whisper.cpp 0.1.0

checksums.yaml

@@ -0,0 +1,7 @@
+---
+SHA256:
+  metadata.gz: 0e8d71af28ba0c9af9fce2d70c86c36b6ebfb2a14d0b93720c87df0102358cc1
+  data.tar.gz: af1711e65ce109501e96aff8ce986764997033e15bc9eb97f9252d7cee9c074f
+SHA512:
+  metadata.gz: 98d417a79f60594c4a903aaaf2661e2b358113013dec48f8afd3d0bc2bf753a0091021cf912f59a87291fcf2fb65c87225807f7315366c89d169ee1b9f711efe
+  data.tar.gz: 54584e200cd1132ed924e9172a558bd5c3831ef4e7a63bf27a7b97e6726e9d74191a7db261dbd9c21ca2debb85e5fca9178727bb2285b4abe04d0d83b58ede2e

data/.gitignore

@@ -0,0 +1,35 @@
+# Ignore bundler config
+/.bundle
+
+/ext/whisper.cpp
+/lib/libwhispercpp.so
+Makefile
+
+# Ignore Gem artifacts
+*.gem
+Gemfile.lock
+
+# Ignore log files
+/log/*.log
+
+# Ignore temp files
+/tmp
+/coverage/
+/doc/
+/.yardoc
+
+# Ignore OS-specific files
+.DS_Store
+Thumbs.db
+
+# Ignore editor-specific files
+.idea/
+/.vscode/
+/*.swp
+
+# Ignore pry history
+.pry_history
+
+# Ignore test coverage reports
+/coverage
+

data/Gemfile

@@ -0,0 +1,6 @@
+source 'https://rubygems.org'
+
+gemspec
+
+ruby File.read('.ruby-version')
+

data/LICENSE.md

@@ -0,0 +1,42 @@
+
+### **8. Additional Notes**
+
+- **Cross-Platform Considerations:** The build process may vary depending on the operating system. The provided `Rakefile` assumes a Unix-like environment.
+- **Error Handling:** Ensure you handle potential errors during the build process, such as network issues when cloning the repository.
+- **Permissions:** You may need appropriate permissions to write to certain directories or to install the CUDA Toolkit.
+
+### **9. Potential Issues and Solutions**
+
+- **Problem:** The `whisper.cpp` repository changes, breaking the build process.
+  - **Solution:** Pin the repository to a specific commit or tag by checking out a specific commit after cloning.
+
+    ```ruby
+    # After cloning
+    sh 'git checkout <commit-hash>'
+    ```
+
+- **Problem:** CUDA is not available on the system.
+  - **Solution:** Modify the build process to skip CUDA support if not available.
+
+    ```ruby
+    # In the Rakefile
+    ENV['GGML_CUDA'] = '1' if cuda_available?
+    ```
+
+    Implement a `cuda_available?` method to check for CUDA availability.
+
+### **10. Optional Enhancements**
+
+- **Automate Dependency Installation:** You could enhance the `Rakefile` to check for and install missing dependencies.
+- **Add Tests:** Implement unit tests to verify the gem's functionality.
+
+---
+
+By updating the `Rakefile` to include steps for cloning and building `whisper.cpp` from the GitHub repository, and adjusting the rest of the gem accordingly, you should be able to build and use the gem as intended.
+
+Please replace `'Your Name'` and `'your.email@example.com'` with your actual name and email in the `whisper.cpp.gemspec` file.
+
+---
+
+**Let me know if you need further assistance or if you encounter any issues during the build process.**
+

data/README.md

@@ -0,0 +1,11 @@
+# Whisper.cpp Ruby Gem
+
+This gem provides Ruby bindings for the `whisper.cpp` library, allowing you to transcribe audio files using the Whisper model.
+
+## Installation
+
+Add this line to your application's Gemfile:
+
+```ruby
+gem 'whisper.cpp'
+

data/Rakefile

@@ -0,0 +1,6 @@
+require 'rake/extensiontask'
+
+Rake::ExtensionTask.new 'whisper.cpp' do |ext|
+  ext.lib_dir = 'lib/whisper'
+end
+

