speech2text 0.3.4 → 0.3.6

data/lib/speech/audio_inspector.rb

@@ -14,7 +14,7 @@ module Speech
 
       def to_s
         s,f = seconds.split('.')
-        sprintf "%.2d:%.2d:%.2d:%.2d", self.hours.to_i, self.minutes.to_i, s.to_i, (f||0).to_i
+        sprintf "%.2d:%.2d:%.2d.%.2d", self.hours.to_i, self.minutes.to_i, s.to_i, (f||0).to_i
         #"#{hours}:#{minutes}:#{seconds}:#{f}"
       end
 
@@ -47,7 +47,13 @@ module Speech
     end
 
     def initialize(file)
-      self.duration = Duration.new(`ffmpeg -i #{file} 2>&1`.strip.scan(/Duration: (.*),/).first.first)
+      out = `ffmpeg -i #{file} 2>&1`.strip
+      if out.match(/No such file or directory/)
+        raise "No such file or directory: #{file}"
+      else
+        out = out.scan(/Duration: (.*),/)
+        self.duration = Duration.new(out.first.first)
+      end
     end
 
   end

data/lib/speech/audio_splitter.rb

@@ -25,13 +25,13 @@ module Speech
       # given the original file from the splitter and the chunked file name with duration and offset run the ffmpeg command
       def build
         return self if self.copied
-        # ffmpeg -y -i sample.audio.wav -acodec copy -vcodec copy -ss 00:00:00:00 -t 00:00:30:00 sample.audio.out.wav
-        offset_ts = AudioInspector::Duration.from_seconds(self.offset)
-        duration_ts = AudioInspector::Duration.from_seconds(self.duration)
+        # ffmpeg -y -i sample.audio.wav -acodec copy -vcodec copy -ss 00:00:00.00 -t 00:00:30.00 sample.audio.out.wav
+        offset_ts = AudioInspector::Duration.from_seconds(self.offset).to_s
+        duration_ts = AudioInspector::Duration.from_seconds(self.duration).to_s
         # NOTE: kind of a hack, but if the original source is less than or equal to 1 second, we should skip ffmpeg
-        puts "building chunk: #{duration_ts.inspect} and offset: #{offset_ts}"
+        #puts "building chunk: #{duration_ts.inspect} and offset: #{offset_ts}"
         #puts "offset: #{ offset_ts.to_s }, duration: #{duration_ts.to_s}"
-        cmd = "ffmpeg -y -i #{splitter.original_file} -acodec copy -vcodec copy -ss #{offset_ts} -t #{duration_ts} #{self.chunk}"# >/dev/null 2>&1"
+        cmd = "ffmpeg -y -i #{splitter.original_file} -acodec copy -vcodec copy -ss #{offset_ts} -t #{duration_ts} #{self.chunk}   >/dev/null 2>&1"
         if system(cmd)
           self
         else
@@ -41,10 +41,9 @@ module Speech
 
       # convert the audio file to flac format
       def to_flac
-        puts "convert: #{chunk} to flac"
-        if system("flac #{chunk}")
-          puts "success?"
-          self.flac_chunk = chunk.gsub(File.extname(chunk), ".flac")
+        chunk_outputfile = chunk.gsub(/#{File.extname(chunk)}$/, ".flac")
+        if system("ffmpeg -i #{chunk} -acodec flac #{chunk_outputfile} >/dev/null 2>&1")
+          self.flac_chunk = chunk.gsub(/#{File.extname(chunk)}$/, ".flac")
           # convert the audio file to 16K
           self.flac_rate = `ffmpeg -i #{self.flac_chunk} 2>&1`.strip.scan(/Audio: flac, (.*) Hz/).first.first.strip
           down_sampled = self.flac_chunk.gsub(/\.flac$/, '-sampled.flac')
@@ -72,7 +71,7 @@ module Speech
 
     end
 
-    def initialize(file, chunk_size=30)
+    def initialize(file, chunk_size=5)
       self.original_file = file
       self.duration = AudioInspector.new(file).duration
       self.size = chunk_size
@@ -86,7 +85,13 @@ module Speech
       #puts "generate: #{full_chunks} chunks of #{size} seconds, last: #{last_chunk} seconds"
 
