Browse
 
Tools
Rss Categories

 Featured
Release Notes

Reference Number: AA-01461 Views: 77455 100 Rating/ 2 Voters


19.3.200 (January 21st, 2022):

 Fixes:

  • Fixed issue with character encoding generated by the grammar based DNN ASR engine.  The results were being returned in the UTF8 character set, but the header specified the results are in ISO-8859-1.  
  • Fixed inconsistencies that arise when switching back and forth between using the legacy and DNN ASR engines.
  • Fixed issue with requests to the legacy ASR engine, for requests using digit-only grammars and not making use of the Legacy digits acoustic model.
  • Moved GRAMMAR_ENGINE configuration setting from the SRE section to the GRAMMAR section of the client_property.conf configuration file.

This release currently available for Centos 7 64-bit and Windows.



19.3.100 (January 3rd, 2022):

 Improvements and New Features:

  • Legacy Grammars: The new end-to-end DNN based recognition engine uses SLMs (Statistical Language Models) by default. However, in this release the new engine is fully integrated with the legacy system and can now be used with your legacy grammars. Grammar-based recognition customers can choose to upgrade the engine to get the new DNN ASR technology and enjoy the latest in innovation and accuracy, while keeping the investment in their grammars.
    • LumenVox ASR supports both the previous engine and new DNN one at the same time. You now have the flexibility to choose to use the old or the new engine for your grammars. You can set controls that dynamically determine which version is used. If you choose the new engine, you have flexibility to use it with either grammars or SLMs. For more information on these settings see: Using the DNN ASR Engine with Grammars
    • Note that the new engine ignores phoneme lexicons and in-line phonetic spelling because the acoustic modeling is no longer phonetic based. Therefore, we don’t require lexicons in this release when using the new engine with your grammars. Grammars using lexicons and phonetic spellings still work, but the alternate pronunciations are ignored. In the future we plan to return to supporting phoneme lexicons in new and improved ways.
  • Windows Support: The DNN ASR service is only supported on Centos/RHEL 7 64-bit (Linux).  However, with this release:
    • The media server, the tuner, and applications using the API can access the new engine from Windows.
    • Additionally, the DNN ASR service can now be run in a Docker container on any OS that supports running Linux containers. Many Windows versions support Linux Docker containers. Therefore, using a Docker installation is a viable option for Windows customers who want to use the new DNN ASR engine without migrating to Linux.
  • Efficiencies & Resources: The DNN ASR engine takes more resources than the previous engine.  Decode operations use, on average, approximately twice the CPU resources.
    • However, the new engine provides significant improvement in accuracy and lower error rate. The engine supports operational efficiencies, quantized storage of the language models, and dynamic usage of ports for call transcription based on load, thus reducing the bill as compared to using fixed ports.
  • Cost Saving Language Licensing: With DNN recognitions (both grammar based and transcription), licensing is now based solely on the language, not dialect. 
    • As an example, a grammar specifying "en-US" can use an "en-GB" license.  The language for a grammar does not need to specify dialect. Specifying “en”, or “es”, or any other 2 letter language code would suffice.
    • To use the DNN ASR engine, you must install separate language packs. See Language Installation
  • Word-Level Confidence Scores: When using transcription mode, you can obtain more detailed ASR results in the interpretation section, including word-level confidence scores as well as word-level timing. See ASR_RESULT_DETAIL_ENABLE
  • Wordlists: Use Wordlists for vocabulary customization. The new engine can accept word and phrase listings, to facilitate recognition of words that the system has difficulty recognizing, such as pharmaceutical names or industry jargon – words unique to a domain or business that are not typically found in a general language model.
    • Wordlists are used to augment language models, not grammars (if you want to augment a grammar you just edit the grammar). However, to accommodate the creation of wordlists in a familiar way, we implemented them using a mechanism similar to grammars. A wordlist definition is straight-forward and looks very similar to a grammar.
    • The words and phrases in the Wordlist can be given a probability boost.
    • See more information to learn how to create wordlists and how to set a probability boost for wordlists, to enhance language models:  Wordlists 

This release currently available for Centos 7 64-bit and Windows.



