Browse
 
Tools
Rss Categories

 Featured
Release Notes

Reference Number: AA-01461 Views: 52970 100 Rating/ 2 Voters

15.1.200 (May 15, 2017):


Fixes:

  • Fixed an Media Server connectivity issue where recognizer SETUP responses did not have the required mode=record parameter in the returned Transport header. This appears to have affected a very small number of users - specifically users running Cisco UCCE when using the MRCPv1 (RTSP) protocol would receive connection failure indications for ASR sessions. Any users running this platform are recommended to upgrade to this version to correct this issue, or alternatively switching to MRCPv2 (SIP) protocol may be an option to workaround the issue.

15.1.100 (January 31, 2017):


Improvements and New Features:

  • Added new SNMP User-Based Security Model (USM) profiles, which are configurable using the Dashboard. These can be used in conjunction with the SNMPv3 secure extensions added in this release.
  • Added Secure (V3) Simple Network Management Protocol handling to the Manager Service, adding new V3_TRAP_PROFILE, V3_ENGINE_ID and V3_CONTEXT options, extending the previous V1 and V2 functionality.
  • Added option to specify the SNMP trap notification type using a new TRAP_PROTOCOL configuration setting. Previously trap notifications sent out were limited to V2 type only. Now all versions, including secure (V3), using USM security profiles has been added.
  • Added support for a new German (de-DE) model, as our ninth supported ASR language. This includes the additions of new German-language boolean, currency, date, digits, number, phone and time "builtin" grammars.
  • Significant improvements to overall pronunciation generation and processing for the Canadian French (fr-CA), British English (en-GB), Brazilian Portuguese (pt-BR), Mexican Spanish (es-MX), Colombian Spanish (es-CO) and Indian English (en-IN) ASR models. These improvements should be noticeable across many ASR applications using these languages and are part of our ongoing efforts to continuously improve our speech recognition performance across all supported languages.
  • Modified the LVSpeechPort client code to issue TTS requests in a distributed (round-robin) mode. This means that API and Media Server users that have more than one TTS Server will see these requests being distributed across all connected servers. Previously, any configured TTS servers were accessed only during failover situations. This change means that TTS servers can now perform load distribution during periods of heavy demand, allowing larger capacity systems to be configured.
  • Added new Speech Tuner functionality to maintain the current cursor position, selection and line being displayed in the Grammar Editor whenever changing views or selecting different grammars. You can now switch between grammars and maintain scroll / select positions between them. This significantly improves usability when working with multiple views and/or multiple grammars.
  • Added a new Command-Line Speech Tuner Data Export API mechanism, allowing users to integrate some of the data processing capabilities within the Speech Tuner into their own processes automation scripts. Licensed Speech Tuner users can now extract a number of data elements, such as audio files, decoded text and confidence scores from a specified .callsre Response File by passing the Speech Tuner a new .tunerapi file describing which data to extract, along with a target location. After performing the operation, the Speech Tuner will shut down, allowing the operation to be repeated as needed within business automation scripts.
  • Added a range of new Media Server features aimed at assisting different types of users, including those primarily working with multi-tenant applications. These features are exposed using new enable-sre-logging, callsre-prefix, callsre-suffix and logging-verbosity Vendor Specific Parameters (VSP) and are designed to allow IVR developers more control over the verbosity and types of logging generated within an individual session, for example if only certain applications or specific sessions should be logged in a certain way, these features now allow fine control over those attributes.
  • Added a new option to include selectable file attachments when sending diagnostic reports to LumenVox, or to disk. This adds more user convenience when network trace logs, Response Files, or any other files need to be sent along with diagnostic reports.

Changes:

  • Changed the mechanism used during diagnostic testing to allow for situations where ASR or TTS servers are in the process of starting (following a restart for example) but are not quite fully operational when the diagnostic tests are started. These diagnostic tests will now wait for up to 2 minutes for these services to respond in this unusual situation.
  • Changed the internal processing of processing URIs (grammars, ssml, etc.) when HTTP redirects (3xx Redirection Response Codes) are encountered. Previously these were returning the redirect message itself, instead of following the redirect to obtain the ultimate file. This was found during internal testing, so may not have been noticed by customers in the field, however it may have impacted redirected fetch operations in earlier versions.
  • Changed the mechanism that releases licenses upon completion of MRCP sessions. Licenses are now released prior to responding to SIP BYE or RTSP TEARDOWN requests. Previously, this processing was handled on a background thread, which could have resulted in a few milliseconds delay between the BYE/TEARDOWN response being sent and the actual license release.
  • Minor change to ASR startup to log configuration items PUBLIC_RULE_ACTIVATION_MODE and UNKNOWN_LANGUAGE_VALUE. These were previously not being recorded in the restart log. 
  • Changed the format of Media Server NLSML responses when processing multiple n-best alternates to only return those above the currently specified confidence threshold. Previously, whenever at least one alternate was above the threshold, all were being returned. This was not consistent with the specification, so this change adheres more closely to the MRCP specification.
  • Minor ASR startup change that now logs out completion time of PreFlightCheck after clearing cached grammars has been completed. Previously customers with many thousands of cached grammars could see significant delays while the cache folders were being cleared during startup immediately following a version update. This additional logging adds more verbosity to this processing.
  • Modified the Speech Tuner to retain the user-flags assigned to any interactions when using "Clear Results". Previously, these flags were being cleared, along with decode and other information. Allowing these user-flags to persist can be beneficial when working with the Speech Tuner.
  • Minor improvements to the diagnostic test being run, which now check for valid NUM_DECODE_THREADS and MACHINE_IP values.
  • Minor change to use updated pnotify.js 3.0.0 library within Dashboard.
  • Reworked the Media Server mechanism used for streaming TTS audio due to a thread locking issue that led to delayed or deferred streaming across different TTS streams. All Media Server TTS users are recommended to upgrade to this version in order to benefit from this change.
  • Improved diagnostics to be more aware when services are not monitored. Also added check and testing for the new German ASR language, when enabled.
  • Dashboard diagnostic report summary now includes network/DNS warnings in the total warnings count at the bottom.  Also increased severity for incorrect MRCP_SERVER_IP value from WARN to ERROR
  • Changed the handling of empty DTMF strings, when for example an input timeout or DTMF-term-char is input when there is no prior DTMF data to process. Previously this generated a benign but potentially confusing error message in the logs, which is now suppressed.