data/ext/extconf.rb

@@ -0,0 +1,107 @@
+# ext/extconf.rb
+
+require 'fileutils'
+require 'rbconfig'
+
+# Set root_dir to the project root directory
+root_dir = File.expand_path('..', __dir__)
+
+# Define the whisper.cpp directory path
+whisper_dir = File.join(__dir__, 'whisper.cpp')
+
+puts "Root Directory: #{root_dir}"
+puts "Whisper.cpp Directory: #{whisper_dir}"
+
+# Clone or update the whisper.cpp repository
+if Dir.exist?(whisper_dir)
+  # If the directory exists, check if it's a git repository
+  Dir.chdir(whisper_dir) do
+    if system('git rev-parse --is-inside-work-tree > /dev/null 2>&1')
+      puts "Updating existing whisper.cpp repository..."
+      # Pull the latest changes
+      system 'git pull' or abort "Failed to update whisper.cpp repository"
+    else
+      # If it's not a git repository, remove it and clone again
+      puts "Removing non-git directory #{whisper_dir}"
+      FileUtils.rm_rf(whisper_dir)
+      puts "Cloning whisper.cpp repository..."
+      Dir.chdir(__dir__) do
+        system 'git clone https://github.com/ggerganov/whisper.cpp.git' or abort "Failed to clone whisper.cpp repository"
+      end
+    end
+  end
+else
+  # Clone the repository
+  puts "Cloning whisper.cpp repository..."
+  Dir.chdir(__dir__) do
+    system 'git clone https://github.com/ggerganov/whisper.cpp.git' or abort "Failed to clone whisper.cpp repository"
+  end
+end
+
+# Verify that the whisper.cpp directory now exists
+unless Dir.exist?(whisper_dir)
+  abort "Failed to find or create the whisper.cpp directory at #{whisper_dir}"
+end
+
+# Now, proceed to build libwhispercpp.so using the whisper.cpp Makefile
+Dir.chdir(whisper_dir) do
+  # Set environment variables for build settings
+  ENV['GGML_CUDA'] = '1'  # Enable CUDA support
+
+  puts "Building libwhispercpp.so with GGML_CUDA=#{ENV['GGML_CUDA']}..."
+
+  # Build libwhisper.a and libggml.a
+  unless system 'make clean && make -j libwhisper.a libggml.a'
+    abort "Failed to build libwhisper.a and libggml.a"
+  end
+
+  # Link the static libraries into a single shared library using g++
+  gcc_command = 'g++ -shared -o libwhispercpp.so ' \
+                '-Wl,--whole-archive libwhisper.a -Wl,--no-whole-archive libggml.a ' \
+                '-L$( [ -d /opt/cuda ] && echo /opt/cuda/lib || echo /usr/local/cuda/lib ) ' \
+                '-lcuda -lcudart -lcublas -lc -lm -lstdc++'
+
+  unless system gcc_command
+    abort "Failed to link libwhispercpp.so"
+  end
+
+  # Verify that libwhispercpp.so was created
+  source_lib = File.join(Dir.pwd, 'libwhispercpp.so')
+  unless File.exist?(source_lib)
+    abort "libwhispercpp.so not found after compilation"
+  end
+
+  # Copy the compiled library to the gem's lib directory
+  FileUtils.cp(source_lib, root_dir)
+
+  puts "Copied libwhispercpp.so to #{root_dir}"
+end
+
+puts "Compilation completed."
+
+# Create a no-op Makefile to prevent RubyGems from attempting further compilation
+makefile_content = <<~MAKEFILE
+  all:
+  	@echo 'libwhispercpp.so already built.'
+  install:
+  	@echo 'libwhispercpp.so already installed.'
+MAKEFILE
+
+File.open('Makefile', 'w') do |f|
+  f.write(makefile_content)
+end
+
+puts "Created a no-op Makefile to prevent further compilation."
+
+# After copying libwhispercpp.so
+
+# Path to the cloned whisper.cpp directory
+cloned_dir = whisper_dir
+
+# Remove the cloned whisper.cpp directory
+puts "Removing cloned whisper.cpp directory at #{cloned_dir}..."
+FileUtils.rm_rf(cloned_dir)
+
+puts "Removed cloned whisper.cpp directory."
+
+

