Monday, June 17, 2013

Speech recognition script for Asterisk that uses Google speech API

Speech recognition for Asterisk

Speech recognition script for Asterisk that uses Google speech API.
This AGI script makes use of Google's speech recognition engine in order to render speech to text and return it back to the dialplan as an asterisk channel variable. See README for a complete list of supported languages.

Dependencies

Perl: The Perl Programming Language
perl-libwww: The World-Wide Web library for Perl
flac: Free Lossless Audio Codec
Internet access in order to contact google and get the speech data.
If you plan to use SSL you will need to install the 'IO-Socket-SSL' Perl module that implements an interface to SSL sockets.

Install

To install copy speech-recog.agi to your agi-bin directory.
Usually this is /var/lib/asterisk/agi-bin/
To make sure check your /etc/asterisk/asterisk.conf file

Usage

agi(speech-recog.agi,[lang],[timeout],[intkey],[NOBEEP]): Records from the current channel untill 3 seconds of silence are detected (setting the 'timeout' argument to -1 disables the silence detenction) or the interrupt key (# by default) is pressed. If NOBEEP is set, no beep sound is played back to the user to indicate the start of the recording. The recorded sound is send over to googles speech recognition service and the returned text string is assigned as the value of the channel variable 'utterance'.
The scripts sets the following channel variables:
status: Return status. 0 means success, non zero values indicating different errors.
utterance: The generated text string.
confidence: A value between 0 and 1 indicating the probability of a correct recognition.Values bigger than 0.90 usually mean that the resulted text is correct.
id: Some id string that googles engine returns, not very useful(?).
Asterisk dialplan examples:
In these examples googletts.agi script is used for speech synthesys:
 
 
;;Simple speech recognition
exten => 1234,1,Answer()
exten => 1234,n,agi(speech-recog.agi,en-US)
exten => 1234,n,Verbose(1,The text you just said is: ${utterance})
exten => 1234,n,Verbose(1,The probability to be right is: ${confidence})
exten => 1234,n,Hangup()

;;Speech recognition demo:
exten => 1235,1,Answer()
exten => 1235,n,agi(googletts.agi,"Say something in English, when done press the pound key.",en)
exten => 1235,n(record),agi(speech-recog.agi,en-US)
exten => 1235,n,Verbose(1,Script returned: ${status} , ${id} , ${confidence} , ${utterance})

;Check return status:
exten => 1235,n,GotoIf($["${status}" = "0"]?success:fail)

;Check the probability of a successful recognition:
exten => 1235,n(success),GotoIf($["${confidence}" > "0.8"]?playback:retry)

;Playback the text:
exten => 1235,n(playback),agi(googletts.agi,"The text you just said was...",en)
exten => 1235,n,agi(googletts.agi,"${utterance}",en)
exten => 1235,n,goto(end)

;Retry in case speech recognition wasn't successful:
exten => 1235,n(retry),agi(googletts.agi,"Can you please repeat more clearly?",en)
exten => 1235,n,goto(record)

exten => 1235,n(fail),agi(googletts.agi,"Failed to get speech data.",en)
exten => 1235,n(end),Hangup()

;;Voice dialing example
exten => 1236,1,Answer()
exten => 1236,n,agi(googletts.agi,"Please say the number you want to dial.",en)
exten => 1236,n(record),agi(speech-recog.agi,en-US)
exten => 1236,n,GotoIf($[$["${status}" = "0"] & $["${confidence}" > "0.8"]]?success:retry)

exten => 1236,n(success),goto(${utterance},1)

exten => 1236,n(retry),agi(googletts.agi,"Can you please repeat?",en)
exten => 1236,n,goto(record)


Under the folder wolfram you can find a sample agi script that in combination with speech-recog.agi sends queries to WolframAlpha and returns the answers as a dialplan variable that can be read back to the user.
See wolfram/README for details.
A diaplan example follows where you can dictate your question to WolframAplha and listen to the answer on your phone.
;WolframAlpha query demo:
exten => 1237,1,Answer()
exten => 1237,n,agi(googletts.agi,"What is your question?",en)

;;Record the question and render it to text:
exten => 1237,n(record),agi(speech-recog.agi,en-US)
exten => 1237,n,GotoIf($[$["${status}" = "0"] & $["${confidence}" > "0.8"]]?success:retry)

;;Submit the question to wolfram:
exten => 1237,n(success),agi(wolfram.agi,"${utterance}")

;;Playback the answer:
exten => 1237,n,agi(googletts.agi,"${wolfram_answer}",en)
exten => 1237,n,goto(end)

;;Retry in case speech recognition wasn't successful:
exten => 1237,n(retry),agi(googletts.agi,"Can you please repeat more clearly?",en)
exten => 1237,n,goto(record)

exten => 1237,n(end),Hangup()

License

The speech-recog script for asterisk is distributed under the GNU General Public License v2.

Download

You can get the latest stable version here.
Develompent snapshots are available in either zip or tar formats.
You can also clone the project with Git by running:
 
$ git clone git://github.com/zaf/asterisk-speech-recog
 
Source:
.... 

No comments:

Post a Comment