Fixes:

  • Fixed a minor non-fatal Media Server exception which was being emitted when all logs were cleared from the Dashboard.
  • Minor correction of the description of the PUBLIC_RULE_ACTIVATION_MODE ASR configuration option in the Dashboard, which was incorrectly showing details relating to grammar cache enabling. This was a regression introduced in the 14.2.100 (April 2016) release.
  • Fixed a number of typos in the Mexican Spanish (es-MX) builtin grammars: The currency grammar, where VEINTIUN was being used in place of the correct word VEINTIÚN (with accented U), also replaced VEITE with the correctly spelled VEINTE in the es-MX time and date grammars. THOUSAND (English) was replaced with the correct MIL (Spanish) in the phone grammar and KILOGRAM (English) was replaced with the correct KILOGRAMO (Spanish) in the number grammar.
  • Fixed a Grammar Editor issue that was incorrectly setting the encoding value to ISO-8859-1 when a UTF-8 grammar was edited using the Speech Tuner. After making changes, saving as UTF-8, later returning to the same grammar would see this reverted to ISO-8859-1 format, which was incorrect (and annoying).
  • Fixed minor warnings about a voice/language mismatch that were being emitted when using the Welsh Gavin or Gwendolyn TTS voices. This may have been most noticeable to Speech Tuner users, or API users accessing the LV_TTS_GetSynthesisWarning... series of functions. This issue related to the unique mixed availability of en-GB and cy-GB for these voices.
  • Modified the Media Server tag; element in the To: header of UDP SIP responses when responding to RE-INVITE requests. Previously, when processing parallel requests, or INVITE requests prior to processing ACK, the tag; was only being sent in response to the first INVITE request, when it was intended to be present in all such responses. This was a minor deviation from the SIP specification, and was unlikely to have affected many users due to the very specific nature of the conditions.
  • Dashboard diagnostics now correctly identifies the License Server associated with a Media Server when performing diagnostic tests if there is no local License Server installed.
  • Dashboard diagnostics now suppresses warning messages about missing configuration files in configurations where such files are not installed, or not monitored.


15.0.400 (October 20, 2016):


Changes:

  • Modified the Media Server behavior when a RECOGNIZE request is received within a few milliseconds of a DEFINE-GRAMMAR that is referenced by the RECOGNIZE request. This release specifically addresses an issue discovered during some Cisco interop testing.
  • Modified the Manager code to no longer log failed connectivity with unmonitored services when processing KPI data. These messages were unnecessary, since connectivity was not expected in those cases.


15.0.300 (October 17, 2016):


Improvements and New Features:

  • Adding support for 32-bit CentOS / RHEL 7. This change now compliments the existing 64-bit CentOS / RHEL 7 support that has been in place since LumenVox 13.1.100 (July 2015), and also coincides with the removal of support for the Linux CentOS / RHEL 5 Operating Systems (both 32 and 64-bit), which are reaching their end of life. All 32-bit Linux users are urged to migrate to CentOS / RHEL 7 32-bit as soon as is possible to avoid any support related issues. Customers may continue using the 14.2.100 version of LumenVox while it remains within its support window if they prefer to continue using CentOS 5 for some reason. 
  • Updated the grammar compilation module within the ASR. Although this is a significant change from a technical implementation perspective, it is unlikely to affect users. Those small numbers of users that may be affected would have previously been using suboptimal grammars, which should be corrected by this change. We recommend users validate their existing grammars against this new version before deploying them into a production environment. If any users encounter issues with this, please work with LumenVox Client Services who may be able to assist. This change may also have the effect of more optimized grammars and fractionally faster compilation times and accuracy, depending on the specific grammar, but may also introduce a very slight negative effect on confidence scoring values. It is likely that most users will not notice any change as a result of this new grammar compilation engine.
  • Added a new SKIP_DETERMINIZE meta property flag option that may be included within ABNF or GrXML grammars to direct the grammar compiler to skip the FST determinize step while compiling. This may be necessary for the very small number of users that might be affected by the above change. Note that running in this mode is suboptimal and likely to use more CPU and memory resources while performing decodes. It is anticipated that most users will not need this.
  • The implementation of OpenSSL in this version of LumenVox has been updated from 1.0.1m to 1.0.1t, which addresses a number of recently detected vulnerabilities. This change is also integral to several other related changes in this version. Note that Linux versions utilize the version of openssl installed on the system, so users are encouraged to update these regularly.
  • Added new SSL_PROTOCOL and CIPHER_LIST options to the manager service, allowing users much finer control over which secure connection protocols and ciphers are used when connecting to the dashboard web interface. New configuration settings for these values have been added to the dashboard manager configuration interface.
  • As part of the range of security enhancements added in this version, we have removed support for SSLv2 and SSLv3 protocols when connecting to the manager. Now the default protocol type will be TLSv1.2, however users may select TLSv1.1 or TLSv1.0 if they choose. Note that TLSv1.0 is no longer deemed secure, so should only be used if there is no other choice. These changes address a number of recently detected vulnerabilities within the openssl code, including the POODLE (CVE-2014-3566) vulnerability.
  • New FLEX_DESIRED_SYNC_HOUR and FLEX_DESIRED_SYNC_MINUTE options added to the License Server (accessible via Dashboard configuration screen too). These options allow users to specify a desired time of the day when flexible license synchronization events occur. By default, these are disabled, leaving the old functionality in place (at 24-hour intervals), however if users wish to impose some control of when these regular synchronization events occur, these values may be useful.
  • Added a new enhanced DNS test as part of the Dashboard diagnostics testing, which includes a new DNS_WARN_THRESHOLD configuration option, used to check the time taken for DNS resolution. Responses slower than this will be shown as warnings in the diagnostic reports. The default setting for this value is 500 milliseconds, but may be adjusted as needed. This is in response to certain customer installations where unwanted delays in DNS resolution have caused grammar fetching issues among other problems. This new feature should highlight any inherent problems that users may unknowingly have within their network configuration.
  • Added new security certificates as well as a new set of LumenVox Licensing Nodes dedicated to HTTPS communication when performing flexible license synchronization. These are now used in conjunction with new License Server changes that implement HTTPS connectivity in a completely different way than was previously performed to assist users that encountered issues when enabling this secure version of the protocol via the new FLEX_REPORTING_PORT_HTTPS setting.
  • Adding a new GRAMMAR_CACHE_FILE_MAX_AGE setting for the manager, which may be used during cleanup to remove cached grammar files after being unused for a specified number of days. This functionality is disabled by default. Once enabled, any files older than the selected number of days will be cleaned from the corresponding Client-Side and ASR cache folders - this may be particularly useful for users that implement a number of dynamic grammars, which can create a lot of unwanted and unused cache files on the local file system.
  • Adding a new RESPONSE_FILE_MAX_AGE setting for the manager, which may be used during cleanup to remove old response files if desired. This functionality is disabled by default. Once enabled, any files older than the selected number of days will be cleaned from the corresponding response folders. Previously, these files and folders have not been monitored, and maintenance was therefore the responsibility of the system administrator. This change allows the manager to remove these files after a period of time, if desired.
  • Added new Call Detail Records (CDR) logging to the Media Server. This will now automatically record various metrics associated with MRCP sessions processed by the Media Server, including the Session ID, start and end times, duration as well as which network ports, licenses and any Logging-Tag values were used. The implementation also uses our common framework, which allows the number of backup copies and the size of log files to be configured using the logs.conf settings file.
  • Added new Administrative Out-Of-Service functionality to the Media Server (also configurable from the dashboard configuration screen). This allows administrators to selectively take LumenVox Media Servers out of service to perform maintenance as needed. When enabled, any new session requests made of the Media Server will be rejected with a configurable response code and any existing sessions will be permitted to continue until they naturally conclude. Administrators can monitor this number of calls and when the count reaches zero, can perform maintenance as needed. Starting up the Media Server with this functionality enabled has the reverse effect. Enabling and disabling this option allows administrators to control where new sessions are to be handled when there are multiple LumenVox stacks configured in an environment.
  • Extended the recently introduced functionality that allows users to clear their LumenVox logs from the Dashboard diagnostic screen. Now whenever the Clear All Logs option is selected, a set of current configuration values will be logged to the corresponding restart logs, allowing the current setup to be more clearly described when debugging.
  • Modified the License Server to use new FLEX_NODE_LIST_HTTPS and FLEX_REPORTING_PORT_HTTPS options whenever USE_FLEX_REPORTING_HTTPS is enabled. This separates the HTTP and HTTPS settings, making configuration simpler for users. This also introduces a new set of flexible licensing nodes as described in the FLEX_NODE_LIST_HTTPS, which have new security certificates and operate independently from the existing flexible licensing nodes currently shared between HTTP and HTTPS traffic. Moving forward, all users should begin using these new flexible license nodes (flexlicense-s01.lumenvox.com thru flexlicense-s10.lumenvox.com) when using HTTPS. The existing flexible licensing nodes will continue serving HTTPS for a transitional period, and in the long term will continue servicing HTTP requests, but HTTPS users are encouraged to migrate to these new more secure machines. The Dashboard now takes the FLEX_NODE_LIST and FLEX_NODE_LIST_HTTPS as well as the appropriate FLEX_REPORTING_PORT or FLEX_REPORTING_PORT_HTTPS into consideration when performing diagnostic tests, based on the current USE_FLEX_REPORTING_HTTPS setting.
  • Added a new Certificate Authority (CA) bundle to be used when working with HTTPS for things like fetching resources (such as grammars) over secure HTTPS connections. The new flexible licensing nodes are compatible with this new CA file, which should allow connectivity to be more compatible for users.
  • Added a new manager option of tracking Key Performance Indicators (KPI) for LumenVox products. This allows users to periodically log out various metrics from LumenVox ASR, TTS and Media Server Services. These values include number of operations, maximum and average time taken for certain operations as well as high-level tracking of the number of errors encountered within an interval. These reports are generated as .csv format, so can be imported to database or spreadsheet applications as needed. The logging period can be configured to anywhere between one minute and one day (at one minute intervals). By default, this new feature is disabled to remain consistent with previous versions.
  • Added new boolean, currency, date, garbage, number, phone and time builtin grammars for the British English (en-GB), Australian English (en-AU), Indian English (en-IN), Colombian Spanish (es-CO), Canadian French (fr-CA) and Brazilian Portuguese (pt-BR) ASR languages. These may now be referenced using the common builtin:grammar/... syntax, assuming appropriate ASR licenses and languages are installed.
  • Added Simple Network Management Protocol (SNMP) handling to the Manager Service, which can now respond to SNMP requests. Protocol version 1 and v2c are supported for SNMP requests and all Trap notifications utilize SNMP v2c protocol, using a configurable COMMUNITY_STRING value.
  • Added SNMP trap notifications to the Manager, based on communication failures, Media Server license errors as well as configurable multi-tiered CPU, Memory and Disk Space alerts. These trap notifications include a number of Variable Binding fields with detailed information about the alerts as well as a way of correlating CLEAR messages with their associated SET messages, allowing System Administrators to easily integrate LumenVox servers into their existing network management systems.
  • New SNMP communication link checking now verifies communications with all ASR, TTS, License, Media Server and Call Indexer services configured on the local machine, as well as any defined remote servers. Any failure or restore of these services, or their communication links will be reported as trap notifications, along with an indication of whether it is a Primary or Backup server. If there is no alternate server available, the verbosity of the notification will be CRIT (Critical Alert), and if an alternate server is available, the severity will be reported as MINOR (Minor Alert).
  • Added a new configurable SNMP heartbeat notification, which can be used by system managers to detect the failure of a LumenVox server.

