      def abort_callback_user_data=: (Object?) -> Object?
+      def abort_callback_user_data: () -> Object?
+
+      # Hook called on new segment. Yields each Whisper::Parakeet::Segment.
+      #
+      def on_new_segment: { (Segment) -> void } -> void
+
+      # Hook called on new token. Yields each Whisper::Parakeet::Token.
+      #
+      def on_new_token: { (Token) -> void } -> void
+
+      # Hook called on progress update. Yields each progress `Integer` between 0 and 100.
+      #
+      def on_progress: { (Integer progress) -> void } -> void
+
+      # Hook called each time before the encoder starts.
+      #
+      def on_encoder_begin: { () -> boolish } -> void
+
+      # Call block to determine whether abort or not. Return `true` when you want to abort.
+      #
+      def abort_on: { () -> boolish } -> void
+    end
+
+    class Segment
+      include Output::Segment
+
+      type deconstructed_keys = {
+        start_time: (Integer | nil),
+        end_time: (Integer | nil),
+        text: (String | nil)
+      }
+
+      # Start time in milliseconds.
+      #
+      def start_time: () -> Integer
+
+      # End time in milliseconds.
+      #
+      def end_time: () -> Integer
+
+      # Text of the segment.
+      #
+      def text: () -> String
+
+      # Yields each Whisper::Parakeet::Token:
+      #
+      #     parakeet.each_segment.first.each_token do |token|
+      #       p token
+      #     end
+      #
+      # Returns an `Enumerator` if no block is given:
+      #
+      #     parakeet.each_segment.first.each_token.to_a # => [#<Whisper::Parakeet::Token>, ...]
+      #
+      def each_token: { (Token) -> void } -> void
+                    | () -> Enumerator[Token]
+
+      # Possible keys: `:start_time`, `:end_time`, `:text`
+      #
+      def deconstruct_keys: (Array[:start_time | :end_time | :text] | nil) -> deconstructed_keys
+    end
+
+    class Token
+      type deconstructed_keys = {
+        id: (Integer | nil),
+        duration_idx: (Integer | nil),
+        duration_value: (Integer | nil),
+        frame_index: (Integer | nil),
+        probability: (Float | nil),
+        log_probability: (Float | nil),
+        start_time: (Integer | nil),
+        end_time: (Integer | nil),
+        word_start: ((true | false) | nil),
+        text: (String | nil),
+      }
+
+      # Token ID.
+      #
+      def id: () -> Integer
+
+      # Index into the model's durations array.
+      #
+      def duration_idx: () -> Integer
+
+      # Actual duration value.
+      #
+      def duration_value: () -> Integer
+
+      # Frame index of the token.
+      #
+      def frame_index: () -> Integer
+
+      # Probability of the token.
+      #
+      def probability: () -> Float
+
+      # Log probability of the token.
+      #
+      def log_probability: () -> Float
+
+      # Start time of the token in milliseconds.
+      #
+      def start_time: () -> Integer
+
+      # End time of the token in milliseconds.
+      #
+      def end_time: () -> Integer
+
+      # Whether this token is the start of a word.
+      #
+      def word_start?: () -> (true | false)
+
+      # Get the token text of the token.
+      #
+      def text: () -> String
+
+      def deconstruct_keys: (Array[:id | :duration_idx | :duration_value | :frame_index | :probability | :log_probability | :start_time | :end_time | :word_start | :text] | nil) -> deconstructed_keys
+    end
+
+    class Model
+      def n_vocab: () -> Integer
+      def n_audio_ctx: () -> Integer
+      def n_audio_state: () -> Integer
+      def n_audio_head: () -> Integer
+      def n_audio_layer: () -> Integer
+      def n_mels: () -> Integer
+      def ftype: () -> Integer
+    end
+  end
+
   module VAD
     class Params
       def self.new: (

data/test/helper.rb

@@ -5,6 +5,8 @@ require_relative "jfk_reader/jfk_reader"
 class TestBase < Test::Unit::TestCase
   AUDIO = File.join(__dir__, "fixtures", "jfk.wav")
 
+  Parakeet = Whisper::Parakeet
+
   class << self
     def whisper
       return @whisper if @whisper