An Engine to translate voice input files between languages. http://blog.moski.me/2012/02/building-siri-arabic-support-arabic.html

voice2voice.rb

# encoding: utf-8

# 
# @Details: blog post http://blog.moski.me/2012/02/building-siri-arabic-support-arabic.html 
# @Install:
#   Make sure to have FFMPEG installed using
#     brew install ffmpeg   or   ports install ffmpeg
# 
#   Install the dependencies
#     gem install speech2text  
# @Example:
#   You can download a sample input audio file from here http://cl.ly/2d220o0I0n2q0H022K0A
#   
#   require './voice2voice.rb'
#   client = Voice2Voice.new('./time.wav', :ar, :en)
#   client.process!
#

require 'speech'
require 'open-uri'
require 'cgi'
require 'json'


# Extend the speech class to support arabic
module Speech
  class AudioToText
    def to_text(lang="ar")
      url = "https://www.google.com/speech-api/v1/recognize?xjerr=1&client=speech2text&lang=#{lang}&maxresults=2"
      splitter = Speech::AudioSplitter.new(file,300) # based off the wave file because flac doesn't tell us the duration
      easy = Curl::Easy.new(url)
      splitter.split.each do|chunk|
        chunk.build.to_flac
        convert_chunk(easy, chunk)
      end
      JSON.parse(File.read(self.captured_file))
    end
  end
end

# A module to translate text using the google translate page.
module Google
  class Translator
    # The URL for the google translate page.
    GOOGLE_TRANSLATE_SERVICE_URL = "http://translate.google.com"
    
    # Define the basic exceptions used in this class
    def self.Exception(*names)
      cl = Module === self ? self : Object
      names.each {|n| cl.const_set(n, Class.new(Exception))}
    end
    
    # Basic exceptions.
    Exception :MissingFromLanguage, :MissingToLanguage, :MissingTextLanguage, :TranslateServerIsDown
    
    def translate(from, to, from_text, options={})
      raise(MissingFromLanguage) if from.nil?
      raise(MissingToLanguage) if to.nil?
      raise(MissingTextLanguage) if from_text.nil?

      begin
        url = GOOGLE_TRANSLATE_SERVICE_URL + "/translate_a/t?client=t&text=#{from_text}&hl=#{from}&sl=auto&tl=#{to}&multires=1&prev=btn&ssel=0&tsel=4&uptl=#{to}&alttl=#{from}&sc=1"
        puts "Translate URL: #{url}"
        agent = 'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_7_2) AppleWebKit/535.2 (KHTML, like Gecko) Chrome/15.0.874.121 Safari/535.2'
        open(URI.escape(url), "User-Agent" => agent) do |stream|
          
          # Read the stream and conver it into a valid json format
          content = stream.read
          s = content.split(',').collect {|s| s == '' ? "\"\"" : s}.join(",")
          
          # parse the json output
          result = JSON.parse(s)
          raise(TranslateServerIsDown) if (!result || result.empty?)
          
          r1 = result[0][0][0]
          r2 = result[0][0][2]
          [r1, r2]
        end
      rescue Exception => e
         raise(TranslateServerIsDown)
      end
    end
  end
end


class Voice2Voice
  attr_accessor :file, :input_text, :output_text
  attr_accessor :from_lang, :to_lang
  
  # Need an Audio File, Language for this file and the
  # desired output language.
  def initialize(file, from=:ar, to=:en)
    self.file = file
    self.from_lang = from
    self.to_lang = to
  end
  
  def process!
    # Convert the Audio file To Text
    # The after math of this function:
    # 1.input_text: will be updated
    #
    voice2Text
    
    #
    # Convert the InputText generated from voice2Text to the to_lang
    # The after math of this function:
    # 1.output_text: will be updated
    #
    tranlsateText
    
    #
    # Generate an 
    #
    text2Voice
  end
  
  def voice2Text
    audio = Speech::AudioToText.new(file)
    self.input_text = audio.to_text(self.from_lang)["hypotheses"].first.first rescue nil
    puts "Arabic: #{self.output_text}"
    self.input_text
  end
  
  def tranlsateText
    return nil if self.input_text.nil?
    translator = Google::Translator.new
    self.output_text = translator.translate(from_lang,to_lang, self.input_text).first rescue nil
    puts "English: #{self.output_text}"
    self.output_text
  end
  
  
  def text2Voice
    url     = "http://translate.google.com/translate_tts?tl=en&q=#{self.output_text}" 
    agent   = 'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_7_2) AppleWebKit/535.2 (KHTML, like Gecko) Chrome/15.0.874.121 Safari/535.2'
    buffer  = open(URI.escape(url), 'UserAgent' => agent).read 

    mp3 = "./" + rand_string() + '.mp3' 
    File.open(mp3, 'w'){|f| f.write buffer}
    mp3
  end
  
  private 
  
  # Generate a random filename.
  def rand_string(size=20)        
    chars = ("a".."z").to_a + ("1".."9").to_a
    str   = Array.new(size, '').collect{chars[rand(chars.size)]}.join
  end
end