Flow and Sphinx woes

The dialplan for the entire flow of our project is finished. It will ask the caller to say the number they wish to dial, then repeat the number back. It will then ask the user if this number is correct. If correct, it will dial the phone number that was stated. If incorrect, the user will be asked to repeat the number again until Sphinx gets it correct, then it will dial it out.

We used Asterisk applications in conjunction with AGI (Asterisk Gateway Interface) which allowed us to use a shell script to execute a commands outside of Asterisk and establish a communication channel between our script and the dialplan.

We still need to clean up some debugging commands in the dialplan, and we need some more wav files for the menu.

SAME PROBLEM

Sphinx is just not reliable whatsoever. I’ve never gotten correct feedback with its speech recognition engine. Even a male-voice tester got correct feedback 1 in 4 times, and I’m pretty sure we could have gone to 1:5.

We need to figure a better solution or workaround for Sphinx.

Riddle of Sphinx

We created a basic dialplan on the Asterisk server that asks the caller to say the number they wish to dial, then using Sphinx speech recognition software translate that to text. We use a unix function called text2wave that will then convert this text into a wav file to be played back to the caller. The caller will then hear the number generated and they can confirm if it is correct or not. If it is, the caller will be notified that they will be connected shortly, and the server will call the number. If it is incorrect, then the caller will be asked to say the number again.

PROBLEM
Currently, Sphinx is acting erratically; i.e. sometime it just doesn’t do anything. Shawn* said it’s because our SIP provider sends it poorly formatted audio files that Sphinx can’t understand. I also noticed another problem with Sphinx. It worked perfectly well with a male voice, but with my voice it doesn’t translate the digits correctly. I’m not sure what we will do to tackle this problem.

We need to investigate these two issues.

*Special thanks to Shawn Van Avery for all his online tutorials and help.

Custom sound files for Asterisk

I recorded some .wav files that will act as our voice-menu using Audacity. I uploaded them to my Asterisk account, and tried to play them using the diaplan. Didn’t work. I played around with smaller .wav files, but nothing worked. I researched online, and it turns out that Asterisk doesn’t accept just ANY .wav file, it needs to have the format and codec for the specific type of file.

According to voip-info.org, “Asterisk has codecs for wav (pcm), gsm, g729, g726, and wav49…However, Asterisk does not understand ADPCM WAV files.”

I used the following command on the Asterisk server to convert our wav files to a format Asterisk will understand and playback:

sox input.wav -r 8000 -c 1 -s -w output.wav resample -ql

Luckily, sox was already installed on the server.

1-800-HelloWorld

We bought a toll-free DID (direct inward dial) number from Vitelity.net which is a SIP based VoIP provider, among other things. Our number is 1-866-457-0074. We wanted a number that had the name of our project in it – PORTEL – but Vitelity didn’t have it.

We added the configuration files needed for the ITP Asterisk Server to Yasmin’s Asterisk account. And we set up our Vitelity account to automatically route our DID to ITP’s Asterisk Server at asterisk.itp.tsoa.nyu.edu, which resolves to 128.122.151.44.

Currently, we have a simple Hello World working. We will be presenting the prototype we developed (see last post) on Thursday, while trying to get more than Hello World working on our Asterisk account.

Proof of concept

Watch a movie of our prototype here.

We developed a prototype, or more of a proof of concept, for Thursday’s class. Currently, using our prototype you can call a phone number by saying the following commands:

“Open Phone”
“607-555-6567″
“Press Enter”

Then the number you said will be dialed, and you’ll be connected.

Here’s how it all works. We have “Dragon Naturally Speaking” installed on the computer which handles all the voice-recognition, and our three commands. The “Open Phone” command opens our perl script; we wrote a perl sript that will accept any phone number beginning with the area code, then open Skype to call this phone number.

C’est tout!

Our design for Portel

Portel is the name of our project. We wanted a design that is light, attractive, and flexible in movement.

 

phone deBUGging

Yesterday, Shlomit and I started to hack into a phone, and this is what we found inside:

We’re probably going to look into softphones now.

What’s out there?

The AblePhone, a voice activated/adaptive phone. Limitations are price and conversations are not private. Dynamic living offers a Voice Activated Phone System which consists of a voice-activated speaker phone and a voice-activated phone dialer for $599; it also shares the same problem as AblePhone of no private conversations.

Asterisk is a free open source PBX and VoIP system. There’s also Tribox (formerly known as Asterisk @ Home), which is an installation package consisting of Asterisk plus other components to have a fully functional system on a PC or on-the-go. Note: Windows users need to install a Virtual Machine and a QEMU processor emulator. Here’s a great tutorial from VoipNet.

For voice recognition, there are a few open source solutions that are mostly Linux/Unix-based machines. Carnegie Mellon University developed SPHINX. Here is a very comprehensive list of other freeware and commerical Speech Recognition software.

HandiPhone is a blog to document our final project for Developing Assistive Technology.