Changes:

  • Significant change to how the Speech Tuner processes results with accented characters. Previous case insensitive comparison would not automatically see accented characters of different cases as being equivalent, so would incorrectly report poorer performance that the application was getting.
  • Internally changed the gcc compiler version from 4.4 to 4.7 when building our Linux versions. This should not affect most users, but may be of interest to certain API developers. This change, as well as providing optimized performance, will allow a number of future enhancements. This change only affects our Linux versions of the product.
  • Minor change to remove unnecessary "v10.4" User Agent string from packets when fetching grammars etc. using HTTP/HTTPS
  • Minor change to clarify the description and intent of the VAD_BARGEIN_THRESHOLD value in client_property.conf
  • Minor change to add logging of FLEX_NODE_LIST to License Server restart log when the service is started and when logs are cleared using the diagnostic screen in the Dashboard.
  • The Manager configuration (including the Dashboard configuration page) now includes a new MACHINE_IP setting. Assigning this to the machines local IP address will allow the manager to correctly identify which services are running locally, and which are remote. This provides better visual feedback in the Dashboard Summary page as well as when performing diagnostic tests, which can be very useful in situations where distributed LumenVox servers co-exist within a local area network.
  • As part of the other security related changes in this release, the default self-signed certificate (SSL_CERT_FILE) that accompanies the manager and is used when serving HTTPS requests has been updated to use a significantly more secure SHA256 2048 algorithm over the previous SHA1/1024-bit version that was used before. Options to replace this self-signed certificate with a valid signed certificate remains the same as previously documented, although using updated, more secure methods is now encouraged.
  • Minor change to LVSpeechPort_Streaming_Example on Windows to avoid using the same intermediate folder for both 32 and 64-bit builds, which now allows both to be more cleanly built.
  • Small modifications to the Linux RPM packages to move the lv_manager daemon from the CORE to the CLIENT package.
  • Added 'zip' dependency to Linux CORE packages, enabling diagnostic report compression when sending reports to LumenVox, which also addresses issues where some customers were not able to upload logs using the dashboard due to this missing third-party zip functionality. Most users likely have this package installed already.
  • In addition to several other changes to the manager, related to the new MACHINE_IP setting, the manager now actively monitors the client_property.conf LICENSE_SERVERS and TTS_SERVERS values as well as the media_server.conf SRE_IP values for changes. Any detected will force a reset of those connections at the manager to keep it in sync with current values. The result of this should be less user confusion when changing these settings without subsequently resetting the manager.
  • Changed the flexible licensing system metrics tracking when switching from primary to backup license servers, where some inconsistent overlap was being reported where the servers were not connected on the same subnet. This issue is unlikely to affect many customers, but has now been resolved.
  • Changed the API function LV_SRE_GetAvailableLicensesCount to now support flexible licensing users. Now users may utilize this function, passing one of the new flexible licensing strings, such as "ASR_PORT" or "ASR_TIER2" to obtain the currently available license count. Previously, this API was only designed to work with the legacy licensing mechanism.
  • Changed the Dashboard diagnostic testing to no longer check connectivity with the now-deprecated LumenVox Subscription License Servers. All users of the Legacy Subscription License Servers should have already migrated to the new Flexible Licensing System.
  • Changed the way in which rules-based pronunciations are implemented within the ASR grammar compilation process. Previously these would be used as alternates whenever a word or phrase was not in our primary lexicon. Testing has shown that the newer Statistical Pronunciation enhancements that were introduced in 13.0.100 (January 2015) outperform the rules-based pronunciations, so now rules-based pronunciations will only be used whenever no alternates exist. This should lead to better overall accuracy as well as reduced ambiguities in ASR decode searches, so giving better throughput too. This change only affects rare, out-of-lexicon phrases, and is likely to have a more noticeable positive impact on non-American English languages.
  • Changed the Phonetic Speller tool in the Speech Tuner to coincide with recent rules-based and grammar compilation changes. This mostly affects multi-word and hyphenated phrases as well as words not in our built-in ASR language lexicons.

