Installation and Administration Guide for the Speech Connect for Cisco Unity Feature
Troubleshooting
Downloads: This chapterpdf (PDF - 105.0KB) The complete bookPDF (PDF - 3.01MB) | Feedback

Troubleshooting

Table Of Contents

Troubleshooting

Speech Connect Recognizes Some Names But Not Others

Speech Connect Does Not Recognize an Employee

Names Tuning sFTP Connection Error

Application Prompts Stutter During Call Session

Recognition Works But Transfers Fail

Calls to the Operator Fail

Changing the Logging Level of SSP

Running the Diagnostic Snapshot Procedure

Changing the Network Speed, Duplex, and Auto Negotiation Settings

Poor Audio Quality Is Experienced in a Call Session

Speech Connect Does Not Answer


Troubleshooting


This chapter describes common troubleshooting problems, and describes the diagnostic snapshot procedure that will help your Technical Support contact isolate a problem with your Speech Connect system.

See the following sections:

Speech Connect Recognizes Some Names But Not Others

Speech Connect Does Not Recognize an Employee

Names Tuning sFTP Connection Error

Application Prompts Stutter During Call Session

Recognition Works But Transfers Fail

Calls to the Operator Fail

Changing the Logging Level of SSP

Running the Diagnostic Snapshot Procedure

Changing the Network Speed, Duplex, and Auto Negotiation Settings

Poor Audio Quality Is Experienced in a Call Session

Speech Connect Does Not Answer

Speech Connect Recognizes Some Names But Not Others

Problem   Until you install your Speech Connect license, the system includes only 50 employee and 50 department records in the speakable directory. This limitation is not enforced during data import and management, but is enforced when the directory is compiled for call recognition. When the grammar is compiled, only the first 50 employees sorted by first name/last name and the first 50 departments sorted by department code can be reached.

Solution   Install your Speech Connect license (see the "Installing the Speech Connect License" section on page 5-2) and rebuild the grammar (see the "Grammar Setup" section on page 6-11).

Speech Connect Does Not Recognize an Employee

Problem   A new employee was added to the system, but when you call Speech Connect and speak the name, recognition fails for that employee.

Solution   Review the following list:

Verify that the employee extension is set up in call control and is working. Call Speech Connect and enter the employee extension on the phone keypad. If a TTS recording of the employee name does not play prior to the transfer, the grammar rebuild and TTS generation process may not have been run yet.

In the Speech Connect interface, click Employees and then click Find/Modify to find the employee record. If no record exists, create the employee record and rebuild the grammar (see the "Grammar Setup" section on page 6-11).

Find the employee record, and verify that an office extension or phone number exists and was entered correctly. Verify that the employee name is spelled correctly and does not contain any illegal characters.

In the employee record, verify that the employee is listed in the external directory (List in External Directory is set to "on"), if required by the Speech Connect application that you are calling. If List in External Directory is set to "no," click Config and then click Application to see which Grammar Package is used by the application. Click Config and then click Grammar to see if the grammar excludes internal-only employees (Include All Employees is set to "no").

The employee record may have been recently added, but the name may not have been added to the speakable directory. To make the name available for speech recognition, rebuild the grammar (see the "Grammar Setup" section on page 6-11).

If the employee record exists and the name was correctly added to the grammar, but it still is not working, try a different pronunciation of the employee name. For example, pronounce the name as it is spelled rather than the way it is supposed to be said. If you are able to reach the person by using an alternate pronunciation of the name, then it is possible that the default phonetic description for the name is inaccurate. If you subscribed to the Names Tuning service, the name will be sent for phonetic improvement on the next scheduled update (usually within 72 hours).

Names Tuning sFTP Connection Error

Problem   When you install the license for the Names Tuning service and validate the sFTP connection (see the "Installing the Names Tuning License" section on page 8-2), you may receive a connection error.

Solution   Check your corporate network Firewall settings. Be sure to allow outbound TCP Port 22 to the Names Tuning service URL that is listed in the licensing section of the Global Settings page in the Speech Connect interface.

Application Prompts Stutter During Call Session

Problem   Under certain circumstances, the load on the Speech Connect system can be high enough to cause application prompts to stutter during a call session. "Stuttering" means that the prompt starts to play, then stops briefly, and then continues while the call session is active. The Cisco MCS7825 chassis model is the most prone to stuttering when system activity or logging level settings are very high.

Solution   Review the following list:

Background processes such as the TTS process, scheduled tasks for Names Tuning, or cron tasks for data loading and management may be running. Be sure to run these processes and tasks during off-peak hours.

To isolate a problem, you may have been asked by a technical representative to increase the Speech Services Platform (SSP) logging level to Verbose. In some configurations, increasing the logging level to Verbose can cause application prompts to stutter. To reduce the possibility of stuttering, change the SSP logging level to a level lower than Verbose if you had previously increased the level in the past (for instructions, see the "Changing the Logging Level of SSP" section). The default SSP log level is "Warning," which should not cause any stuttering.

Recognition Works But Transfers Fail

Problem   When you call in to Speech Connect and say an employee or department name, the name is recognized but the call does not transfer.

