whispercpp 1.3.0 → 1.3.2

data/sig/whisper.rbs

@@ -0,0 +1,480 @@
+module Whisper
+  interface _Samples
+    def length: () -> Integer
+    def each: { (Float) -> void } -> void
+  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
+
+  LOG_LEVEL_NONE: Integer
+  LOG_LEVEL_INFO: Integer
+  LOG_LEVEL_WARN: Integer
+  LOG_LEVEL_ERROR: Integer
+  LOG_LEVEL_DEBUG: Integer
+  LOG_LEVEL_CONT: Integer
+
+  def self.lang_max_id: () -> Integer
+  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
+
+  class Context
+    def self.new: (path | ::URI::HTTP) -> instance
+
+    # transcribe a single file
+    # can emit to a block results
+    #
+    #   params = Whisper::Params.new
+    #   params.duration = 60_000
+    #   whisper.transcribe "path/to/audio.wav", params do |text|
+    #     puts text
+    #   end
+    #
+    def transcribe: (string, Params) -> self
+                  | (string, Params) { (String) -> void } -> self
+
+    def model_n_vocab: () -> Integer
+    def model_n_audio_ctx: () -> Integer
+    def model_n_audio_state: () -> Integer
+    def model_n_text_head: () -> Integer
+    def model_n_text_layer: () -> Integer
+    def model_n_mels: () -> Integer
+    def model_ftype: () -> Integer
+    def model_type: () -> String
+
+    # Yields each Whisper::Segment:
+    #
+    #   whisper.transcribe("path/to/audio.wav", params)
+    #   whisper.each_segment do |segment|
+    #     puts segment.text
+    #   end
+    #
+    # Returns an Enumerator if no block given:
+    #
+    #   whisper.transcribe("path/to/audio.wav", params)
+    #   enum = whisper.each_segment
+    #   enum.to_a # => [#<Whisper::Segment>, ...]
+    #
+    def each_segment: { (Segment) -> void } -> void
+                    | () -> Enumerator[Segment]
+
+    def model: () -> Model
+    def full_get_segment: (Integer nth) -> Segment
+    def full_n_segments: () -> Integer
+
+    # Language ID, which can be converted to string by Whisper.lang_str and Whisper.lang_str_full.
+    #
+    def full_lang_id: () -> 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) -> 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) -> Integer
+
+    # 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+.
+    #
+    #   full_get_segment_text(3) # => "ask not what your country can do for you, ..."
+    #
+    def full_get_segment_text: (Integer) -> String
+
+    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
+    # 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.
+    #
+    def full: (Params, Array[Float] samples, ?Integer n_samples) -> self
+            | (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.
+    # 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
+  end
+
+  class Params
+    def self.new: (
+      ?language: string,
+      ?translate: boolish,
+      ?no_context: boolish,
+      ?single_segment: boolish,
+      ?print_special: boolish,
+      ?print_progress: boolish,
+      ?print_realtime: boolish,
+      ?print_timestamps: boolish,
+      ?suppress_blank: boolish,
+      ?suppress_nst: boolish,
+      ?token_timestamps: boolish,
+      ?split_on_word: boolish,
+      ?initial_prompt: string | nil,
+      ?diarize: boolish,
+      ?offset: Integer,
+      ?duration: Integer,
+      ?max_text_tokens: Integer,
+      ?temperature: Float,
+      ?max_initial_ts: Float,
+      ?length_penalty: Float,
+      ?temperature_inc: Float,
+      ?entropy_thold: Float,
+      ?logprob_thold: Float,
+      ?no_speech_thold: Float,
+      ?new_segment_callback: new_segment_callback,
+      ?new_segment_callback_user_data: Object,
+      ?progress_callback: progress_callback,
+      ?progress_callback_user_data: Object,
+      ?encoder_begin_callback: encoder_begin_callback,
+      ?encoder_begin_callback_user_data: Object,
+      ?abort_callback: abort_callback,
+      ?abort_callback_user_data: Object,
+      ?vad: boolish,
+      ?vad_model_path: path | URI,
+      ?vad_params: Whisper::VAD::Params
+    ) -> instance
+
+    # params.language = "auto" | "en", etc...
+    #
+    def language=: (String) -> String # TODO: Enumerate lang names
+
+    def language: () -> String
+    def translate=: (boolish) -> boolish
+    def translate: () -> (true | false)
+    def no_context=: (boolish) -> boolish
+
+    # 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).
+    #
+    def single_segment: () -> (true | false)
+
+    def print_special=: (boolish) -> boolish
+
+    # 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.
+    #
+    def print_progress: () -> (true | false)
+
+    def print_realtime=: (boolish) -> boolish
+
+    # 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.
+    #
+    def print_timestamps=: (boolish) -> boolish
+
+    def print_timestamps: () -> (true | false)
+
+    def suppress_blank=: (boolish) -> boolish
+
+    # If true, suppresses blank outputs.
+    #
+    def suppress_blank: () -> (true | false)
+
+    def suppress_nst=: (boolish) -> boolish
+
+    # If true, suppresses non-speech-tokens.
+    #
+    def suppress_nst: () -> (true | false)
+
+    def token_timestamps=: (boolish) -> boolish
+
+    # If true, enables token-level timestamps.
+    #
+    def token_timestamps: () -> (true | false)
+
+    def split_on_word=: (boolish) -> boolish
+
+    # 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
+
+    # 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)
+
+    def diarize=: (boolish) -> boolish
+
+    # If true, enables diarization.
+    #
+    def diarize: () -> (true | false)
+
+    def offset=: (Integer) -> Integer
+
+    # Start offset in ms.
+    #
+    def offset: () -> Integer
+
+    def duration=: (Integer) -> Integer
+
+    # Audio duration to process in ms.
+    #
+    def duration: () -> Integer
+
+    def max_text_tokens=: (Integer) -> Integer