Fixes:

  • Fixed problem related to invalid processing of Random Sentence Generator functionality within the Speech Tuner affecting certain specific types of grammars. This issue did not affect grammar processing, and was isolated to the Speech Tuner tool.
  • Fixed some minor issues with the builtin:grammar/currency for the es-MX language. Specifically, the older version did not correctly process "ocho dólares y cero centavos".
  • Fixed a minor problem when shutting down speech port client applications on CentOS 7, which could cause occasional exceptions to be logged while shutting down due to a synchronization issue. This has now been addressed, and may result in these types of application shutting down slightly faster.
  • Fixed an issue where the manager would not clean resource files if LOG_FILE_MAX_AGE was set to 0 (disabled). The intent was that this cleanup operation should have been independent of this configuration, so now this is the case. These resource files are typically fairly small, so it is unlikely to have caused any significant impact on user systems; however the new cleanup operation is preferred. Users with older version of LumenVox can simply specify a non-zero value for the LOG_FILE_MAX_AGE setting to enable resource file cleanup until they are able to update to this version.
  • Fixed an issue calculating installed memory in 32-bit Operating systems if the value was above 4GB. This has minimal impact on performance, only affecting diagnostic logging, which under-reported the amount of available memory.
  • Fixed an issue in SimpleTTSClient when processing SSML requests larger than approximately 4 KB, which were causing a buffer overrun issue in the test code due to a fixed length buffer. This only affected the sample code, and not any core functionality.


14.2.100 (Apr 11, 2016):

 

Improvements and New Features:

  • Added new TTS1 Polish voices Jacek (Male), Agnieszka (Female), Ewa (Female) and US English voice Justin (Male Child).
  • Support for these new TTS voices was added to the Speech Tuner auto-complete functionality within the TTS Editor.
  • Added new client_property.conf settings options to the Media Server configuration page in the Dashboard. This now allows users to modify all Media Server and underlying Client Property configuration settings from within the Dashboard. These new settings are conveniently shown below the existing Media Server options, in the "Client Property Configuration" section on the page.
  • The configuration settings view within the Dashboard now also has additional setting information available as expandable dropdowns beside certain settings where needed. This provides users more verbose information about settings, offering further assistance when making configuration changes.
  • Further improved the Dashboard configuration page by compressing the underlying code, making it significantly smaller and therefor faster when using this page.
  • Changed the Dashboard to allow the Manager Service to be restarted if necessary. This restarting option is now available one the Summary and Configuration Pages.
  • Added a new option to display the last successful License Synchronization date when viewing the Licensing Page within the Dashboard. This option is only visible when using the Flexible Licensing mechanism, and may be helpful when determining whether successful synchronization with the LumenVox Licensing Nodes in the cloud is occurring.
  • Added a new 'Class' column to the Interaction List displayed within the Speech Tuner. This contains the Accuracy Classification of each interaction, based on the currently selected Confidence Threshold. The currently selected Confidence Threshold can be changed by clicking on one of the Confidence Histograms. Once a Confidence Threshold is selected, these classifications show how each interaction would be classified if the selected Threshold was used, in terms of In Grammar (IG), Out Of Grammar (OOG) as well as Correct-Accept (CA), False-Accept (FA) and so on, from an accuracy classification perspective. This information will now also be exported if users select the "Save List To File..." option for the Interaction List, and may therefore be helpful when compiling external accuracy reports.
  • Added new builtin:grammar/boolean support for the Brazilian Portuguese ASR language

Changes:

  • The Speech Tuner now allows users to edit the settings for their selected License Server within the application configuration dialog instead of needing to edit the corresponding client_property.conf file on disk as was previously needed. Any changes made within the Speech Tuner dialog are permanently saved to the corresponding LICENSE_SERVERS setting of the conf file on disk.
  • Modified the Speech Tuner export audio functionality to add a new option of utilizing any Logging-Tag entries derived from imported Response Files when naming the saved audio files to disk. If selected, the corresponding Logging-Tag (if present) will be used as the filename prefix where appropriate. If no appropriate Logging-Tag is detected in this mode, the specified default prefix string will instead be used.
  • Modified some Error Code messages to return ASR in place of the previously used SRE references as part of our ongoing branding updates to standardize on the use of the more familiar ASR term.
  • Changed the wording of the MRCP diagnostic report results on the Diagnostic Page of the Dashboard. These changes correct minor typos and also clarify the meaning of certain test failure conditions.
  • Modified the License Server's Configuration Page within the Dashboard to expose the FLEX_NODE_LIST configuration option that was previously not present.

Fixes:

  • Fixed incorrect behavior of special NULL rule handling when used in certain optional parameter conditions due to incorrect weighting being applied. This is unlikely to have affected many customers and was only discovered during internal aggressive testing.
  • Fixed some minor typos in the Dashboard's Diagnostic page within the diagnostic report.
  • Fixed a bug in the Dashboard's Diagnostic testing algorithm that prevented the correct TTS language from being selected when running certain tests. Previously a default value of en-US language was selected when this may not have been appropriate and caused some confusion when no American English voices/licenses were available.
  • Fixed an unusual bug related to Flexible Licensing Tier selection, which appears to have affected a single customer, where transition from Tier2 to Tier3 licenses were prevented in very specific conditions based on the grammar vocabulary size.
  • Fix for SSML processing of say-as / interpret-as "telephone", "phone", "vxml:phone" and "vxml:time" elements, where carriage-return and tab characters within the element were being unexpectedly read aloud.


14.1.300 (Feb 16, 2016):

 

Improvements and New Features:

  • Updated the LumenVox End User License Agreement (EULA) documentation accompanying all products to the latest version.