data/lib/whisper/audio_processor.rb

@@ -0,0 +1,28 @@
+require 'open3'
+require 'tmpdir'
+
+module Whisper
+  class AudioProcessor
+    def self.convert_to_float_array(file_path)
+      # Use ffmpeg to convert audio to 16-bit PCM WAV at 16kHz mono
+      wav_file = "#{Dir.tmpdir}/temp_#{Time.now.to_i}_#{rand(1000)}.wav"
+      cmd = [
+        'ffmpeg', '-y', '-i', file_path,
+        '-ar', '16000', '-ac', '1', '-f', 'wav', wav_file
+      ]
+      stdout_str, stderr_str, status = Open3.capture3(*cmd)
+      raise "ffmpeg error: #{stderr_str}" unless status.success?
+
+      # Read the WAV file and extract the PCM data
+      data = File.binread wav_file
+      # Skip the WAV header (44 bytes) and unpack the PCM data
+      pcm_data = data[44..-1].unpack 's<*' # Little-endian 16-bit signed integers
+
+      # Normalize and convert to float32
+      pcm_data.map { |sample| sample / 32768.0 }
+    ensure
+      File.delete wav_file if File.exist? wav_file
+    end
+  end
+end
+

data/lib/whisper/model.rb

@@ -0,0 +1,76 @@
+require_relative '../whisper'
+require_relative 'audio_processor'
+
+module Whisper
+  class Model
+    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
+
+    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
+
+      n_segments = Whisper.whisper_full_n_segments @ctx
+      case format.downcase
+      when 'plaintext'
+        transcript = ''
+        n_segments.times do |i|
+          segment_text = Whisper.whisper_full_get_segment_text @ctx, i
+          transcript += segment_text
+        end
+        transcript
+      when 'srt'
+        srt_content = ''
+        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
+
+          srt_content += "#{i + 1}\n"
+          srt_content += "#{format_time_srt start_time} --> #{format_time_srt end_time}\n"
+          srt_content += "#{segment_text.strip}\n\n"
+        end
+        srt_content
+      else
+        raise "Unsupported format: #{format}"
+      end
+    end
+
+    def close
+      Whisper.whisper_free @ctx
+    end
+
+    private
+
+    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
+    end
+  end
+end
+

data/lib/whisper.cpp.rb

@@ -0,0 +1,9 @@
+require_relative 'whisper'
+require_relative 'whisper/audio_processor'
+require_relative 'whisper/model'
+
+module Whisper
+  # The module is intentionally left empty here.
+  # All classes and modules are defined in their respective files.
+end
+