+
+    # Max tokens to use from past text as prompt for the decoder.
+    #
+    def max_text_tokens: () -> Integer
+
+    def temperature=: (Float) -> Float
+    def temperature: () -> Float
+    def max_initial_ts=: (Float) -> Float
+
+    # See https://github.com/openai/whisper/blob/f82bc59f5ea234d4b97fb2860842ed38519f7e65/whisper/decoding.py#L97
+    #
+    def max_initial_ts: () -> Float
+
+    def length_penalty=: (Float) -> Float
+    def length_penalty: () -> Float
+    def temperature_inc=: (Float) -> Float
+    def temperature_inc: () -> Float
+    def entropy_thold=: (Float) -> Float
+
+    # Similar to OpenAI's "compression_ratio_threshold"
+    #
+    def entropy_thold: () -> Float
+
+    def logprob_thold=: (Float) -> Float
+    def logprob_thold: () -> Float
+    def no_speech_thold=: (Float) -> Float
+    def no_speech_thold: () -> Float
+
+    # Sets new segment callback, called for every newly generated text segment.
+    #
+    #   params.new_segment_callback = ->(context, _, n_new, user_data) {
+    #     # ...
+    #   }
+    #
+    def new_segment_callback=: (new_segment_callback) -> new_segment_callback
+    def new_segment_callback: () -> (new_segment_callback | 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 progress callback, called on each progress update.
+    #
+    #   params.new_segment_callback = ->(context, _, progress, user_data) {
+    #     # ...
+    #   }
+    #
+    # +progress+ is an Integer between 0 and 100.
+    #
+    def progress_callback=: (progress_callback) -> progress_callback
+
+    def progress_callback: () -> (progress_callback | 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 when the encoder starts.
+    #
+    def encoder_begin_callback=: (encoder_begin_callback) -> encoder_begin_callback
+
+    def encoder_begin_callback: () -> (encoder_begin_callback | 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 to check if the process should be aborted.
+    #
+    #   params.abort_callback = ->(user_data) {
+    #     # ...
+    #   }
+    #
+    #
+    def abort_callback=: (abort_callback) -> abort_callback
+
+    def abort_callback: () -> (abort_callback | 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
+
+    # Enable VAD
+    #
+    def vad=: (boolish) -> boolish
+
+    def vad: () -> (true | false)
+
+    # Path to the VAD model
+    def vad_model_path=: (path | URI | nil) -> (path | URI | nil)
+
+    def vad_model_path: () -> (String | nil)
+
+    def vad_params=: (Whisper::VAD::Params) -> Whisper::VAD::Params
+    def vad_params: () -> (Whisper::VAD::Params)
+
+    # Hook called on new segment. Yields each Whisper::Segment.
+    #
+    #   whisper.on_new_segment do |segment|
+    #     # ...
+    #   end
+    #
+    def on_new_segment: { (Segment) -> void } -> void
+
+    # Hook called on progress update. Yields each progress Integer between 0 and 100.
+    #
+    def on_progress: { (Integer progress) -> void } -> void
+
+    # Hook called on encoder starts.
+    #
+    def on_encoder_begin: { () -> void } -> void
+
+    # Call block to determine whether abort or not. Return +true+ when you want to abort.
+    #
+    #   params.abort_on do
+    #     if some_condition
+    #       true # abort
+    #     else
+    #       false # continue
+    #     end
+    #   end
+    #
+    def abort_on: { (Object user_data) -> boolish } -> void
+  end
+
+  class Model
+    def self.pre_converted_models: () -> Hash[String, Model::URI]
+    def self.new: () -> instance
+    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_text_ctx: () -> Integer
+    def n_text_state: () -> Integer
+    def n_text_head: () -> Integer
+    def n_text_layer: () -> Integer
+    def n_mels: () -> Integer
+    def ftype: () -> Integer
+    def type: () -> String
+
+    class URI
+      def self.new: (string | ::URI::HTTP) -> instance
+      def to_path: -> String
+      def clear_cache: -> void
+    end
+  end
+
+  class Segment
+    # Start time in milliseconds.
+    #
+    def start_time: () -> Integer
+
+    # End time in milliseconds.
+    #
+    def end_time: () -> Integer
+
+    # Whether the next segment is predicted as a speaker turn.
+    def speaker_next_turn?: () -> (true | false)
+
+    def text: () -> String
+    def no_speech_prob: () -> Float
+  end
+
+  module VAD
+    class Params
+      def self.new: (
+        ?threshold: Float,
+        ?min_speech_duration_ms: Integer,
+        ?min_silence_duration_ms: Integer,
+        ?max_speech_duration_s: Float,
+        ?speech_pad_ms: Integer,
+        ?samples_overlap: Float
+      ) -> instance
+
+      # Probability threshold to consider as speech.
+      #
+      def threshold=: (Float) -> Float
+
+      def threshold: () -> Float
+
+      # Min duration for a valid speech segment.
+      #
+      def min_speech_duration_ms=: (Integer) -> Integer
+
+      def min_speech_duration_ms: () -> Integer
+
+      # Min silence duration to consider speech as ended.
+      #
+      def min_silence_duration_ms=: (Integer) -> Integer
+
+      def min_silence_duration_ms: () -> Integer
+
+      # Max duration of a speech segment before forcing a new segment.
+      def max_speech_duration_s=: (Float) -> Float
+
+      def max_speech_duration_s: () -> Float
+
+      # Padding added before and after speech segments.
+      #
+      def speech_pad_ms=: (Integer) -> Integer
+
+      def speech_pad_ms: () -> Integer
+
+      # Overlap in seconds when copying audio samples from speech segment.
+      #
+      def samples_overlap=: (Float) -> Float
+
+      def samples_overlap: () -> Float
+      def ==: (Params) -> (true | false)
+    end
+  end
+
+  class Error < StandardError
+    attr_reader code: Integer
+
+    def self.new: (Integer code) -> instance
+  end
+end