       (full_chunks-1).times do|chunkid|
-        chunks << AudioChunk.new(self, chunkid * self.size, self.size)
+        if chunkid > 0
+          chunks << AudioChunk.new(self, chunkid * self.size, self.size)
+        else
+          off = (chunkid * self.size)-(self.size/2)
+          off = 0 if off < 0
+          chunks << AudioChunk.new(self, off, self.size)
+        end
       end
 
       if chunks.empty?
@@ -94,7 +99,7 @@ module Speech
       else
         chunks << AudioChunk.new(self, chunks.last.offset.to_i + chunks.last.duration.to_i, self.size + last_chunk)
       end
-      puts "Chunk count: #{chunks.size}"
+      #puts "Chunk count: #{chunks.size}"
 
       chunks
     end

data/lib/speech/audio_to_text.rb

@@ -2,63 +2,80 @@
 module Speech
 
   class AudioToText
-    attr_accessor :file, :rate, :captured_json, :captured_file
+    attr_accessor :file, :rate, :captured_json
+    attr_accessor :best_match_text, :score, :verbose, :segments
 
-    def initialize(file)
+    def initialize(file, options={})
+      self.verbose = false
       self.file = file
-      self.captured_file = self.file.gsub(/\.wav$/,'.json')
       self.captured_json = {}
+      self.best_match_text = ""
+      self.score = 0.0
+      self.segments = 0
+
+      self.verbose = !!options[:verbose] if options.key?(:verbose)
+    end
+
+    def to_text(max=2,lang="en-US")
+      to_json(max,lang)
+      self.best_match_text
     end
 
-    def to_text
-      url = "https://www.google.com/speech-api/v1/recognize?xjerr=1&client=speech2text&lang=en-US&maxresults=10"
+    def to_json(max=2,lang="en-US")
+      self.best_match_text = ""
+      self.score = 0.0
+      self.segments = 0
+
+      url = "https://www.google.com/speech-api/v1/recognize?xjerr=1&client=speech2text&lang=#{lang}&maxresults=#{max}"
       splitter = Speech::AudioSplitter.new(file) # 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
-
-    def clean
-      File.unlink self.captured_file if self.captured_file && File.exist?(self.captured_file)
+      self.best_match_text = self.best_match_text.strip
+      self.score /= self.segments
+      self.captured_json
     end
 