19.2.500 (November 16th, 2021):

Fixes:

  • Fixed errors happening during service shutdown
  • Fixed errors related to accessing grammars via TLS (#4079)
    • Updated CA root certificates
    • Update openssl to version 1.1.1l on Linux
    • Updated curl to 7.79.1 on Linux
  • Fixed issue where Media Server could crash while under load, when using save_waveform and the waveform url prefix began with "file://" (#3841)
  • The Media Server can now retransmit SIP 200 OK messages if the ACK message is not received (#4046)  details
  • Cleaned up various issues with Dashboard Diagnostic tests


19.2.400 (September 9th, 2021):


Fixes:

  • Confidence value normalization was adjusted for the DNN based recognition engine to eliminate artificially low values being returned in some cases.
  • Two-letter language values can used in the grammar triggering transcription mode with the DNN recognition engine.  So for example, the language code "en" could be used, instead of "en-US" or "en-GB"
  • Several issues with running the diagnostic tests in the Dashboard related to the new DNN recognition models have been fixed.
  • Security improvements made to the Dashboard
    • Removed unneeded options in HTTP OPTION response  (PUT, DELETE, CONNECT, WebDAV)
    • Verified that these options are completely removed from code base.
    • Added security headers in HTTP response to ban any cross site scripting, use of frames, or mixing of transport security.
    • Setting autocomplete=off in login field

Known Issues:

  • Dashboard Diagnostic Test "MRCPv1 ASR DNN Test" may return result "Unable to obtain ASR license for language 'undefined'", as long as the other MRCP ASR DNN tests pass, this can be ignored.



19.2.100 (June 28th, 2021):


 Improvements and New Features:

  • Introducing a completely new end-to-end DNN based recognition engine for statistical language and transcription processing. 
    • The new speech engine features a single acoustic model for each language, encompassing all dialects for a given language.  When specifying the language, unless there is a special dialect specific acoustic model, the dialect portion of the language code is ignored.  For example, specifying the language codes en-US or en-GB will working equally well for US or British speakers.
    • The engine can return partial results for a continuous stream of audio.  Partial results include tentative results that may change as more audio is streamed.  The new grammar meta tag PARTIAL_RESULTS and new vendor specific MRCP header delay-partial-decode facilitate this functionality within MRCP sessions.
    • Acoustic model training requires only audio and a text transcript.  A phonetic lexicon is no longer required.
    • The method used to set up a transcription ASR session has been modified.  A new grammar meta tag TRANSCRIPTION_ENGINE is used to begin a transcription session and to choose which version of the ASR engine to use.  (The default is to use the new engine.)
    • For the ASR service, transcription is only supported on Centos 7 64bit.  Windows clients (Media Server and API applications) however, can handle transcription requests as long as they are configured to access, via the network, an ASR service on a centos 7 64bit server.
    • The new transcription ASR service requires a different license.  Licenses for the previous versions of the transcription engine will need to be updated before using the new engine.
  • Significant improvements to AMD and CPA processing.
    • CPA now ignores any specified Wind Back Time (STREAM_PARM_VAD_WIND_BACK). This setting is only useful when running ASR interactions and could previously adversely affect CPA interactions, so is now ignored in favor of internal settings when using CPA
    • Users of CPA and AMD are now strongly encouraged to specify configurations using updated and recommended grammar files, which include updated default settings. This will also be the recommended method in the future. Existing applications may continue using API or other options to specify CPA and AMD setting values, however using grammars file (as described by our Grammars in CPA and AMD article) will continue to override any other settings specified
    • Added a new CPA_MAX_TIME_FROM_CONNECT configuration option, allowing users to specify a maximum time limit to apply to CPA responses. See our CPA_MAX_TIME_FROM_CONNECT article for details.
    • Previously when using CPA, specifying incorrect STREAM_PARM_BARGE_IN_TIMEOUT or STREAM_PARM_END_OF_SPEECH_TIMEOUT settings could adversely affect results in a difficult to diagnose way. Now these settings are automatically set to values 10 seconds greater than the largest CPA value, thus avoiding such unwanted influence. This is done internally within the code, so no application changes are needed to benefit from this change.
    • New BUSY tone detection was added to AMD functionality allowing detection of international busy tones, when enabled, using the new BUSY_CUSTOM_ENABLE configuration option as described in our Customizing CPA and Tone Detection article. Note that busy tones will only be detected within the first 7.5 seconds of the processed audio stream.
  • Speech Tuner has many changes to coincide with AMD and CPA changes as well as to simplify transcription of AMD and CPA interactions, using internally generated grammars when processing these interaction types. See our Running Tests article which describes the new CPA and AMD settings available within the Speech Tuner.
    • Speech Tuner now automatically utilizes streaming mode internally when processing both AMD and CPA interaction types, regardless of the mode selected (i.e. uses streaming mode when 'Load Files' is the selected mode), since this is required in order to correctly analyze these interaction types. Previously if an incompatible mode was selected, these types of interaction remained unprocessed.
    • Speech Tuner now has progress overlaid on the task bar button to more easily visualize the status of lengthy operations, such as when decoding, without needing to explicitly switch to the application.
    • Significant optimization of the Speech Tuner to reduce latency when changing views, most notably when loading and displaying audio, especially lengthy and/or complex audio waveforms. This removes a stalling issue that affected waveform rendering, leading to noticeable delays transitioning between interactions, especially in the Transcriber or Call Browser views, which are now significantly more responsive.
  • SimpleMrcpClient was modified to allow inline grammar text to be sent, rather than the previous implementation which only supported using hosted grammar URIs, which are more complex to utilize. Now both URIs or inline grammar text can be specified using the new -inline_grammar option. See our SimpleMRCPClient article for more details of this and other available settings.
  • Added new customer examples for SimpleCPAClient in C and C++. These are focused at assisting new CPA customers in quickly understanding how CPA can be utilized. These samples include audio and grammar settings files allowing various combinations of CPA and AMD functionality to be easily exercised.
  • Added new information to the Speech Tuner Answers page when displaying Call Properties for an interaction, consistent with updated timing and Begin/End frame markers for CPA. See the new CPA Tuning article for details.