Changes:

  • Changed the Speech Tuner Matching Files Update Dialog to be less sensitive. Now the list of possible grammars presented excludes any grammars that do not match the prior text of the grammar being modified.
  • Removed a minor security loophole within the Dashboard that could allow system monitoring data to be exposed to non-authenticated users via the Dashboard's Monitoring tab.
  • Small cosmetic change to the Licensing page of the Dashboard to correctly show the license tree expansion after certain operations such as "Resync Info"

Fixes:

  • Fixed a minor bug in the Speech Tuner, introduced as part of the 14.1.100 changes relating to the modification of a root grammar within a callsre file, where the root status was incorrectly removed. Speech Tuner users should upgrade to this version from 14.1.100 to take advantage of this change and avoid the issue. This issue only occurs when editing root grammars in the version 14.1.100 Speech Tuner.
  • Fixed a Media Server issue where Content-ID is specified with angle brackets during DEFINE-GRAMMAR (i.e. 'session: <grammar111@grammar.store>' and later used during a RECOGNIZE request without the brackets (i.e. 'session: grammar111@grammar.store'). This appears to have affected a small number of users, so modifications have been made accept this formatting.
  • Fixed a small typo with the recently introduced Diagnostics page of the Dashboard when an MRCPv2 synthesis failure is reported. Previously, the RTSP port was being shown in the error string instead of the SIP port that was actually used.
  • Fixed a reported issue relating to occasional Media Server exceptions occurring while streaming TTS audio. This appears to been due to an unusual buffering issue that this change addresses.


14.1.100 (Jan 11, 2016):

Improvements and New Features:

  • A new diagnostics tab has been added to the Dashboard utility. This can be used to check the system configuration and test functionality. Options are also available to globally change logging verbosity to debug or default levels as needed, or clear all LumenVox logs. An option to send a diagnostic report to LumenVox, or save it to disk is also included.
  • Dashboard now includes options to disable monitoring certain LumenVox services on the local machine, as needed. This allows users to prevent indicating issues with services they do not intend to use. One example of this is that the seldom-used Call Indexer Service is now unmonitored by default. These new options coincide with the new diagnostic tests that are now available.
  • Dashboard configuration screen now includes Basic, Intermediate and Advanced filtering options, allowing new or occasional users better clarity as to which settings should be altered. Now that there are a lot of possible options for some of the LumenVox services, this should make it easier for users to manage.
  • The Summary page on the Dashboard now optionally shows the IP addresses of remote LumenVox services that may be configured.
  • Many improvements added to the lv_show_config utility, including an option that displays the current values as well as the default values for all configuration settings reported. Previously only the current setting was shown, which did not clearly indicate which settings were at their defaults.
  • Additional improvements to the lv_show_config utility were added to better support mixed language installations. Now users can specify the ASR and TTS languages they wish to test (asr_lang and tts_lang options respectively), instead of the utility defaulting to American English.
  • Along with other significant changes to the lv_show_config utility, results being returned were updated to be more verbose and clearer to users whenever an issue is detected. For example, previously where a generic failure message would be shown, the output will now be clearer, indicating a missing license or language pack for example.
  • The lv_show_config utility was also updated to automatically skip over SIP or RTSP testing if those ports are marked as disabled (set to 0 in media_server.conf). This change removes user confusion while running this utility.
  • Added new builtin:grammar/digits grammars for Australian English, British English, Indian English, Colombian Spanish, Canadian French and Brazilian Portuguese. These optional grammars were previously not provided.
  • Added a new process status checking WebAPI to the manager server allowing users to check the status of LumenVox processes by calling /apt/status/asr etc. and receiving a corresponding HTML page response. This may be of particular interest to those users that have requested something that they could use to integrate to their monitoring systems, such as SNMP or other management tools, as used by many datacenters.
  • Added a new LOGGING_VERBOSITY configuration option to manager.conf, allowing the manager service to adjust logging verbosity, in a similar way to our other services. This, in conjunction with some logging cleanup results in easier to read manager_app.txt files.

Changes:

  • The Manager Service will now automatically skip attempting to clean up log files that are marked as read-only. This only affects logging reports, since previous versions would report attempts to clean up such files.
  • Configuration files for License Server, TTS Server and ASR Server are now correctly located and reported by lv_show_config. Previously on Windows-only installations this was omitted since these files could be installed anywhere on the system as determined by the user. Linux machines are unaffected - the configuration files are always located in a fixed location.
  • Manager now reports if unable to control services due to permission issues. Previously, such controls were disabled with no indication as to the reason.
  • The "Phonetic Speller" and "Random Sentence Generator" dialogs in the Speech Tuner now automatically scale correctly if Windows users configure non-default screen resolution font settings.
  • The Media Server now processes the Speech-Language header for TTS SPEAK requests, which now allows text/plain requests to be able to specify and use the user's desired language. This is only needed if not using SSML markup - if SSML markup is used (the most common usage), this header value will be ignored. Previously when using text/plain TTS requests, these would only use the default (SYNTHESIS_LANGUAGE) specified in client_property.conf
  • Changed the Recognition-Timeout characteristics within the Media Server. Previously whenever this timer expired, the recognition was terminated with a Recognition-Complete and a completion cause of Recognition-Timeout. We found that it is possible to improve performance by attempting a decode at this time if sufficient audio has been received and a valid result is achievable. If insufficient audio is available, or the decode yields no result, the response will be the same as before, however if a valid response is available, it will be returned instead. This may dramatically help users encountering barge-out issues related to certain noisy conditions.
  • Implemented a minor change to unify our third-party library notifications. These are now all contained in a single ThirdPartyLegalNotices.txt file distributed will all of our packages.
  • Added option to configure LICENSE_SERVERS setting (within client_property.conf) for Media Server in the Dashboard.

Fixes:

  • Fixed a number of minor bugs relating to MRCP connectivity problems when running the SimpleMRCPClient utility.
  • Removed a bug in the manager configuration settings related to Access Control List (ACL), which prevented the use of the + and - modifiers. This is now corrected, allowing the correct ACL attributes to be specified per our documentation.
  • Fixed a significant issue in the speech tuner relating to incorrectly resolved grammar references when a particular combination of parent-child-grandchild grammars are used and multiple slightly different instances of these relatives exist, with or without references to custom lexicons. This problem only affected a very small number of Speech Tuner users, but the impact was significant, where grammar references were not correct. The internal structure and processing of these grammar references was completely redone to address this issue. Now if sibling grammars are edited within the Speech Tuner, users will be presented a dialog, indicating the affected references, and given a choice of whether to modify those references, or only the current one.
  • Fixed a client-side grammar caching issue, where duplicate entries could be spawned, causing unnecessary duplicate copies to be saved into response files. This issue is unlikely to have been noticed by users, however this behavior used more disk space and memory than was needed, and is therefore less optimal. We encourage users to upgrade to this version to make use of this change.
  • Fixed a minor manager startup issue, which may have resulted in incorrect log reports of problems accessing the ADMIN_PORT.
  • Fixed an issue in the Media Server when processing a BARGE-IN-DETECTED message from an RTSP client, when no active SPEAK request was being processed. Previously, the acknowledgement response to this notification did not have the required Content-Type header and may have caused some platforms to incorrectly process this response. This incorrect behavior has been present since version 10.0 (February 2011)


14.0.100 (Sept 25, 2015):

 

Improvements and New Features:

  • Added support for Brazilian Portuguese (pt-BR) ASR language.
  • Added a new mechanism to report warnings detected during grammar loading. These are issues that insufficient to trigger a grammar loading error, so may have previously gone unnoticed (despite appearing in the log files). This new mechanism introduces three new API functions exposing this functionality: LV_SRE_ReturnGrammarNumWarnings,LV_SRE_ReturnGrammarFirstWarningStringLength and LV_SRE_ReturnGrammarFirstWarningString.
  • Added support for new Brazilian Portuguese ASR language to the Speech Tuner as well as the new grammar warning mechanism. Now any syntax or similar warnings generated when loading grammars will be displayed to the Speech Tuner users.