Solution   In the Speech Connect interface (click Config > Dialing), configure dialing rules for outbound dialing of internal extensions and external phone numbers. For instructions, see the "Setting Up Dialing Rules" section on page 5-1.

Calls to the Operator Fail

Problem   Speech Connect recognizes employee names and departments and transfers the calls successfully, but calls to the operator fail.

Solution   In the Speech Connect interface (click Config > Application), configure applications with operator phone numbers for day, night, weekend, and holiday modes. For instructions, see the "Day and Night Settings" section on page 6-6 and the "Weekend and Holiday Settings" section on page 6-8.

Changing the Logging Level of SSP

Revised July 2008

Problem   To isolate a specific problem, you may need to change the Speech Services Platform (SSP) logging level.


Caution Changing the logging level should only be done under the direction of a Cisco technical representative.

Solution   You can not change the SSP logging level through the web interface, and instead must make the change from the command line. Do the following "To Change the SSP Logging Level" procedure.

To Change the SSP Logging Level


Step 1 Log in to the Speech Connect system as scadmin.

Step 2 Enter su - to become the root user.

Step 3 Enter su - seaa to become the seaa user.

Step 4 Enter cd config to change to the config directory.

Step 5 Edit the seaa_settings file by using an editor such as vi. For example, enter vi seaa_settings.

Step 6 Change the SSP_LOGLEVEL line to the desired level. Verify that the logging level is in double quotes, as in the following example:

SSP_LOGLEVEL="Verbose"


Note The available error levels are Warning (minimum), Info, and Verbose (Maximum). Normally, this value should be set to Warning. A technical representative may ask you to set it to Info or Verbose while troubleshooting.


Step 7 Save the file and exit the editor.

Step 8 Run commit_ssp_settings.sh.

Step 9 Exit back to the root user.

Step 10 Exit back to the scadmin user.

Step 11 Run sc_sys restart to restart Speech Connect.

The SSP software will now be running with the desired log level.


Running the Diagnostic Snapshot Procedure

The Speech Connect Diagnostic Snapshot procedure is designed to retrieve and package diagnostic information. The procedure collects system information and some current trace files into a compressed tar file. The snapshot can then be forwarded to Technical Support for further analysis. This procedure will only be required at the request of your Technical Support contact.

You can obtain the diagnostic snapshot file from the Speech Connect interface. If the interface is not available, you can run the procedure from the command line.


Note Diagnostic snapshot files may be too large to e-mail. Your Technical Support contact can advise you on the best file transfer method.


To Run the Diagnostic Snapshot from the Speech Connect Interface


Step 1 In the Speech Connect interface, click Config and then click Global Settings.

Step 2 In the Reporting section, click Save Diagnostic Information. The system builds the diagnostic snapshot, and provides a status message. In the status dialog box, you can view the file or save it to the location of your choice.

Step 3 Send the snapshot file to your Technical Support contact.


To Run the Diagnostic Snapshot from the Command Line


Step 1 From an SSH client, log in to the Speech Connect server as scadmin.

Step 2 At the command line prompt, run [scadmin@myhost ~]$ sudo sc_sys diag-snapshot.

This will output the following:

Initiated Snapshot:
Snapshot Complete: /opt/cisco/speechconnect/logs/SpeechConnect-snapshot.tar.gz
[scadmin@myhost ~]$

Step 3 From an sftp client on your desktop, connect to the Speech Connect server as scadmin.

Step 4 Change the directory to the /opt/cisco/speechconnect/logs directory.

Step 5 Download the snapshot file to your desktop by entering get SpeechConnect-snapshot.tar.gz.

Step 6 Send the snapshot file to your Technical Support contact.


Changing the Network Speed, Duplex, and Auto Negotiation Settings

Problem   The network speed, duplex, and auto negotiation settings should be correct but in some cases may not be set correctly at installation. If the settings are not correct, they must be modified after the system has been built.

Solution   To set the network speed, duplex, and auto negotiation settings and make the settings permanent, you need to edit the /etc/sysconfig/network-scripts/ifcfg-eth0 file. Do the following procedure.

To Edit the ifcfg-eth0 File


Step 1 Open the file # vi /etc/sysconfig/network-scripts/ifcfg-eth0.

Step 2 Append the following line:

ETHTOOL_OPTS="speed 100 duplex full autoneg off"

Step 3 Save and close the file. The new line will set the eth0 device to 100Mbs, full duplex, with the auto negotiation off at boot time.

Step 4 Reboot the server, or restart the networking service by entering the following line:

# /etc/init.d/network restart


Note This will disconnect all ssh or ftp sessions.


Depending on traffic and load, it may take up to one minute for the new settings to take effect.


Poor Audio Quality Is Experienced in a Call Session

Added February 2008

Troubleshooting audio quality issues on a VOIP network can be time consuming and complex. Audio issues can be introduced anywhere along the path of the audio conversation. This section cannot encompass the entire methodology for isolating and troubleshooting audio quality issues. However you can verify certain configuration settings, and failing that, audio can be captured at the Speech Connect interface for examination and possible trouble isolation.