Changes:

  • Added REUSE_SIP_TCP_SOCKET configuration option.  When set to true, the SIP TCP socket used with MRCP v2, is left open when all sessions have been closed.  Some platforms expect the TCP socket to be reused for subsequent sessions.
  • Modified Speech Tuner "NO INPUT" behavior for AMD interactions. Previously these were hard-coded as "SPEECH", which was misleading. Now InputText and SI are configurable within settings (alongside other new AMD settings), meaning that users can now specify which "NO INPUT" values are used. This helps with backward compatibility and allows for more flexibility.
  • Modified Speech Tuner audio stream processing to be 10x faster than before, likely increasing CPU utilization, while dramatically improving throughput. Note that the number of simultaneous decode processing threads can still be defined in advanced settings to control overall CPU use when running the Speech Tuner.
  • Removed CPA and AMD references from MultiThreadedStreamingExample code, since there are now dedicated samples when working with these types of interaction. All users are encouraged to utilize the new SimpleCPAClient applications when working with CPA and AMD
  • Removed previously deprecated STREAM_PARM_END_OF_SPEECH_DETECTION option. This was not widely used and was deprecated some time ago so should not affect any customer implementations, but was needed as part of the CPA enhancements.
  • Internally uses a more optimal value of 0 for STREAM_INIT_DELAY when using CPA.
  • The removal of the Wind-Back-Time setting when processing CPA improves calculation of begin/end frames allowing more accurate analysis of CPA results within the Speech Tuner, especially for very short-duration interactions. 
  • The entire start of the audio stream is now saved to Response (callsre) Files when CPA mode is selected (no longer trimming silence chunks from the beginning), also allowing clearer analyses within the Speech Tuner.
  • Speech Tuner configuration option sliders now move in appropriate increments and can be adjusted using keyboard arrows for finer control.
  • The Transcriber view within the Speech Tuner now shows the begin/end frame and word markers for CPA and AMD interactions, making it much clearer to visualize when related classifications happen within an audio stream.
  • Added a new hidden configuration setting in client_propert.conf (CALLBACK_THREAD_STACK_SIZE_INCREASE) which specifies how much extra memory should be allocated as the stack size for the status change callback thread with the c/c++ API.  This can be helpful for cases where other coding languages are used to access the API (Java for example).