Changes:

  • Modified the Phonetic Speller in the Speech Tuner to correctly process multi-word phrases. Previously these multi-word phrases were being processed, however the method used was not identical to that used by the ASR during grammar compilation, so may have yielded slightly different results.
  • Modified the way the Speech Tuner lists grammar references. Previously, the complete grammar URIs were displayed, which often obscured the filename portion of the reference. Now, users can choose between displaying the entire URIs or the filename portions. Both can always be displayed by viewing the properties of a selected grammar, but displaying only the shortened version may be easier for many users. This setting can be toggled as desired from a new "Show Grammar Paths" option in either the View menu, or the grammar list context menu.
  • Changed the Speech Tuner license checking behavior. Now if the license date has expired, a message to that effect will be displayed prior to exiting. Note that using the Speech Tuner requires an annual subscription license that must be renewed each year.
  • Minor enhancement to the Licensing Tab in the Dashboard to include Expand-All and Collapse-All options to allow users more convenient display options for the license tree.

Fixed:

  • Fixed a Media Server issue that caused an exception when processing SIP UPDATE requests from unknown or previously closed sessions.
  • Changed the Media Server to re-implement SIP responses when there is no message-body in the message (such as a response from SET-PARAMS). A change in 13.1.100 affected the MRCPv2 processing for Avaya (and possibly other platforms), preventing correct operation. This change now ensures compatible responses are returned.


13.1.100 (July 23, 2015):

 

Improvements and New Features:

  • Added support for Linux RHEL 7 / CentOS 7 Operating Systems, and deprecated support for Windows 2003.
  • Added sysvstatus option to Linux init.d scripts to be consistent across supported Linux platforms with the introduction of RHEL 7/CentOS 7 support. This small change was needed since the semantics of status changed due to the use of systemd.
  • Added new Portuguese (pt-PT) Female 'Catina' TTS voice and new Russian (ru-RU) Male 'Vasili' voice.
  • Migrated Windows development to Visual Studio 2013 from VS2008 in previous versions. This means that the runtime libraries we are built against are now using the VS2013 versions, and our sample projects for Windows have also been updated to reflect this change. Along with our deprecated support for Windows XP in 13.0 and now Windows 2003 in 13.1, we can use more optimal coding algorithms, some of which are being introduced in this release. Most customers should not notice any significant difference in this migration.
  • Added support for seldom used 'OPTIONS' request (for ASR and TTS) in RTSP protocol within the Media Server.
  • Rebuilt TTS1 acoustic models, introducing a number of overall enhancements, improved currency interpretation and pronunciation of numerous words, including proper names and expansion of abbreviations as well as road numbers. Improved apostrophe handling and unit expansion and time/date ranges. Improved American address expression, state names and contractions.
  • Added TTS1 processing for radio stations and non-native characters (such as ê). Improved pronunciation of hyphenated compounds (e.g. pay-as-you-go) and improved homograph disambiguation. Improved reading of serial numbers, chemical elements and NCAA team names among other pronunciation improvements.
  • LVShowConfig -tts_test will now succeed even if the default TTS voice is not installed or licensed. As long as at least one voice can be used, the test will complete successfully. This addresses unusual issues where users had not updated the default voice setting, typically when using a non en-US voice or language.
  • Rebuilt the ASR acoustic models to provide a slight improvement in overall accuracy.
  • Modified the internal architecture of the Media Server to run more efficiently using less threads, which should allow it to scale to even higher port densities than before. These changes also include a number of memory management optimizations to reduce memory fragmentation of extremely heavily loaded installations.
  • Modified the Windows installation to set the Media Server service to start automatically whenever the system reboots. Previously this was assigned as a manual start, since not all users utilize this service, however we have changed this setting to be more accommodating to the MRCP / Media Server users. Any users that do not utilize the Media Server can reduce the overhead of having that service running by disabling it in the Windows Services Manager. See out knowledge base article on Windows Installation for more details on how to do this. Note that on Linux systems, this automatic startup option was already in place for the Media Server.

Changes:

  • Changed the response to requests in RTSP protocol within the Media Server to return a simpler "404 Not Found" instead of the previous "486 Busy Here" when processing DESCRIBE or SETUP requests. This new response is more optimal and more consistent with the MRCP standard. If any user wrote code to expect the "486 Busy Here", we recommend changing to anticipate either, or use the normalized "404 Not Found" now being returned. In the interim, users can add and enable the hidden 'send_486_replies' flag in the MRCPv1 section of media_server.conf to continue receiving the "486 Busy Here" responses (this is for backward compatibility only)
  • Changed the behavior of the Media Server when RECOGNITION-START-TIMERS requests are received whenever the recognizer is not in a RECOGNIZING state. Such requests will now receive a "402 Method not valid in this state" error response, where previously the request could have been accepted and the timers started. This new functionality is more consistent with the MRCP specifications. Other timers also now check the current recognizer state prior to being processed (such as DTMF activity), which is a more robust overall implementation.
  • The ECMAScript engine used for processing Semantic Interpretation data was updated to use the latest Spider Monkey version 31 library. This was also modified to reduce dependence on system installed libraries which have caused occasional problems in the past.
  • Various optimizations and improvements were made to SimpleMRCPClientto minimize stalling or errant packet processing.
  • Added 'Content-Type' for legacy RTSP 486 replies from Media Server. This header was previously missing.
  • Changed MultiThreadedStreamingExample (C and C++) examples to print out the parameters used to the console, giving users a little more detail about which audio and grammar files and number of threads are being used.
  • Removed the RTSP idle timeout by default from the Media Server. This was previously added to address issue with certain client applications opening many ports and then abandoning them, which would use up unnecessary resources. This new default, although a change in behavior, is unlikely to affect any systems. If someone wishes to re-enable this non-standard functionality, setting the enable_rtsp_idle_timeout to 1 in the media_server.conf settings will enable the old behavior. This option is retained for backwards compatibility only, and its use should be avoided due to the (small) additional overhead it introduces. When enabled, this timer will trigger an RTSP connection shutdown if no requests are received within 10 seconds of the connection being established.
  • Add support for file://C:/blah URIs (with two slashes following the file:). These are now being treated as if they were file:///C:/blah URIs (three slashes following the file:). The URI with two slashes is technically invalid, but to remain consistent with other vendors which support these non-standard URI formats, we have added this support. Please see our URIs and LumenVox knowledge base article for more information relating to supported URI formats, or https://en.wikipedia.org/wiki/File_URI_scheme for more explicit details and examples of valid file URI schemes.