data/lib/whisper.rb

@@ -0,0 +1,181 @@
+require 'ffi'
+
+module Whisper
+  extend FFI::Library
+
+  # Load the shared library
+  lib_path = File.expand_path '../../libwhispercpp.so', __FILE__
+  ffi_lib lib_path
+
+  # Enums and Constants
+  WHISPER_SAMPLING_GREEDY = 0
+  WHISPER_SAMPLING_BEAM_SEARCH = 1
+
+  # Enums for alignment heads preset
+  enum :whisper_alignment_heads_preset, [
+    :WHISPER_AHEADS_NONE,
+    :WHISPER_AHEADS_N_TOP_MOST,
+    :WHISPER_AHEADS_CUSTOM,
+    :WHISPER_AHEADS_TINY_EN,
+    :WHISPER_AHEADS_TINY,
+    :WHISPER_AHEADS_BASE_EN,
+    :WHISPER_AHEADS_BASE,
+    :WHISPER_AHEADS_SMALL_EN,
+    :WHISPER_AHEADS_SMALL,
+    :WHISPER_AHEADS_MEDIUM_EN,
+    :WHISPER_AHEADS_MEDIUM,
+    :WHISPER_AHEADS_LARGE_V1,
+    :WHISPER_AHEADS_LARGE_V2,
+    :WHISPER_AHEADS_LARGE_V3
+  ]
+
+  # Structs Definitions
+
+  # whisper_ahead struct
+  class WhisperAhead < FFI::Struct
+    layout(
+      :n_text_layer, :int,
+      :n_head, :int
+    )
+  end
+
+  # whisper_aheads struct
+  class WhisperAheads < FFI::Struct
+    layout(
+      :n_heads, :size_t,
+      :heads, :pointer # Pointer to array of WhisperAhead
+    )
+  end
+
+  # whisper_context_params struct
+  class WhisperContextParams < FFI::Struct
+    layout(
+      :use_gpu, :bool,
+      :flash_attn, :bool,
+      :gpu_device, :int,
+      :dtw_token_timestamps, :bool,
+      :dtw_aheads_preset, :whisper_alignment_heads_preset,
+      :dtw_n_top, :int,
+      :dtw_aheads, WhisperAheads,
+      :dtw_mem_size, :size_t
+    )
+  end
+
+  # whisper_token_data struct
+  class WhisperTokenData < FFI::Struct
+    layout(
+      :id, :int32,
+      :tid, :int32,
+      :p, :float,
+      :plog, :float,
+      :pt, :float,
+      :ptsum, :float,
+      :t0, :int64,
+      :t1, :int64,
+      :t_dtw, :int64,
+      :vlen, :float
+    )
+  end
+
+  # greedy sampling parameters
+  class WhisperGreedyParams < FFI::Struct
+    layout(
+      :best_of, :int
+    )
+  end
+
+  # beam search sampling parameters
+  class WhisperBeamSearchParams < FFI::Struct
+    layout(
+      :beam_size, :int,
+      :patience, :float
+    )
+  end
+
+  # whisper_full_params struct
+  class WhisperFullParams < FFI::Struct
+    layout(
+      :strategy, :int,
+      :n_threads, :int,
+      :n_max_text_ctx, :int,
+      :offset_ms, :int,
+      :duration_ms, :int,
+      :translate, :bool,
+      :no_context, :bool,
+      :no_timestamps, :bool,
+      :single_segment, :bool,
+      :print_special, :bool,
+      :print_progress, :bool,
+      :print_realtime, :bool,
+      :print_timestamps, :bool,
+      :token_timestamps, :bool,
+      :thold_pt, :float,
+      :thold_ptsum, :float,
+      :max_len, :int,
+      :split_on_word, :bool,
+      :max_tokens, :int,
+      :debug_mode, :bool,
+      :audio_ctx, :int,
+      :tdrz_enable, :bool,
+      :suppress_regex, :pointer,
+      :initial_prompt, :pointer,
+      :prompt_tokens, :pointer,
+      :prompt_n_tokens, :int,
+      :language, :pointer,
+      :detect_language, :bool,
+      :suppress_blank, :bool,
+      :suppress_non_speech_tokens, :bool,
+      :temperature, :float,
+      :max_initial_ts, :float,
+      :length_penalty, :float,
+      :temperature_inc, :float,
+      :entropy_thold, :float,
+      :logprob_thold, :float,
+      :no_speech_thold, :float,
+      :greedy, WhisperGreedyParams,
+      :beam_search, WhisperBeamSearchParams,
+      :new_segment_callback, :pointer,
+      :new_segment_callback_user_data, :pointer,
+      :progress_callback, :pointer,
+      :progress_callback_user_data, :pointer,
+      :encoder_begin_callback, :pointer,
+      :encoder_begin_callback_user_data, :pointer,
+      :abort_callback, :pointer,
+      :abort_callback_user_data, :pointer,
+      :logits_filter_callback, :pointer,
+      :logits_filter_callback_user_data, :pointer,
+      :grammar_rules, :pointer,
+      :n_grammar_rules, :size_t,
+      :i_start_rule, :size_t,
+      :grammar_penalty, :float
+    )
+  end
+
+  # Function Bindings
+
+  # Initialize context with params
+  attach_function :whisper_init_from_file_with_params, [:string, WhisperContextParams.by_value], :pointer
+
+  # 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
+
+  # Free functions
+  attach_function :whisper_free, [:pointer], :void
+
+  # Full transcription function
+  attach_function :whisper_full, [:pointer, WhisperFullParams.by_value, :pointer, :int], :int
+
+  # Number of segments
+  attach_function :whisper_full_n_segments, [:pointer], :int
+
+  # Get segment text
+  attach_function :whisper_full_get_segment_text, [:pointer, :int], :string
+
+  # Get segment start and end times
+  attach_function :whisper_full_get_segment_t0, [:pointer, :int], :int64
+  attach_function :whisper_full_get_segment_t1, [:pointer, :int], :int64
+end
+