Fixes:

  • Fixed previously incorrect default SIT INPUT TEXT values to their appropriate values. Users were highly unlikely to encounter this issue since grammars specify them. Specifically, if a grammar does not specify these default text values (which it should) an internal set of values is used. These were incorrectly set to SIT instead of their more specific values (such as "SIT NO CIRCUIT DISTANT"). Nevertheless, these default values were wrong and have been corrected.
  • Modified Speech Tuner to return correct Tone types, rather than a generic "BEEP" during decodes for AMD interactions.
  • Changed Speech Tuner AMD processing to show more appropriate "NO INPUT" instead of "SPEECH" when barge-in-timeout reached. New configuration options for AMD Input Text, SI and which tone detect algorithms are enabled. Semantic Interpretation processing for AMD interactions is now matched to new these new configuration settings and is no longer forced to hard-coded values (which would not have matched non-standard grammars previously)
  • Fixed an issue where the TTS Server IP address was not being correctly applied if the speech tuner setting differed from the value in client_property.conf, preventing the Text To Speech View from working correctly.
  • Fixed issue where the logging-tag was not put into the CDR file when the logging-tag was set as a header in the MRCP RECOGNIZE or SYNTHESIZE message instead of via the MRCP SET_PARAM message.

This release available for Centos 7 64-bit.


Known Issues:
  • Using a two letter language code with the new DNN transcription engine can in some circumstances result in errors, recommend always using the four letter language specifier (example: use "en-US", not "en")
  • The confidence score normalization for DNN transcription sometimes result in artificially low scores.  (This will be addressed very shortly will a follow up point release.)



19.0.200 (March 1, 2021):


Fixes:

  • Fixed issue with setting the vendor specific setting secure_context in MRCP session.  The value being set within the session was being ignored
  • Fixed small memory leak when streaming audio to speech recognition service when doing transcription.
  • Removed unused settings in the configuration page in Dashboard for NLU provider configuration.
  • Small UX improvements to the configuration page in Dashboard for NLU provider configuration.
  • Fixed issue with vendor specific setting sticky-save-waveform.  sticky-save-waveform is used to over-ride the standard save-waveform MRCP setting.  When sticky-save-waveform was set to true within a session, the audio was being recorded, but the URL to the recorded audio was not being passed back in the MRCP messages.  


19.0.100 (January 31, 2021):


Improvements and New Features:

  • Introducing the LumenVox Natural Language Understanding (NLU) Gateway.  
    • This new service allows applications to take advantage of third-party services that provide natural language processing for supplied text.  
    • Google DialogFlow, Amazon Lex, Microsoft LUIS and IBM Watson are supported.
    • This supports the intent classification and entity determination, as well as conversational bot interaction.  
    • The LumenVox gateway directly communicates with various NLU processing cloud services and sends the text transcription of the spoken audio (using the LumenVox Transcription Engine).
    • The NLU results are returned to the speech application within the semantic interpretation typically returned for grammar based speech recognition.  
    • The results from the various vendors are normalized to a common format.  
    • The NLU Gateway handles all communication and authentication.  
    • This allows existing and new speech applications take advantage of NLU processing with very little to no extra programming.
    • Configuration for NLU end-points is done through the dashboard.
    • Activating NLU results for a transcription interaction is done via grammar meta tags.

  • Lumenvox Transcription ASR now supports up to 15 minutes of audio within a single interaction.
  • Audio sent for transcription is now streamed to the ASR server, greatly reducing the time needed after all audio is streamed until the full transcript is available.
  • Handles case of ASR server going offline during a transcription interaction.  Client will switch to batch mode and send complete audio to an alternate ASR server.

