rhubarb-lip-sync/rhubarb/src/recognition/g2pRules.rb

# This script reads transformation rules for English G2P from a text file
# and generates C++ code.
# The rule format is described here: http://www.zompist.com/sounds.htm

$characterClasses = {}

def formatRule(searchValue, replaceValue, contextBegin, contextEnd)
  return nil if replaceValue == searchValue

  # Special case: search and replace values are character classes
  if $characterClasses.has_key?(searchValue) && $characterClasses.has_key?(replaceValue)
    searchCharacters = $characterClasses[searchValue]
    replaceCharacters = $characterClasses[replaceValue]
    count = [searchCharacters.length, replaceCharacters.length].min
    result = ''
    0.upto(count - 1) do |i|
      subrule = formatRule(searchCharacters[i], replaceCharacters[i], contextBegin, contextEnd)
      if subrule
        result << ' ' if !result.empty?
        result << subrule
      end
    end
    return result
  end
  
  # Special characters
  searchValue = Regexp.escape(searchValue)
  contextBegin = Regexp.escape(contextBegin)
  contextEnd = Regexp.escape(contextEnd)
  replaceValue.gsub!(/\$/, '$$')
  
  # Anchors
  contextBegin.sub!(/^\\\#/, '^')
  contextEnd.sub!(/\\\#$/, '$')
  
  hasContextBegin = contextBegin != ''
  hasContextEnd = contextEnd != ''
  regexString = searchValue;
  if hasContextBegin
    regexString = "(#{contextBegin})" + regexString
    replaceValue = "$1" + replaceValue
  end
  if hasContextEnd
    regexString = regexString + "(#{contextEnd})"
    replaceValue = replaceValue + (hasContextBegin ? "$2" : "$1")
  end
  
  # Optional parts
  regexString.gsub!(/\\\((.*?)\\\)/, '(?:\\1)?')
  
  # Fold repeated characters/classes
  regexString.gsub!(/([w])\1\1\1\1/, '\\1{5}')
  regexString.gsub!(/(\w)\1\1\1/, '\\1{4}')
  regexString.gsub!(/(\w)\1\1/, '\\1{3}')
  regexString.gsub!(/(\w)\1/, '\\1{2}')
  
  # Character classes
  regexString.gsub!(/./) do |ch|
    $characterClasses.has_key?(ch) ? "[#{$characterClasses[ch]}]" : ch
  end
  
  # C++ string escaping
  regexString.gsub!(/[\\"]/, '\\\\\\\\')
  replaceValue.gsub!(/[\\"]/, '\\\\\\\\')
  
  return "{ wregex(L\"#{regexString}\"), L\"#{replaceValue}\" },"
end

# Read rules
lines = File.read('../../lib/soundchange/english.sc', :encoding => 'iso-8859-1').split(/\r?\n/)

# Add supplementary rules
lines.push(
  '* There are a number of cases not covered by these rules.',
  '* Let\'s add some reasonable fallback rules.',
  'a/â/_',
  'e/@/_',
  'i/ë/_',
  'o/ö/_',
  'q/k/_'
)

# Parse character class definitions
characterClassLineCount = 0
lines.each_with_index do |line, index|
  # Skip comments
  next if line.start_with? '*'
  
  match = /^(.)=(.+)$/.match(line)
  if match
    characterClassLineCount = index + 1
  else
    break
  end
  
  name = match[1]
  value = match[2]
  $characterClasses[name] = value
end

# Parse rules and convert them to C++
File.open('g2pRules.cpp', 'w:UTF-8') do |file|
  file.print "// Generated by #{__FILE__}; don't modify by hand!\n\n"
  lines.drop(characterClassLineCount).each do |line|
    # Handle comments
    comment = /^\*(.*)$/.match(line)
    if comment
      file.puts "//#{comment[1]}"
      next
    end

    # Handle rules
    rule = /^(.+)\/(.*)\/(.*)_(.*)$/.match(line)
    if rule
      searchValue = rule[1]
      replaceValue = rule[2]
      contextBegin = rule[3]
      contextEnd = rule[4]
      
      file.puts formatRule(searchValue, replaceValue, contextBegin, contextEnd)
      next
    end
    
    raise "Invalid rule: #{line}"
  end
end