data/script/console

@@ -0,0 +1,8 @@
+#!/usr/bin/env ruby
+
+require 'bundler/setup'
+require 'pry'
+require_relative '../lib/whisper.cpp'
+
+Pry.start
+

data/whisper.cpp.gemspec

@@ -0,0 +1,24 @@
+Gem::Specification.new do |spec|
+  spec.name          = 'whisper.cpp'
+  spec.version       = '0.1.0'
+  spec.authors       = ['Braulio Oliveira']
+  spec.email         = ['brauliobo@gmail.com']
+
+  spec.summary       = 'Ruby bindings for whisper.cpp'
+  spec.description   = 'A Ruby gem that provides bindings to the whisper.cpp library for speech transcription.'
+  # spec.homepage      = 'http://example.com/whisper.cpp'
+  spec.license       = 'MIT'
+
+  # Use git ls-files to specify files to include in the gem
+  spec.files = `git ls-files -z`.split("\x0")
+
+  # Specify the extension to build
+  spec.extensions = ['ext/extconf.rb']
+
+  # Dependencies
+  spec.add_dependency 'ffi', '~> 1.15'
+
+  spec.add_development_dependency 'rake-compiler'
+  spec.add_development_dependency 'pry'
+end
+

metadata

@@ -0,0 +1,99 @@
+--- !ruby/object:Gem::Specification
+name: whisper.cpp
+version: !ruby/object:Gem::Version
+  version: 0.1.0
+platform: ruby
+authors:
+- Braulio Oliveira
+autorequire:
+bindir: bin
+cert_chain: []
+date: 2024-09-27 00:00:00.000000000 Z
+dependencies:
+- !ruby/object:Gem::Dependency
+  name: ffi
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '1.15'
+  type: :runtime
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '1.15'
+- !ruby/object:Gem::Dependency
+  name: rake-compiler
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
+  type: :development
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
+- !ruby/object:Gem::Dependency
+  name: pry
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
+  type: :development
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
+description: A Ruby gem that provides bindings to the whisper.cpp library for speech
+  transcription.
+email:
+- brauliobo@gmail.com
+executables: []
+extensions:
+- ext/extconf.rb
+extra_rdoc_files: []
+files:
+- ".gitignore"
+- Gemfile
+- LICENSE.md
+- README.md
+- Rakefile
+- ext/extconf.rb
+- lib/whisper.cpp.rb
+- lib/whisper.rb
+- lib/whisper/audio_processor.rb
+- lib/whisper/model.rb
+- script/console
+- whisper.cpp.gemspec
+homepage:
+licenses:
+- MIT
+metadata: {}
+post_install_message:
+rdoc_options: []
+require_paths:
+- lib
+required_ruby_version: !ruby/object:Gem::Requirement
+  requirements:
+  - - ">="
+    - !ruby/object:Gem::Version
+      version: '0'
+required_rubygems_version: !ruby/object:Gem::Requirement
+  requirements:
+  - - ">="
+    - !ruby/object:Gem::Version
+      version: '0'
+requirements: []
+rubygems_version: 3.5.3
+signing_key:
+specification_version: 4
+summary: Ruby bindings for whisper.cpp
+test_files: []