Fixes:

  • Increased time-out value for communication with Lumenvox cloud flexible license server.  Previous low timeout value was causing a number of successful flexible license synchronizations to incorrectly report failures, prompting a retry attempt.
  • Previous version stopped logging SIP messages related to MRCP v2 sessions.  This version logs full SIP messages again.
  • Under certain heavy load conditions when using MRCP V1, resource handles were not being released.  This version fixes that issue.
  • Fixes issue with Dashboard diagnostic testing when using just one type of license.  Previously diagnostic would report no ASR or TTS licenses installed, when only a single type of license was installed (ex. TTS only).

Changes

  • Removing support for Centos 6 / RHEL 6

Known Issues:

  • The Media Server secure_context setting value does not save when changed in the dashboard.  To change this setting, direct editing of the media_server.conf file is required.  This will be fixed in the next version.



18.0.450 (September 1, 2020):

Fixes:

  • Fixed issue where callsre interactions for a single session would get saved into multiple callsre files.  This happened when the config settings for enable_sre_logging where set to zero, and the MRCP vendor specific parameter (enable-sre-logging) was used to turn on call sre logging for a session.
  • Fixed issue when the vendor specific parameter (callsre-prefix) was used with MRCP, to supply a custom prefix to callsre files, only the first callsre file for a session had the custom prefix.

This release available for Centos 7 64-bit and Windows.


18.0.400 (May 4, 2020):

Improvements and New Features:

  • Added Media Server support for MRCPv2 over the TLS protocol. This includes support for secure SIP (SIPS) connections and encrypted audio packet transmission over the SRTP protocol. 
  • Added support for Windows Server 2019 Operating Systems, and deprecated support for Windows Server 2008 and Windows 7.
  • All MRCP headers can now be overwritten as a Vendor-specific-parameter. If a header appears in both the header and the Vendor-specific-parameter header, the one in the Vendor-specific-parameter will override the other setting. This is to allow customer the ability to change MRCP properties that a given platform may not allow. 
  • A new Vendor-specific-parameter called "com.lumenvox.sticky-save-waveform" has been added. It can take a value of either "true" or "false". If set to true, regardless of the save-waveform header value, the save-waveform option will be set to true for the remainder of the MRCP session. Likewise, if set to false, regardless of the save-waveform header value, the save-waveform will be set to false for the remainder of the MRCP session.
  • Our SimpleMRCPClient has been updated with a new vsp command line option the will append any text provided to the Vendor-specific-parameter header.
  • Our SimpleMRCPClient has been updated to allow testing of TLS features.
  • Windows product Installer have been stream line to allow easier product installation.
  • Migrated Windows development to Visual Studio 2017 from VS2013 in previous versions. This means that the run-time libraries we are built against are now using the VS2017 versions, and our sample projects for Windows have also been updated to reflect this change. 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 Diagnostic support for MRCPv2 over the TLS protocol functionality. In addition to recently released changes, the Dashboard's Diagnostic capabilities were extended to include specific tests for this functionality. 
  • Added ability to backup and restore configuration settings from the diagnostic page of the Dashboard.
  • Added support to load alternate language models when using short utterance transcription.

Changes:

  • The implementation of OpenSSL in this version of LumenVox has been updated from 1.0.2k to 1.1.1d, which addresses a number of recently detected vulnerabilities. 
  • The ECMAScript engine used for processing Semantic Interpretation data was updated to use the latest Spider Monkey version  70.0.1 library.
  • Updated internal DNS mechanisms for better and more reliable name resolution. Should be mostly unnoticed by users.
  • Updated how the Media Server internally processes TCP as part of ongoing longer-term enhancements. This set of changes should be slightly more efficient than previous implementations, but should be mostly unnoticed by users.
  • Update ASR diagnostic tests to not run with a license that only has CPA only and not ASR.
  • The LumenVox icon used for applications was updated.
  • Default TLS certificates for the Dashboard web page and SIPS are now created during installation.  Previously, the default certificates were shipped with the installation.

