Class: Whisper::Model

Inherits:
Object
  • Object
show all
Defined in:
lib/whisper/model.rb

Defined Under Namespace

Classes: TranscriptionResult

Instance Method Summary collapse

Constructor Details

#initialize(model_path) ⇒ Model

Returns a new instance of Model.



9
10
11
12
13
14
15
16
17
 
# File 'lib/whisper/model.rb', line 9

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?
end

Instance Method Details

#close ⇒ Object 



67
68
69
 
# File 'lib/whisper/model.rb', line 67

def close
  Whisper.whisper_free @ctx
end

#transcribe_from_file(audio_file_path, format: 'plaintext') ⇒ Object 



19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
 
# File 'lib/whisper/model.rb', line 19

def transcribe_from_file(audio_file_path, format: 'plaintext')
  # Load audio file and convert to float array
  audio_data = Whisper::AudioProcessor.convert_to_float_array audio_file_path

  # 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

  # 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

  # Call the whisper_full function
  result = Whisper.whisper_full @ctx, params, samples_ptr, n_samples
  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
  output = ''
  case format.downcase
  when 'plaintext'
    n_segments.times do |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

      output += "#{i + 1}\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