Fixed:

  • Fixed a problem with TTS synthesis where an ambiguous language 'en' specification was failing to utilize the correct language per our published Voice Language Selection Priority list. Now any partially specified language identifiers will follow the correct order of precedence.
  • Fixed the TTS language/voice ordering for Gwendolyn. As a female, she should precede any male voice in that language. Previously this voice preceded Gavin, which was inconsistent with our order of precedence rules, so this has been changed. Again, these rules only apply if an ambiguous language, gender or voice-name selection is made, so this change is unlikely to impact many users.
  • Fixed the TTS language/voice ordering for European Portuguese to have precedence over Brazilian Portuguese.
  • Fixed an unusual problem with the ASR that was throwing an exception when a certain custom lexicon was being used. The problem was isolated to the 13.0 version of LumenVox and also specific to RHEL6 Linux.
  • Fixed a Speech Tuner issue where Call data was being incorrectly shown on the Call Volume graph in the Summary view. The number of Calls shown was significantly lower that the correct values. This issue was a regression issue with the version 13.0 release.
  • Fixed an issue with SSML <say-as> telephone number processing for TTS1 voices. Previously the cadence of the numbers spoken was not correct. Now numbers are correctly grouped as expected and any following extension number or abbreviation is correctly pronounced.
  • Fixed a TTS stalling issue where a .wav file with a "LIST" chunk is specified as the reference. Additional (unsupported) chunk types are now ignored as needed to prevent such stalls in future.
  • Fixed a logging bug in LVShowConfig when logging an MRCPv2 failure. It would incorrectly report the RTSP port instead of the SIP port in the failure log which could lead to some user confusion.


13.0.400 (May 22, 2015):

 

Improvements and New Features:

This is a maintenance release which addresses the issues listed below, and should be considered a recommended update for users who may encounter these specific problems. No other functionality was changed, so there is no need to update to this version if you are unaffected by these changes.

  • Correction for a regression introduced in 13.0.100 for TTS playback artifacts affecting second and subsequent SPEAK requests within active sessions. Due to a recent optimization, utilizing an audio buffer, up to 2 seconds of audio artifact from an earlier canceled synthesis stream could be played prior to the subsequent (expected) stream being played. This affects MRCPv1 and MRCPv2 sessions where TTS playback was canceled due a STOP request or a BARGE-IN-OCCURRED request.
  • A correction to grammar URI parsing relating specified MRCP grammar URIs that contain more than one parameter in the URI string was made. Previously only parameters separated using ampersands was being parsed correctly, and parameters separated using semicolons were being pruned and ignored. The correct behavior of allowing either parameter separator is now implemented. This issue was isolated to the Media Server MRCP processing, and did not affect C/C++ API users.
  • A correction was made to ASR lexicon URI processing, which was a regression introduced with version 13.0.100. Specifically, lexicon URIs that were pointing to files on the local file system and had optional parameters, e.g. type=backup, were not being parsed correctly, causing this optional parameter to be misinterpreted and cause a grammar load failure to be emitted. This affected both absolute URIs using the file:// prefix and relative URIs from grammars loaded from the local file system. Web-based URIs, e.g. the ones using the http:// prefix, were not affected. These issues have now been resolved.
  • Corrected a Speech Tuner licensing issue when using the new Flexible licensing mechanism. Specifically, if the License Server port number was present in the configuration IP address string, this was causing connectivity issues, which prevented the Speech Tuner from starting correctly.


13.0.300 (April 9, 2015):

 

Improvements and New Features:

This is a maintenance release which addresses the two issues listed below, and should be considered a recommended update for users who may encounter these specific Media Server connectivity problems. No other functionality was changed, so there is no need to update to this version if you are unaffected by these changes.

  • Changed the ordering of media lines (m=xxx) in SDP responses to SIP INVITE packets to match the order represented in the original requests. Previously the LumenVox Media Server responded with the application and audio media lines in a fixed order, which caused issues on a specific IVR platform, so this change means the Media Server will now respond with media lines listed in the same order as they appear in the original request. This change makes the LumenVox Media Server more compliant with the IETF RFC3388 specification describing Grouping of Media Lines.
  • Added support for MRCP sequence number 0, which was previous being treated as an erroneous / invalid value. This change allows platforms to make requests using Request-ID 0 in both MrcpV1 and MrcpV2, and resolves a related problem seen with a specific IVR platform during initial integration testing. This addresses an ambiguous definition in the IETF MRCPv2 draft, and now offer support for the “0” MRCP request-id value for both ASR and TTS requests.


13.0.100 (January 6, 2015):

 

Improvements and New Features:

  • Added support for Windows 8 and Windows Server 2012 Operating Systems, and deprecated support for Windows XP
  • Added new Norwegian(nb-NO) Female 'Mathilde' TTS voice. Also included support for this new voice in the Speech Tuner
  • New SSML processing warning reports are implemented. Previously whenever SSML issues were encountered during TTS synthesis, if some output was possible, the synthesis result would simply be returned with no indication of possible syntax errors in the request. Now additional diagnostic warnings can be used to determine whether a referenced file was not found, or some minor syntax error was skipped over. Increasing the logging verbosity level at the client and server will show an increasing amount of detail when processing these warnings.
  • Added new C API functions to access the new SSML warnings that are now available following a TTS synthesis:
  • Added new C++ API functions to access the new SSML warnings that are now available following a TTS synthesis:
  • Added a new warning tab to the Speech Tuner SSML Editor view to indicate the presence of any SSML warnings that were detected during synthesis.
  • Added new acoustic models complete with advanced pronunciation generation algorithms for all supported ASR languages. Users of grammars that contain words already in our comprehensive lexicons will not really notice any difference, but grammars containing out-of-lexicon words and phrases should benefit from these changes. Such grammars and applications may include people and place names, foreign or unusual words.
  • A new and improved Canadian French acoustic model was produced that should perform slightly better than previous versions.
  • As part of our pronunciation enhancements in this version, we have extended our inline phonetic support and also clarified and improved our ASR custom lexicon support.
  • In addition to our new pronunciation enhancements, this version also includes a significant improvement to confidence scoring performance for all supported ASR languages. Significantly lower confidence scores will be seen for Out-of-Grammar and Incorrect utterances. There is also better separation between the correct and incorrectly recognized utterances making it easier for the application developer to pick good confidence thresholds or ranges for confirmation prompts. Our internal testing shows an average relative improvement of over 10% in application performance.
  • Added new command line argument -ip to SimpleTTSClient, allowing a specific TTS server's IP address to be specified
  • Improved how the Speech Tuner processes and stores grammars and references within its sandbox area. This resolves odd referenced grammar and lexicon issues where those references were n-levels deep. This issue only affected the Speech Tuner's sandbox references.
  • Added new 'Close Tuner Database' menu option in the Speech Tuner to make it clearer to new users how to close the current database. This performs the same operation as 'New Tuner Database' but may be more intuitive.
  • Speech Tuner now allows the use of middle-mouse button to close selected grammar and SSML tabs when using those editors.
  • Speech Tuner now double-clicking selected grammar and SSML documents to open them in the appropriate editor.
  • When saving Speech Tuner files in situations where filters are active, users are now prompted to choose whether to save all interactions and remember the active filters, or to save only the unfiltered interactions.
  • Speech Tuner now allows bulk assignment / removal of the OOC marker for selected interactions.
  • Improved Speech Tuner transcription to reduce latency when moving from one interaction to another. This should be most noticeable to users working extended periods with the Transcriber.
  • Added new performance metrics to the Speech Tuner Confidence Histogram to provide additional in-grammar and out-of-grammar statistics.
  • Improved the Speech Tuner Transcriber View to allow users to configure how function keys can be mapped to shortcut markers on the Marker Toolbar, which should make the transcription process easier. Now users can assign any marker to any of the available function keys, rather than use the previously predefined assignments.
  • Added new Decoded SI list to the Speech Tuner's Summary View, enabling a different way to visualize the statistics of the data.

