gst-pocketsphinx-client.rb

require 'gst'
require 'thread'
Gst.init


class Recognizer
	

	def initialize()
		@result = ""
		# construct pipeline
		@pipeline = Gst::Parse.launch("appsrc name=appsrc ! audioconvert ! audioresample ! pocketsphinx name=asr ! filesink location=file.wav")
		
		# define input audio properties
		@appsrc = @pipeline.get_child("appsrc")
		caps = Gst::Caps.parse("audio/x-raw-int,rate=16000,channels=1,signed=true,endianness=1234,depth=16,width=16")
		@appsrc.set_property("caps", caps)
		
		# define behaviour for ASR output
		asr = @pipeline.get_child("asr")
		asr.signal_connect('partial_result') { |asr, text, uttid| 
			#puts "PARTIAL: " + text 
			@result = text 
		}
		asr.signal_connect('result') { |asr, text, uttid| 
			#puts "FINAL: " + text 
			@result = text 	
			@queue.push(1)
		}
		
		
		@queue = Queue.new
		
		# This returns when ASR engine has been fully loaded
		asr.set_property('configured', true)
		#@pipeline.pause
  	end
  	
	# Get current (possibly partial) recognition result
	def result
		@result
	end
  	
  	# Call this before starting a new recognition
  	def clear()
  		@result = ""
  		@queue.clear
  		@pipeline.pause
  	end
  	
  	# Feed new chunk of audio data to the recognizer
  	def feed_data(data)
  		@pipeline.play  		
		buffer=Gst::Buffer.new
	  	buffer.data = data
	  	@appsrc.push_buffer(buffer)
	end
	
	# Notify recognizer of utterance end
	def feed_end()
		@appsrc.end_of_stream()
	end
	
	# Wait for the recognizer to recognize the current utterance
	# Returns the final recognition result
	def wait_final_result()
		@queue.pop
		@pipeline.stop
		return @result
	end
end



rec = Recognizer.new 


files = ["/home/tanel/devel/cmusphinx/trunk/pocketsphinx/test/data/goforward.raw", "/home/tanel/devel/cmusphinx/trunk/pocketsphinx/test/data/something.raw"]


files.each { |f|

	rec.clear()
	File.open(f, "r") { |f|
		# read data in chunks and feed to recognizer	
		while buff = f.read(2*16000)
			puts "."
		    rec.feed_data(buff)
		    # this is not really needed, it simulates time delay from real audio stream
		    sleep(1)
		    puts "Current result: #{rec.result}"
	    end
	}
	
	rec.feed_end()
	result = rec.wait_final_result()
	puts "OK, recognition done: " + result 
	
}