   protected
 
     def convert_chunk(easy, chunk, options={})
-      puts "sending chunk of size #{chunk.duration}..."
+      puts "sending chunk of size #{chunk.duration}..." if self.verbose
       retrying = true
       retry_count = 0
-      while retrying && retry_count < 5
-        #easy.verbose = true
+      while retrying && retry_count < 3 # 3 retries
+        easy.verbose = self.verbose
         easy.headers['Content-Type'] = "audio/x-flac; rate=#{chunk.flac_rate}"
         easy.headers['User-Agent'] = "https://github.com/taf2/speech2text"
-        #puts chunk.inspect
         easy.post_body = "Content=#{chunk.to_flac_bytes}"
-        easy.on_progress {|dl_total, dl_now, ul_total, ul_now| printf("%.2f/%.2f\r", ul_now, ul_total); true }
+        if self.verbose
+          easy.on_progress {|dl_total, dl_now, ul_total, ul_now| printf("%.2f/%.2f\r", ul_now, ul_total); true }
+        end
         easy.on_complete {|easy| puts }
         easy.http_post
-        #puts easy.header_str
-        #puts easy.body_str
         if easy.response_code == 500
-          puts "500 from google retry after 0.5 seconds"
+          puts "500 from google retry after 0.5 seconds" if self.verbose
           retrying = true
           retry_count += 1
           sleep 0.5 # wait longer on error?, google??
         else
-          # {"status":0,"id":"ce178ea89f8b17d8e8298c9c7814700a-1","hypotheses":[{"utterance":"I like pickles","confidence":0.92731786}]}
+          # {"status":0,"id":"ce178ea89f8b17d8e8298c9c7814700a-1","hypotheses":[{"utterance"=>"I like pickles", "confidence"=>0.59408695}, {"utterance"=>"I like turtles"}, {"utterance"=>"I like tickles"}, {"utterance"=>"I like to Kohl's"}, {"utterance"=>"I Like tickles"}, {"utterance"=>"I lyk tickles"}, {"utterance"=>"I liked to Kohl's"}]}
           data = JSON.parse(easy.body_str)
           self.captured_json['status'] = data['status']
           self.captured_json['id'] = data['id']
           self.captured_json['hypotheses'] = data['hypotheses'].map {|ut| [ut['utterance'], ut['confidence']] } 
-          puts self.captured_json.inspect
-          File.open("#{self.captured_file}", "wb") {|f| f << captured_json.to_json }
+          if data.key?('hypotheses') && ['hypotheses'].first
+            self.best_match_text += " " + data['hypotheses'].first['utterance']
+            self.score += data['hypotheses'].first['confidence']
+            self.segments += 1
+          end
           retrying = false
         end
         sleep 0.1 # not too fast there tiger
       end
+      puts "#{segments} processed: #{self.captured_json.inspect}" if self.verbose
+      self.captured_json
     ensure
       chunk.clean
     end

data/lib/speech/version.rb

@@ -1,6 +1,6 @@
 # -*- encoding: binary -*-
 module Speech
   class Info
-    VERSION='0.3.4'
+    VERSION='0.3.6'
   end
 end

data/test/audio_splitter_test.rb

@@ -6,10 +6,10 @@ require 'speech'
 class SpeechAudioSplitterTest < Test::Unit::TestCase
 
   def test_audio_splitter
-    splitter = Speech::AudioSplitter.new("samples/i-like-pickles.wav", 1)
+    splitter = Speech::AudioSplitter.new(File.expand_path(File.join(File.dirname(__FILE__),"samples/i-like-pickles.wav")), 1)
 
-    assert_equal '00:00:03:52', splitter.duration.to_s
-    assert_equal 3.52, splitter.duration.to_f
+    assert_equal '00:00:03.51', splitter.duration.to_s
+    assert_equal 3.51, splitter.duration.to_f
 
     chunks = splitter.split
     assert_equal 3, chunks.size

data/test/audio_to_text_test.rb

@@ -4,39 +4,19 @@ $:.unshift File.expand_path(File.dirname(__FILE__) + '/../lib')
 require 'speech'
 
 class SpeechAudioToTextTest < Test::Unit::TestCase
-  def test_audio_to_text
-    audio = Speech::AudioToText.new("samples/i-like-pickles.wav")
-    captured_json = audio.to_text
-    assert captured_json
-    assert captured_json.key?("hypotheses")
-    assert !captured_json['hypotheses'].empty?
-    assert captured_json.keys.include?('status')
-    assert captured_json.keys.include?('id')
-    assert captured_json.keys.include?('hypotheses')
+  def setup
+    super
+  end
 
-    assert_equal "I like pickles", captured_json['hypotheses'].first.first
-    assert captured_json['hypotheses'].first.last > 0.9
-#    {"hypotheses"=>[["I like pickles", 0.92731786]]}
-#    puts captured_json.inspect
-  ensure
-    audio.clean
+  def test_audio_to_text
+    audio = Speech::AudioToText.new(File.expand_path(File.join(File.dirname(__FILE__),"samples/i-like-pickles.wav")))
+    assert_equal "I like pickles", audio.to_text
   end
 