Changes:

  • Modified the way in which our SimpleMRCPClient and lv_show_config/LvShowConfig utilities detect the localhost IP address in Linux Operating systems. Previously, the first interface returned that was not localhost was used, but now a reverse lookup of the hostname is performed, which is a more reliable method. The corresponding articles for SimpleMRCPClient and lv_show_config have been updated to include a new "Network related warning" in both, which explain this.
  • Modified LVLicenseManager to reload / re-synchronize licenses in the LumenVox License Server when running in Flexible licensing mode. This extends the '-r [ip-address]' functionality that was previously available in non-Flexible modes.
  • Media Server TTS performance was improved significantly by increasing the amount of cached synthesized audio prior to being streamed out. The amount of benefit users will encounter depends largely on the length of audio being synthesized, as well as the overall system load; however our testing shows noticeable improvement in most cases.
  • Improved overall TTS synthesis performance. Less system resources are now used during TTS synthesis, which means higher port densities should be achievable on production servers. Internal testing indicates in excess of 10% overall improvement in synthesis performance in most cases, which should be reflected in the corresponding TTS port densities achievable.
  • Minor TTS pronunciation improvements to German, English, Polish, Swedish, Danish, Russian and Spanish language voices
  • Modified en-US builtin grammars for date, number, time and phone to improve in-grammar coverage and overall performance when using any of these.
  • Improved SSML 1.1 and 'lookup' support for TTS1 synthesis. Now the 1.1 version will be correctly passed along to the synthesizer.
  • Modified the Speech Tuner to permit OOC utterances to have transcription text associated with it. This is done by using ++OOC++ marker at the beginning of the transcript string, which can now be toggled.
  • Modified the Linux versions of SimpleMRCPClientSimpleASRClient and SimpleTTSClient to optionally look for sample audio, grammar and SSML files in /usr/share/lumenvox/client/data/ if not present in the current directory.
  • Modified the Speech Tuner to indicate the confidence threshold seen on screen when saving the histogram image to file. Previously the threshold indicated may not have been what was selected.
  • Modified the Speech Tuner to make the Phonetic Speller tool accessible from all views, instead of being limited to the Grammar Editor View.
  • Modified the Speech Tuner to remember user selections when working with the Phonetic Speller and Advanced Filtering dialogs. User selections will now be remembered the next time these dialogs are used.
  • Speech Tuner Summary View now shows utterance list filtering options as unavailable when multiple selections are made. Previously the operation was still disabled, but visibly appeared as though this was a valid option (it's not)
  • Added a new return code (LV_ACQUIRING_LICENSE_FAILED) as a possible return value to LV_TTS_WaitForSynthesis. This will be triggered if a necessary voice license was not available.
  • Removed the deprecated END_OF_SPEECH_DETECTION option from client_property.conf - this setting should no longer be used. Use the VAD_EOS_DELAY setting instead.
  • Added logging of diagnostic issues when attempting to resolve external URI references. Increased logging verbosity will show more detail. This information can now provide more specific detail when URI issues occur, such as when an HTTPS handshake failure occurs with the remote server, which will now indicate the nature of the URI failure, rather than simply a generic URI failure message.
  • Improved Speech Tuner auto-indent handling within the Grammar Editor and SSML Editor views. Now the correct combination of whitespace characters is inserted, where previously only spaces would be inserted when adding an open-tag marker.
  • Changed the way in which OpenSSL is initialized within our implementation to provide better thread-safety. This change is not associated with any reported errors from the field, but is a preventative measure against possible issues occurring in the future.
  • Added a new warning message to alert Speech Tuner users when their license is due to expire within 30 days. This will be shown at startup of the Speech Tuner, but otherwise not affect any functionality. As always, Speech Tuner license expiration dates can be seen using the Dashboard utility's Licensing page.

Fixes:

  • Fixed a problem with Media Server to allow more case-insensitive support for the 'channel-identifier' and other headers when using MRCPv2.
  • Fixed a problem with Media Server when processing MRCPv2, where recognition would be incorrectly aborted if START-INPUT-TIMERS was received after barge-in was already detected. This change brings MRCPv2 into alignment with our existing MRCPv1 implementation when this situation is encountered.
  • Added missing example files in Linux: 1234.ulaw,8587070707.pcmSimpleTTSClient.ssml and ABNFDigits.gram are now located in the /usr/share/lumenvox/client/data/ folder.
  • Added missing example file to the Windows Engine package to install a sample SSML document named SimpleTTSClient.ssml
  • Fixed a minor issue when accessing shortcut menus in Windows installations, which would always run the batch file after changing to %LVBIN% folder, which was not always the desired or expected behavior. Now the batch files will operate in the expected folder (whichever was selected).
  • Fixed a bug when loading external referenced grammars via HTTPS URI. The parameters of SSL_VERIFYPEER and CERTIFICATE_AUTHORITY_FILE were not being passed on from the parent grammar to the externally referenced child grammars, leading to grammar load failures since default values were being used when resolving those child references, which would not be appropriate in non-default configurations.
  • Fixed a minor bug to remove an unwanted quote mark when using auto-complete in the SSML Editor of the Speech Tuner, when selecting 'spell-out'
  • Fixed a minor bug in the Speech Tuner Call Browser View where 'Clear Selection' and 'Re-evaluate Selection' options did not operate as intended for the selected interactions.
  • Fixed a minor bug in the Speech Tuner Tuning Wizard metrics for Confidence tuning to use the correct statistics and colors in comparisons against the baseline and optimal thresholds.
  • Fixed ASR handling of optional parameters in URIs that appear after the question mark for referenced grammars and lexicons.
  • Fixed TTS SSML parsing issue where & appearing within quotes were being incorrectly processed (within URI references for example).
  • Fixed a minor issue in the Speech Tuner when displaying decoded results overlaid on the audio control's waveform image. If decodes were run within the tuner (using the Tester View), only the resulting decode results would be shown, and not the original. Now switching between the original and output versions is correctly shown.
  • Made a minor css style change to correct a word-wrapping issue on the Licensing page in the Dashboard
  • Made a minor css style change to alter the color of expired maintenance dates from red to orange to reduce the perceived severity on Licensing page in the Dashboard
  • Fixed suppression of HTTPS URIs in the Media Server logs