Fixes:

  • The base URI used for loading Lexicon resources from a grammar is now set to the base URI for the document requesting the lexicon.  Previously the base URI from the root grammar was being used.
  • Fixed an issue causing the Media Server to crash when the "@" symbol appeared outside brackets in the SIP header.
  • The SAVE_WAVEFORM toggle in media server configuration page of the Dashboard now loads with the proper value.
  • For the German language, if a digits only grammar used lower case umlauts characters (example "ü"), the special higher accuracy digits acoustic model was not being automatically used.  This has been fixed.
  • Fixed issue where config files were getting double comment header messages
  • Fixed issue with the Dashboard, where the Manager service would not start when the License Server is using an alternate communication port
  • Fixed issue when failing over to backup License Server.  When the network connection was abruptly cut off, the fail-over could take too long, causing loss of licensing.  This works correctly now.

17.0.650 (June 5, 2020):


Fixes:

  • Fixed issue where callsre interactions for a single session would get saved into multiple callsre files.  This happened when the config settings for enable_sre_logging where set to zero, and the MRCP vendor specific parameter (enable-sre-logging) was used to turn on call sre logging for a session. When the vendor specific parameter (callsre-prefix) was also being used in this condition, only the first interaction got a custom prefix, subsequent interactions were saved into separate callsre files with no prefix.  This point release fixes this issue.  This point release is only available for Windows and Centos-7 64bit.

17.0.600 (July 12, 2019):


Fixes:

  • Fixed threading deadlocking issue in the Media Server.  In rare cases, when a session was closed while TTS synthesis was still active, the session cleanup could trigger a deadlock.  This would result in the Media Server no longer being able to accept new MRCP sessions.  This bug was introduced into the code base in Lumenvox version 15.0.

17.0.400 (March 23, 2019):


Improvements and New Features:

  • Added Diagnostic support for LumenVox Transcription ASR  functionality. In addition to recently released changes, the Dashboard's Diagnostic capabilities were extended to include specific tests for this functionality. New diagnostic tests include checks for licensing, functionality as well as confirmation that the base language pack for all Transcription ASR languages being used are correctly installed. As a reminder, when using a Transcription ASR language, the corresponding "regular" ASR language pack for that language much also be installed.
  • Improvements and optimizations to the Dashboard performance using compressed JavaScript and CSS references, leading to faster page loading times.

Fixes:

  • Fixed an issue specific to using recent LumenVox Transcription ASR, where when working with decodes following the use of Transcription ASR, further results were formatted as Transcription ASR result instead of being consistent with the request type (Transcription ASR or non-Transcription ASR). This situation persisted for the duration of each active port or session. This change corrects this incorrect behavior to allow a mix of Transcription and non-Transcription ASR decodes within a single speech port or MRCP session (assuming appropriate Tier4 licenses for the language being used is available)

17.0.200 (February 28, 2019):


Improvements and New Features:

  • Added support for LumenVox Transcription ASR functionality. Note that this is only supported on 64-bit RHEL 7 / CentOS 7 Operating Systems. These LumenVox Transcription ASR models will need to be installed in addition to their base ASR language models in order to utilize this functionality. Customers wishing to take advantage of this new functionality will also need to purchase appropriate TIER4 licenses for supported languages.
  • Added LumenVox Transcription ASR support mentioned above for the following languages:
    • American English (en-US)
    • British English (en-GB)
    • Mexican Spanish (es-MX)
    • Colombian Spanish (es-CO)
    • Brazilian Portuguese (pt-BR)
  • Added a new OUT_OF_SERVICE configuration option for the ASR service, allowing system administrators to enter maintenance mode from the Dashboard, which permits currently pending requests to be completed, but any new requests will be rejected (to be potentially handled by other ASR servers in the cluster). A message is reported to the asr_server_app.txt file when the ASR may safely be shut down after all pending requests have been completed. Starting the service in this mode will have a similar effect (rejecting any new requests). Note this functionality is NOT backward compatible, so will only work if both client and server are version 17.0 or later and have these changes. This functionality allows ASR servers to be taken down and brought up in a controlled manner when situated in a production cluster.
  • Added a new feature to the ASR load-balancing mechanism to actively route ASR requests based on the language specified. Previously, load-balancing and fail-over required all ASR servers to be configured for all possible ASR languages needed for all applications being run. This change means that ASR servers can be configured in clusters according to the languages they need to serve. This allows users to have certain ASR servers configured for single or multiple languages (including the new LumenVox Transcription ASR languages) and have the load-balancing software automatically determine how best to route requests. Note that this change is not backward compatible, so users wishing to take advantage of this should upgrade all of their LumenVox components.
  • Consistent with our recent merger, the LumenVox icon used for applications was updated.

