Release notes 6.3.0

Release date: 21st November 2025

Summary

This page highlights all the changes, new features and bugs addressed within the LumenVox Containers version 6.3.0 release. This change affects Speech products. This version is not available for Voice biometric products, this will be made available in upcoming releases. 

This release builds upon the 6.2.0 release - see https://lumenvox.capacity.com/article/892684/release-notes-6-2-0 .

These release notes include changes as part of the following patch releases: asr-6.2.1, asr-6.2.2, grammar-6.2.2, grammar-6.2.3, itn-6.2.1, license-6.2.1, license:6.2.1-beta, neural-tts-6.2.1, mrcp-api-6.2.1, neural-tts-6.2.2, neural-tts-6.2.3, neural-tts-6.2.4, vad-6.2.1, vad-6.2.2, vad-6.2.3, vad-6.2.4-beta, vad-6.2.5-beta

Highlights

1. Further enhancements to Redis to assist with scaling

   • Under load, high latency in Redis mitigated by reducing pull/push notification threads to Redis improving performance and reducing no-input errors
   • Notifications to stream updates within services streamlined
   • ASR & VAD listeners modified to reestablish dropped connections
   • Improved logging implemented to troubleshoot dropped connections
2. MRCP service changes made to address TLS under load producing inconsistent results. NOTE: If running with TLS enabled interactions on a single MRCP-API instance must be limited to 100-150
3. New TTS voices for Galician & Valencian & revised voice for Portuguese
4. SSML supported added for Swedish & Dutch TTS voices
5. Grammar modifications 

   • Modified cache hit rate to improve loading of grammars (resolve cache misses)

   •  Reduce URL fetch roundtrips during loading

   • Implement grammar cache versioning mechanism 

6. TTS tester in the Analysis portal modified to allow for TTS synthesis with plain text or SSML
7. Prometheus metrics added to track active interaction requests for various products at a session level

   • session_active_asr_requests

   • session_active_tts_requests

   • session_active_cpa_requests

   • session_active_amd_requests

   • session_active_transcription_requests

   • session_active_enhanced_transcription_requests

8. Enhancements made CPA and AMD to cater for new Apple Call Screening
9. Improvements made for scaling of CPA & AMD interactions
10. Continued security patch updates

What’s new on LumenVox Cloud 6.3.0 

New features

• ITN updated to return start and end time of redacted words

• New TTS voices for Galician & Valencian & revised voice for Portuguese (Bel)

• TTS tester in the Analysis portal modified to allow for TTS synthesis with plain text or SSML
  
  

Updates

• Further enhancements to Redis to assist with scaling

  • Under load, high latency in Redis mitigated by reducing pull/push notification threads to Redis improving performance and reducing no-input errors

  • Notifications to stream updates within services streamlined

  • ASR & VAD listeners modified to reestablish dropped connections

  • Improved logging implemented to troubleshoot dropped connections

• MRCP service changes made to address TLS under load producing inconsistent results. NOTE: If running with TLS enabled interactions on a single MRCP-API instance must be limited to 100-150

• SSML supported added for Swedish & Dutch TTS voices

• Grammar modifications 

  • Modified cache hit rate to improve loading of grammars (resolve cache misses)

  •  Reduce URL fetch roundtrips during loading

  • Implement grammar cache versioning mechanism 

• Prometheus metrics added to track active interaction requests for various products at a session level

  • session_active_asr_requests

  • session_active_tts_requests

  • session_active_cpa_requests

  • session_active_amd_requests

  • session_active_transcription_requests

  • session_active_enhanced_transcription_requests

• Enhancements made CPA and AMD to cater for new Apple Call Screening

• Improvements made for scaling of CPA & AMD interactions

• Security patches made to address high & critical security scanned vulnerabilities

• Updated the Helm chart so that the management-api pod receives the same RabbitMQ TLS settings as the other services by default.

• Enhancements made to the new 6.2 enhanced transcription functionality to address semantic integration issues for Apple

• Changes made to apply audio score in the absence of the GPU ASR model

• Enhancements made to improve potential blocking scenarios by increasing the audio buffer size and to some internal go-routines.

• Additional Prometheus metrics to track reported no-input errors”

  • vad_stream_subscribe_duration – Histogram of latencies for Redis stream subscription, which monitors subscription latency. The expected value here is less than a millisecond.

  • vad_transcoding_duration — Histogram of latencies for transcoding audio chunks.

  • vad_processing_duration — Histogram of latencies for engine processing of audio chunks.