-  def test_short_audio_clip
-    audio = Speech::AudioToText.new("samples/i-like-pickles.chunk5.wav")
-    captured_json = audio.to_text
-    assert captured_json
-    assert captured_json.key?("hypotheses")
-    assert !captured_json['hypotheses'].empty?
-    #{"status"=>0, "id"=>"552de5ba35bb769ce3493ff113e158a8-1", "hypotheses"=>[["eagles", 0.7214844], ["pickles", nil], ["michaels", nil], ["giggles", nil], ["tickles", nil]]}
-    assert captured_json.keys.include?('status')
-    assert captured_json.keys.include?('id')
-    assert captured_json.keys.include?('hypotheses')
-    puts captured_json.inspect
-    assert_equal "eagles", captured_json['hypotheses'][0].first
-    assert_equal "pickles", captured_json['hypotheses'][1].first
-    #assert captured_json['confidence'] > 0.9
-  ensure
-    audio.clean
+  def test_longer_audio
+    audio = Speech::AudioToText.new(File.expand_path(File.join(File.dirname(__FILE__),"/SampleAudio.wav")), :verbose => true)
+    puts audio.to_text
+    puts audio.score
+    puts audio.segments
   end
 end

metadata

@@ -1,7 +1,8 @@
 --- !ruby/object:Gem::Specification
 name: speech2text
 version: !ruby/object:Gem::Version
-  version: 0.3.4
+  version: !binary |-
+    MC4zLjY=
   prerelease: 
 platform: ruby
 authors:
@@ -9,12 +10,11 @@ authors:
 autorequire: 
 bindir: bin
 cert_chain: []
-date: 2011-04-05 00:00:00.000000000 -04:00
-default_executable: 
+date: 2012-10-07 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: curb
-  requirement: &2163558380 !ruby/object:Gem::Requirement
+  requirement: !ruby/object:Gem::Requirement
     none: false
     requirements:
     - - ! '>='
@@ -22,10 +22,15 @@ dependencies:
         version: '0'
   type: :runtime
   prerelease: false
-  version_requirements: *2163558380
+  version_requirements: !ruby/object:Gem::Requirement
+    none: false
+    requirements:
+    - - ! '>='
+      - !ruby/object:Gem::Version
+        version: '0'
 - !ruby/object:Gem::Dependency
   name: json
-  requirement: &2163557840 !ruby/object:Gem::Requirement
+  requirement: !ruby/object:Gem::Requirement
     none: false
     requirements:
     - - ! '>='
@@ -33,7 +38,12 @@ dependencies:
         version: '0'
   type: :runtime
   prerelease: false
-  version_requirements: *2163557840
+  version_requirements: !ruby/object:Gem::Requirement
+    none: false
+    requirements:
+    - - ! '>='
+      - !ruby/object:Gem::Version
+        version: '0'
 description: Super powers of Google wrapped in a nice Ruby interface
 email: todd.fisher@gmail.com
 executables:
@@ -51,12 +61,10 @@ files:
 - test/audio_splitter_test.rb
 - test/audio_to_text_test.rb
 - test/SampleAudio.wav
-- test/samples/i-like-pickles.json
 - test/samples/i-like-pickles.wav
 - Rakefile
 - README.rdoc
 - speech2text.gemspec
-has_rdoc: true
 homepage: https://github.com/taf2/speech2text
 licenses: []
 post_install_message: 
@@ -77,8 +85,9 @@ required_rubygems_version: !ruby/object:Gem::Requirement
       version: '0'
 requirements: []
 rubyforge_project: 
-rubygems_version: 1.6.2
+rubygems_version: 1.8.24
 signing_key: 
 specification_version: 3
 summary: Speech to Text Library
 test_files: []
+has_rdoc: 

data/test/samples/i-like-pickles.json

@@ -1 +0,0 @@
-{"captured_json":[["I like pickles",0.92731786]],"confidence":0.92731786}