Changes:

  • Changed the method of loading multiple ASR languages so that any errors encountered during startup are better represented in the list of available languages shown in the DashboardSpeech Tuner and diagnostic testing. This corrects an issue where some customers attempting to load a mismatched or incorrect language were seeing incorrect error messages during diagnostic testing.
  • Added support for unconventional use of a space that follows the colon within grammar session URI within MRCP requests (note: we believe this syntax is incorrect, however we have seen some customers using this). For example "session: request1@form-level.store" (with space after colon), instead of the more correct "session:request1@form-level.store" (without space) will now be handled appropriately.
  • Updated how the Media Server internally processes RTP as part of ongoing longer-term enhancements. This set of changes should be slightly more efficient than previous implementations, but should be mostly unnoticed by users.
  • Modified the Manager to mask passwords in the configuration page by default. There is now a toggle button to display the password when needed. Also various optimizations were made to the configuration page (minimized js and css) so that the page should load faster.
  • Modified sample SimpleASRClient_cpp and SimpleASRClient_c code to allow the use of NBest, DecodeTimeout and SaveSoundFiles Settings from client_property.conf without overriding these values within the application, although users may continue doing so if they wish.
  • Updated the ThirtdPartyLegalNotices.txt document to include periodic changes.

Fixes:

  • Fixed an issue specific to using the Speech Tuner with Windows 10, where the trailing period when importing callsre (Response) files using "FolderSelection." was being ignored. It seems Windows 10 is (potentially incorrectly) removing the trailing period from the name, which caused this functionality to break. This change accommodates the unusual Windows 10 behavior to restore this functionality.
  • Fixed an issue related to Speech Tuner grammar handling, when two or more buffer grammars were loaded for same interaction, only one would appear in the Tuner grammar list.
  • Fixed an issue related to Speech Tuner when loading a grammar, where the root rule was being used in the grammar "name" appearing in the grammar list. It should have displayed the label, which is now corrected.
  • Fixed an issue where when using compiled grammars in the Speech Tuner the descriptive text of the compiled grammar was being saved to the local temp directory instead of in the binary compiled grammar itself. Users implementing pre-compiled grammars are advised to update to this version of the Speech Tuner.
  • Fixed a bug in the Manager where setting the PROCESS_MONITOR_LIST to anything other than the default caused the dashboard web-service to not start.
  • Fixed an issue specific to Cisco Virtual Voice Browser where the SDP lines were not terminated with the expected carriage-return and line-feed characters, which forced a parsing issue. This ultimately caused the Cisco system to not receive a response to ACK messages, causing apparently "stuck" sessions. Additional logging was added to more readily identify similar issues if they happen again. Customers running Cisco Virtual Voice Browser are encouraged to upgrade to this version to take advantage of this change.
  • Fixed TTS1 pronunciation for the state NE (Nebraska) which was incorrectly being pronounced as "North East" when using SSML say-as interpret-as "address" with format as "us-state". Note that if format="us-state" is not specified, the appropriate pronunciation would be "North East" since the request is still an address. This was a regression introduced in release version 13. Users expecting state abbreviations to be read out correctly when using the interpret-as="address" and format as "us-state" should consider upgrading to take advantage of this change.
  • Fixed an issue with CPA which required both meta values "STREAM|VAD_EOS_DELAY" and "STREAM|PARAM_VAD_EOS_DELAY" to be present in a grammar in order to change the VAD_EOS_DELAY setting being used. "STREAM|VAD_EOS_DELAY" is the correct value, however the issue forced the use of the default 1200ms instead of the expected value being specified if only one meta tag value was specified. The workaround for users prior to this fix is to assign the same value to both meta tags. From now on, users wishing to change this delay should use the "STREAM|VAD_EOS_DELAY" meta tag in the grammar. This issue should not affect many users.