• Updated logging to log deployment ID more often within log files which should assist with future troubleshooting

• General enhancements to license reporting & refactoring of the license cache

• The following Go SDK enhancements have been made

  • OAuth & support for session grammar loads added to Go SDK

  • Add error detection for finalized calls

  •  Add support for InteractionBeginProcessing

  • Updated non-audio examples to use NO_AUDIO_RESOURCE

• Added transcoder to VAD to handle any sample rate

• Resolved TTS issue with lexicon not working if SSML is missing the 'xml:id' attribute

• Resolved grammar caching issues when switching from 4.X to 5.X ASR model

• ITN modified to not fully capitalize email addresses, names and addresses

• CPA modified to cater for new Apple call screening comfort music which was triggering a false positive beep detection

• AMD issue where the beep tone does not exist but we detected a beep with high confidence score resolved

• Vulnerability scan issues on the admin & deployment portal resolved

• Various minor updates to the Admin & Deployment portal made e.g. updating help information, fixes to analysis reports, and improvements to user functions in the analysis portal

• RabbitMQ TLS warning messages in Management API pods resolved

• Resolved Built-in DTMF Boolean grammar not working through MRCP

• Issue with IPA phonemes not being applied in empty tags for neural TTS resolved

• Issues with incorrect semantic values being returned with enhanced transcription resolved

• Empty grammars now report an error and fail to compile

• Exception in http handler of grammar service resolved

• TTS issue reported referencing pre-recorded audio in SSML resolved

• TTS issue with empty audio tag not working resolved

• Issue with Megan voices not working if this is the only neural TTS voice installed resolved

• Improvements made to the English model to improve accuracy on spelled out email addresses

• ITN & VAD memory leak resolved

• Issue with grammar service crashes when an unreachable grammar URL is specified resolved

• Large audio files can now be archived

• Issues with capitalization in ITN reported fixed

• Fixed session configuration default setting for sessionArchiving which was not displaying

• Grammar Counter grammar_active_load_requests becoming negative resolved

Installation notes

The following helm chart can be used

Helm Chart

Note that for MRCP there is no helm chart but a docker compose file. MRCP will run on its own Docker virtual machine which will integrate into the Kubernetes cluster.

Run the following command helm repo update to update the helm charts

Note: if using TTS we recommend you add the following toggle into the values file legacyEnabled. To enable legacy TTS this must be set to True, and False to enable the new neural TTS. The new neural TTS voices must be loaded in the values file in order for the models to be retrieved from S3 

 ttsLanguages:
   - name: "en_us" 
     legacyEnabled: false
     voices:
       - name: "jeff"    
         version: "4.0.0"       
       - name: "megan"         
         version: "4.0.0"

If installing ITN, the following changes are required in the helm charts:

itnLanguages:
  - name: "en"
  - name: "es"

Key installation guide changes:

LumenVox recommends that a minimum of version 1.31 Kubernetes.

The following versions of required software are supported:

• PostgreSQL: 17.4 (recommended)

• MongoDB: 8.0.11 (minimum)

• Redis: 8.0.3 (minimum)

• RabbitMQ: 4.1.1(recommended)

• NGINX: 1.12.4

• Linkerd: edge 25.8.4
  
  

The full software support matrix can be found here: LumenVox Kubernetes software support matrix 

Upgrade procedures

Upgrade or migration from previous versions is supported. Please contact LumenVox to discuss. See notes above if upgrading from 4.7.

If you are performing an upgrade, you need ensure that your NGINX versions are updated from 1.11 to 1.12.

If installing ITN, the following changes are required in the helm charts:

itnLanguages:       
   - name: "en"        
   - name: "es"

Updated API guide

APIs for all speech products available on version 6.2 can be obtained here:  View full size  https://developer.lumenvox.com/6.3.0/

Information for voice biometric products relates to version 3.4.0-3.4.3 

Model versions as part of the release

• ASR - 4.1.0 (4.1.1 English)

• 5.0.0 New encoder & decoder English model released offering better accuracy but will require additional ASR pod scaling 

• TTS - 4.0 (Neural TTS) sample rate 24 & 16 - can be down sampled to 8kHz (note change). Legacy TTS models will still run under version 1.0. Further voice enhancements are currently being released in TTS voices version 4.0.1 so LumenVox recommends that clients cater for 4.0.X)

• VB - 2.1.15

• VB incorporates Selene 2.4.3 which was integrated into the Container stack
  
  

Model version changes

• 4.0.0 Neural TTS model released for Galician (Xela) and Valencian (Marta)

• 4.0.0 Neural TTS voices for Swedish & updated to support SSML