News: Downloadable PDA Demos

Introduction

Technical Description

How To

Documents

Product Brief

White Paper

Supported Platforms

Consulting

License Terms

NFL Demo

Downloads

Secure Travel Demo

FAQ's for Users
& Developers

Registration Form and License Agreement

SDK Access

Client Access

Feedback

Other Sprex Products/Services

Exit

• How do I install the face demo download onto my PDA?
• I hear a stutter on 'Replay' (local).
• I can't tell which jack is which on the back of the sound card.
• I don't see the "Record" button.
• I see a record button but the VU meter doesn't move at all when I hold Record down and speak into the microphone.
• I see a record button and the VU meter moves a little, but only a little, when I hold Record down and speak into the microphone.
• If it still doesn't work, try the Java Console
• Java console displays an init error: "Japplet notinited"
• Java console displays a permission or security error
• Applet is very slow loading, Java console displays IOException error when trying to write to socket.
• It runs but it gives the wrong answer.

• How do I install the face demo download onto my PDA?
  First download the latest facedemo .zip file to your computer. Then unzip the .zip file by right clicking on the .zip file in Windows Explorer and selecting Extract All... and then extracting all the files so you can use them for extra help and to install on your Pocket PC 2002 PDA.

  Once extracted, read the instruction file facedemoguide.txt or facedemoguidehtml (whichever you prefer) for more details about how to install the face demo. You can use Notepad, Wordpad or most other text editors to review the documentation. You may even wish to install it on your PDA for more convenient access (e.g., using Pocket Word).

• I hear a stutter on 'Replay' (local).
  Often, this symptom for Java-based demos is accompanied by 'Netplay' (remote replay) being completely empty. In either case (or even in the case of the blue right-VU meter getting stuck at a level higher than zero), we recommend first pressing and holding the 'Talk' button down for five seconds while speaking anything into the microphone, then releasing 'Talk', and then waiting ten seconds while your audio has a chance to be sent across the Internet and your recognition result generated. Repeat this (hold 'Talk', speak five seconds, release 'Talk') up to four times (or until the red left VU meter shows good slow-speed movements reflecting your speech). After these steps, press 'Replay' and 'Netplay': if both replay your most recent audio utterance, then you can use the demo as normal. If this doesn't work, then you may need to exit your browser (typically IE) and restart it. In extreme cases, you may even benefit from rebooting your computer. Typically, all of these cases are transient or intermittent or persistent Java Virtual Machine issues in dealing with audio input (work-arounds are listed above).

  We recommend that you use our PocketPC-based demos, which don't have these Java Virtual Machine issues.

• I can't tell which jack is which on the back of the sound card
  It is sometimes difficult to read the jack labels stamped onto the back of your sound card (or on a side of your computer case). Here are the standard colors and ordering of jacks on the back of sound cards:

  BLUE	Line In
  RED	Mic In
  GREEN	Line Out
  BLACK	Speaker Out
  RS232	Joystick (15-pin connector)

  So the "line in" jack is the farthest away from the 15-pin joystick connector, and the "mic in" jack is one step closer.

• "I don't see the "Record" button.
  The applet is not running.
  • Enable Java in your browser settings.

    Your browser (usually Netscape Navigator or MS Internet Explorer) must be set to allow Java applets to execute.

    Within MSIE::Tools(menu)::Internet Options(menu selection):: Security(tab page):: Internet(icon selection):: Custom Level(button):: Settings:: Microsoft VM:: Java Permissions:: uncheck the "disable java" checkbox

  • Make sure Java 2 is installed.

    Java can be enabled within the browser even though Java 2 isn't yet installed or properly set up. Our ANSR Java demos are all Java 2 applets, with the ability to record audio from your sound card. Thus the Java 2 plugin must be installed, and Java's security settings must specify that audio recording is allowable. Should our Java demo not run, your Java Runtime Environment (JRE) may need to be updated. JRE allows our Java demos to run on Windows, and is available from Sun at http://www.java.com/en/download/manual.jsp. See the Java installation instructions for details.

• "I see a record button but the VU meter doesn't move at all when I hold Record down and speak into the microphone."
  Audio is not being recieved by the applet. This could be caused by:
  • The mic is switched off (if it has an on/off switch).
  • The mic volume control is set to zero (if it has a volume control).
  • The mic is not plugged in.
  • The mic is plugged into line in and the OS is set to do recording from mic in.
  • The mic is plugged into mic in and the OS is set to do recording from line in.
  • The OS volume controls are set to "mute"
  • The mic is plugged into the wrong jack. Generally, a microphone with a preamp should be plugged into line in, not mic in. And a microphone with no battery or power supply usually should be plugged into mic in, not line in. You might have to experiment, and try both ways.

  • Or, Java thinks it can't record. Set the Java security policy. Detailed instructions here. Note that updating Internet Explorer or Netscape Navigator can change the java.policy file, which will then need to be modified (again) as described in detail here.
    • Some machines have installed Java 2 in multiple locations. It is possible to fixed the wrong one, so check for others. (Start::Search::For Files Named::java.policy).
    • Is the file saved as plain text?
      