16.0.200 (July 25, 2018):


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

  • The Media Server was modified to address a specific load-related issue detected on some systems running the Broadsoft Broadworks platform when using a specific load profile, which caused a very small number of calls to be dropped under significant load. The change is a minor alteration to internal timing when processing INVITE requests.

16.0.100 (February 15, 2018):


Improvements and New Features:

  • Added support for a new Italian (it-IT) model, as our tenth supported ASR language. This includes the additions of new Italian-language boolean, currency, date, digits, number, phone and time "builtin" grammars.
  • Added a new VOICE_MAPPING configuration option to the TTS Server, allowing a comma-separated list of user-defined name mapping from application-specific names to corresponding LumenVox TTS voice names (i.e. Joe=Chris,Mary=Jackie) to be specified. Requests made using application-specific voice names will be correctly mapped to the actual LumenVox voice prior to synthesis. This feature may be particularly useful for users migrating legacy applications from different vendors to LumenVox servers, where the original voice name does not match up directly to LumenVox TTS voice names. This feature means that developers can avoid making changes to existing applications simply to migrate them over to LumenVox.
  • Added new SNMP trap messages for lvAlarmSyncFailure and lvAlarmSyncRestore. First occurrence issues a Minor severity alert. After 10 consecutive sync attempts fail, a second (Critical) alert will be raised. Restore messages will be sent as appropriate when sync is restored (when the license server is able to synchronize with the licensing nodes). Customers will need to update their LUMENVOX-SNMP-MIB to include these new definitions. These traps are enabled with the existing "ENABLE_COMM_TRAPS" setting in the manager (not ENABLE_LICENSE_TRAPS).
  • Added a new German Digits acoustic model, which provides enhanced ASR performance when working with German digits-only recognition. Any recognitions using the "de-DE" language specifier that contains only German numerals (zero through nine) will automatically use this acoustic model internally (users do not need to change their applications in order to receive the benefit of this enhanced performance).

Changes:

  • The implementation of OpenSSL in this version of LumenVox has been updated from 1.0.1t to 1.0.2k, 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.
  • Updated the builtin currency grammar for American English (en-US) to remove unwanted phrases and greatly simplify the grammar to optimize performance.
  • Optimized the way in which SIP messages are processed within the Media Server when multiple messages are bundled into a single packet by network optimization mechanisms.
  • Improved parsing of SSML <say-as> time and vxml:time "interpret-as" modifiers, due to many users incorrectly including the colon within the SSML string being synthesized by the TTS engine, or using incorrect combinations of suffix. Now multiple variants of time with or without the colon and with various AM/PM suffixes, such as "1234a", "12:23AM", "1234a.m.", etc. should be handled more consistently. This effectively adds more flexibility to how users can specify time strings. To clarify - the previous implementation was correct, however many users were not sending the correct syntax, and were therefore not getting expected results.
  • The previously deprecated Call Indexer service will no longer be supported and has been removed from the installation packages
  • Adding support for the Windows 2016 Server Operating system

Fixes:

  • Fixed the diagnostic test which confusingly reported NUM_DECODE_THREADS as 0 when the ASR was not running. Now reports a much clearer message of 'Skipped NUM_DECODE_THREADS check - ASR is not running'
  • Fixed a Media Server issue where failure to fetch SSML documents located on external HTTP servers in a timely manner was causing undesired behavior. Specifically, if the fetch took an unusually lengthy amount of time to respond, and during that time, the session was closed by the client, and internal exception was forced.
  • Minor fix to the Dashboard diagnostic connectivity test 1 (attempting to reach www.lumenvox,.com) when an HTTP 301 response code was used (from a web service redirect).
  • Fix for LOG_TTS_EVENTS and VISEME_GENERATION values not being correctly updated in the Client Property section of the Media Server configuration settings in the dashboard.
  • Fixed an obscure license usage tracking bug due to internal variable value overrun during certain calculations, resulting in the under-reporting of the seconds used counter. Some clearer logging was also added in the affected area.
  • Fixed an obscure license usage tracking bug due to a license request being made after a port shutdown had already begun. The result of this would be an indication of a license being "stuck" in use.