Problem   Poor audio quality is experienced in a call session.

Solution   If poor audio quality is experienced in a call session, try the following solutions:

Verify that the Cisco Unified Communications Manager Media Termination Point is not being used. See the "To Verify That the Cisco Unified Communications Manager Media Termination Point Is Not Being Used" procedure.

Disable VAD. See the "To Disable VAD" procedure.

Capture and rebuild an RTP stream from the Speech Connect Console. See the "To Capture and Rebuild an RTP Stream from the Speech Connect Console" procedure.

To Verify That the Cisco Unified Communications Manager Media Termination Point Is Not Being Used


Step 1 Calls using the Cisco Unified Communications Manager Software MTP are sensitive to any load on Cisco Unified CM. When at all possible, use a hardware MTP resource or disable RFC2833 support in the Device Settings for a particular endpoint. This will keep Cisco Unified CM out of the RTP path. For details, see the "Integrating with Cisco Unified Communications Manager Version 5.x" section on page 10-9.

Step 2 See the "Media Termination Point Configuration" chapter of the Cisco Unified CallManager Administration Guide, Release 5.0(4), available at http://www.cisco.com/en/US/docs/voice_ip_comm/cucm/admin/5_0_4/ccmcfg/b04mtp.html.

Step 3 See the "Cisco Unified IP Phone Configuration" chapter of the Cisco Unified CallManager Administration Guide, Release 5.0(4), available at http://www.cisco.com/en/US/docs/voice_ip_comm/cucm/admin/5_0_4/ccmcfg/b06phone.html.


To Disable VAD


Step 1 Voice Activity Detection is not supported by Speech Connect and must be turned off in the Cisco Unified Communications Manager or PIMG/TIMG gateway. See the "Integrating with Cisco Unified Communications Manager Version 5.x" section on page 10-9 for instructions on disabling VAD.

Step 2 See the "Trunk Configuration" chapter of the Cisco Unified CallManager Administration Guide, Release 5.0(4), available at http://www.cisco.com/en/US/docs/voice_ip_comm/cucm/admin/5_0_4/ccmcfg/b06trunk.html.


To Capture and Rebuild an RTP Stream from the Speech Connect Console

Perform the following steps to capture and rebuild an RTP stream from the Speech Connect console. The stream can then be exported for playback.


Step 1 Log into the Speech Connect console as root.


Note Ethereal (now called Wireshark) is installed on the Speech Connect server and must be run graphically from the console. This method will not work via SSH.


Step 2 When logged in, click the Actions menu at the top of the screen, and select Run Application.

Step 3 In the Application Name box, enter Ethereal. If you did not log in as root, you will be prompted for the root password. Otherwise Ethereal will launch.

Step 4 In the Ethereal window, click Capture > Options. The Capture Options window appears.

Step 5 Verify that eth0 is selected in the Interface menu, and click Start. The Capture Status window appears.

Step 6 Ethereal is now capturing. Place a call into Speech Connect. When the call is completed and disconnected, click Stop. Ethereal will then load the captured packets.

Step 7 Click Statistics > RTP > Show All Streams. Ethereal will present a list of all captured RTP streams.

Step 8 Each audio conversation will consist of two RTP streams. Find one of the streams of the audio call, highlight it, and click Find Reverse. Ethereal will find the applicable reverse stream.

Step 9 Click Analyze. Ethereal will launch the RTP Stream Analysis window.

Step 10 Click Save Payload. The Payload window appears.

Step 11 Select .au as the format, and verify that the Both radio button is selected. Give the payload file a name, and click OK.

Step 12 A playable .au file is now in the /root directory. Speech Connect cannot play this file so it must be exported to a user PC through ftp, sftp or a USB drive plugged into the Speech Connect server. On the PC, this file can be played by using Windows Media Player or any other audio player that supports the .au format.

This file will be an accurate reconstruction of the actual conversation that occurred on the Speech Connect network interface card. Incoming packets are captured before the Speech Connect server processes them in any way, and outgoing packets are captured after Speech Connect performs any processing. If there are any audio issues heard in the outgoing stream from Speech Connect to the caller (for example a Speech Connect prompt), it can be determined that the problem is a Speech Connect issue. If there are any audio issues in the incoming stream (caller talking to Speech Connect), it can be determined that the problem is in the network or sending device.


Speech Connect Does Not Answer

Added May 2009

Problem   The web administration console erroneously permits the use of space characters in application names. Calls to DNIS numbers that are assigned to an application that has one or more space characters in the name will not answer.

Solution   Create a new application with no space characters in the name by copying any of the existing applications.

To Create a New Application With No Space Characters in the Name


Step 1 Log on to the web administration console as scadmin.

Step 2 Click Config > Application.

Step 3 Select the application that you want to copy.

Step 4 Click Create New from This Application.

Step 5 In the pop-up box that opens, enter a new name with no space characters, and then click OK.

Step 6 Click Config > DNIS.

Step 7 Change the Application Name for the failing DNIS and then click Update.

Step 8 Call the DNIS number to verify that the new application name works correctly.