    • Is the new permission inserted in the right place in the file?
    • Is the final "};" still located at the end of the file.

• "I see a record button and the VU meter moves a little, but only a little, when I hold Record down and speak into the microphone."
  
  • The mic is too far from your mouth. Some microphones specify as little as 1cm distance from the corner of your mouth to the microphone element.
  • The mic volume control is set too low (if it has a volume control).
  • The OS volume controls are set too low.
  • The mic is plugged into line in but it can generate a strong enough signal for mic in. Some microphones generate a very weak signal, and they might need a pre-amplifier.

• "If it still doesn't run, try the Java Console."
  Try turning on the Java console. As any of our demo applets runs, it prints out status messages to the Java console, so you can more precisely diagnose what is or is not going on.

  How to display the Java console The Java Console displays status messages that tell what's going on while any of our demos is running. Here's how to turn on the Java Console.

  For MSIE, pull down the Tools menu, select the Internet Options item, click on Advanced, scroll down to Microsoft VM, and check the Java console enable checkbox.

  For Netscape Navigator, pull down the Communicator menu, select the Tools item which brings up a sub-menu, and select the Java Console item.

  If you see healthy-looking messages on the Java console, you should feel comfortable that things are working properly.

  Java Console Messages
  • The Java console display should say it is running the Java Plug-In version 1.3.1 or greater. If it is an older version, install a newer version.
  • The applet checks if it can record the correct audio format. The console should say

    Fmt ok for line 0: PCM_SIGNED, 16000.0 Hz, mono, little-endian audio data

    If not:
    • The security settings may not be enabled.
      See here for detailed instructions.
    • The sound card may not allow recording at 16000 samples per second., It is unusual, but it sometimes occurs that a low-end sound card may not be able to do recording at the required rate of 16000 samples per second. Since the speech recognition models used in these demos are based on, and therefore require as input, 16000 Hz data, this means your sound card is not compatible with the demos. Upgrade your sound card, or simply try another computer.

  • Java console displays an init error: "Japplet notinited"
    This probably means that the Java hasn't been given permission to record audio. To be sure, turn on the Java Console (see here) and reload, looking for permission or security error messages like the following one:
  • Permission or security error
    If the following message appears on your Java Console, Java hasn't been told to allow audio recording.

    Java.security.AccessControlException: access denied 
    (javax.sound.sampled.AudioPermission record)
    

    See the security settings part of the setup instructions for details on this.

• Applet loads and runs and responds to mouse clicks very slowly; the Java console displays IOException errors when trying to write to socket.
  Tom says: This combination of symptoms has occurred to me when (virtual) memory is nearly filled up. What happens has a few steps. First, memory fills up with what may be zombie processes, so there is only a tiny amount actually available for use (in my case only 13M). Second, the Java environment then can't do much because it needs more memory, for example starting the applet quickly (I had to wait up to 30 seconds before buttons displayed on the applet), or simultaneously recording from the audio card and sending data over the network. Third, because it is so slow, the server exceeds its timeout while waiting for data to arrive, and kills the channel, and then when the applet finally tries to write the data, the channel is already destroyed and the IOException is raised. So that's how these (and other) symptoms arise.

  To solve the problem, I reduced my memory usage by other processes. The simple method is to reboot, but you can also (under Windows 2000) do a control-alt-delete, select Task Manager, select Processes, look for large processes that aren't actually doing anything (for example I found a 90M RealAudio process that hadn't been playing any audio for several days, filling up the memory space), and select and kill the process. If you're not careful you could kill essential processes that would prevent the entire system from working (e.g., I killed explore.exe which prevented further display, so that meant I had to reboot).

• It runs but it gives the wrong answer
  
  • Can ANSR hear you clearly? If ANSR can't hear you, it won't be able to recognize what you say, and you'll get the wrong answer, or no answer at all. Anything from your voice to the microphone to the computer's settings can prevent your voice from coming through clearly. Go through the Audio checklist to make sure.

  • Is ANSR trained to understand your accent?
    • Speaker Adaptation: ANSR recognition results can normally be improved through speaker adaptation. The web demos do not include this capability at present, but the product allows developers to build applications which can periodically re-adapt ANSR's internal expectations to match a user's particular pronunciation pattern. In this way, even native Chinese speakers speaking English can be recognized with adequate accuracy even in demanding, high-accuracy applications such as financial amount entry -- where errors are potentially costly.
    • Language Models: Native speakers of different languages normally pronounce corresponding sounds differently. For this reason, ANSR needs to be trained on each different language.

      ANSR has pronunciation models trained for Spanish, German, Japanese, UK English, and US English. The demonstration system is using US English models, so if you are not a native speaker of American English, you can unfortunately expect poor recognition results from the demo; however, excellent results should be obtainable with language-appropriate models, which we can provide for you, or you can build yourself.

      Sprex customers are able to develop models for other languages using the HTK toolkit from Cambridge University and can plug them into ANSR applications.

      For this reason, customers like SINTEF, in Norway, purchase ANSR: they can now develop Norwegian speech recognition systems for networked applications, with HTK based acoustic models used in ANSR's action-oriented, network-distributed speech application architecture.