data/tests/helper.rb

@@ -0,0 +1,35 @@
+require "test/unit"
+require "whisper"
+require_relative "jfk_reader/jfk_reader"
+
+class TestBase < Test::Unit::TestCase
+  AUDIO = File.join(__dir__, "..", "..", "..", "samples", "jfk.wav")
+
+  class << self
+    def whisper
+      return @whisper if @whisper
+
+      @whisper = Whisper::Context.new("base.en")
+      params = Whisper::Params.new
+      params.print_timestamps = false
+      @whisper.transcribe(TestBase::AUDIO, params)
+    end
+  end
+
+  private
+
+  def whisper
+    self.class.whisper
+  end
+
+  module BuildOptions
+    load "ext/options.rb", self
+    Options.include self
+
+    def enable_config(name)
+    end
+
+    def arg_config(name)
+    end
+  end
+end

data/tests/jfk_reader/.gitignore

@@ -0,0 +1,5 @@
+Makefile
+jfk_reader.o
+jfk_reader.so
+jfk_reader.bundle
+jfk_reader.dll

data/tests/jfk_reader/extconf.rb

@@ -0,0 +1,3 @@
+require "mkmf"
+
+create_makefile("jfk_reader")

data/tests/jfk_reader/jfk_reader.c

@@ -0,0 +1,68 @@
+#include <ruby.h>
+#include <ruby/memory_view.h>
+#include <ruby/encoding.h>
+
+static VALUE
+jfk_reader_initialize(VALUE self, VALUE audio_path)
+{
+  rb_iv_set(self, "audio_path", audio_path);
+  return Qnil;
+}
+
+static bool
+jfk_reader_get_memory_view(const VALUE obj, rb_memory_view_t *view, int flags)
+{
+  VALUE audio_path = rb_iv_get(obj, "audio_path");
+  const char *audio_path_str = StringValueCStr(audio_path);
+  const int n_samples = 176000;
+  float *data = (float *)malloc(n_samples * sizeof(float));
+  short *samples = (short *)malloc(n_samples * sizeof(short));
+  FILE *file = fopen(audio_path_str, "rb");
+
+  fseek(file, 78, SEEK_SET);
+  fread(samples, sizeof(short), n_samples, file);
+  fclose(file);
+  for (int i = 0; i < n_samples; i++) {
+    data[i] = samples[i]/32768.0;
+  }
+
+  view->obj = obj;
+  view->data = (void *)data;
+  view->byte_size = sizeof(float) * n_samples;
+  view->readonly = true;
+  view->format = "f";
+  view->item_size = sizeof(float);
+  view->item_desc.components = NULL;
+  view->item_desc.length = 0;
+  view->ndim = 1;
+  view->shape = NULL;
+  view->sub_offsets = NULL;
+  view->private_data = NULL;
+
+  return true;
+}
+
+static bool
+jfk_reader_release_memory_view(const VALUE obj, rb_memory_view_t *view)
+{
+  return true;
+}
+
+static bool
+jfk_reader_memory_view_available_p(const VALUE obj)
+{
+  return true;
+}
+
+static const rb_memory_view_entry_t jfk_reader_view_entry = {
+  jfk_reader_get_memory_view,
+  jfk_reader_release_memory_view,
+  jfk_reader_memory_view_available_p
+};
+
+void Init_jfk_reader(void)
+{
+  VALUE cJFKReader = rb_define_class("JFKReader", rb_cObject);
+  rb_memory_view_register(cJFKReader, &jfk_reader_view_entry);
+  rb_define_method(cJFKReader, "initialize", jfk_reader_initialize, 1);
+}