Release Notes 7.0.0

Release date: 17th February 2026

Summary

This page highlights all the changes, new features and bugs addressed within the Capacity Private Cloud (formerly LumenVox Containers) version 7.0.0 release. This change affects Speech products. This version is not available for Voice Biometric products; this will be made available in future releases.

This release builds upon the 6.3.0 release — see https://privatecloud.capacity.com/article/294286/release-notes-6-3-0

These release notes include changes as part of the following patch releases: grammar-6.3.1

Highlights

1. ASR modelling backend was updated to address security vulnerabilities. This is a breaking change, so new ASR models for all languages need to be downloaded and installed (ASR model version 7.X).
   • The English ASR language model has had its vocabulary expanded and improvements have been made for alpha numeric repetition.
   • All existing ITN models have also been updated to cater for new ASR engine.
2. High-Definition ASR models released for English, Spanish, French, Italian, Portuguese and Japanese with higher accuracy.
   • The ASR models include pronunciation model updates for Apple Hero word recognition.
3. Changes made to allow clients to dynamically switch between standard and High-Definition models.
4. A custom Apple Language and Acoustic model is released.
5. New Neural TTS released featuring improved quality voices and functionality (all neural voices to be used as version 8)
   • 19 new high quality US English voices released including 2 new Capacity feature voices (Caspian & Aurora).
   • New TTS voices released for Catalan, Galician and Valencian.
   • TTS now offers over 100 voices across 30 languages & dialects
6. CPA & AMD updated to detect Apple call screening.
7. Capacity's new Neuron service powering NLU intent & slot detection has been released. A new Neuron testing tool has also been added to the Deployment Portal.
8. A Beta version of the new Post Processing Profiles feature is released into the Deployment Portal for users to add regex rules for ASR post processing.
9. Admin and Deployment Portals have been restyled and rebranded to Capacity.
10. The Admin Portal now provides language support in Spanish, French, German and Portuguese.
11. Admin settings are now available to limit access to custom models per deployment, which may be accessed in the Admin Portal
12. Grammar testing tool now includes a parse tree output and provides better grammar error feedback, as well as support for built-in grammars and grammars loaded via URL.
13. File Store functionality has been added into the Container stack to assist clients with local storing and retrieval of grammars, phrase lists, SSML, Lexicons etc. This has been added to the Deployment Portal.
14. Sentiment Analysis services were moved out of the NLU pod to ITN thus no longer requiring GPUs to run
15. Language Identification now allows for configuration to limit the scope of languages detected.
16. Istio service mesh support was added. Linkerd remains the default service mesh.
17. Persistent volume was refactored to simplify the storage structure
18. Various vulnerability scan issues resolved
19. Support was added for 2-channel raw audio (ULAW, ALAW, PCM16).
20. The Grammar manager has been refactored and includes improved error handling & messaging.
21. Update Neural TTS, Resource, License and Grammar services to apply http proxy environment variables when connecting to a web server
22. Redis & rabbit were optimized to improve performance when scaling. We recommend that session, vad and lumenvox-api pods not exceed 100 concurrent interactions per pod
23. Introduced an updated ECMAScript 2023 (ES14) engine used when processing SISR grammars.
24. Helm chart changes were made to allow for auto scaling and resource reservations (for nodes & pods) and limitations for all services.
25. Go-SDK released from Beta

What's new on Capacity Private Cloud 7.0.0

New features

• ASR modelling backend was updated including upgrade of base OS Ubuntu to 24.04 image and Python to version 3.12 to address security vulnerabilities. This is a breaking change, so new ASR models for all languages need to be downloaded and installed (ASR model version 7.X).
  • The English ASR language model has had its vocabulary expanded and improvements have been made for alpha numeric repetition.
  • All existing ITN models have also been updated to cater for new ASR engine.
• High-Definition ASR models released for English, Spanish, French, Italian, Portuguese and Japanese with higher accuracy.
  • The ASR models include pronunciation model updates for Apple Hero word recognition.
  • The following helm chart changes are used to enable hidef models:

asrLanguages:
  - name: "en"
    version: "7.0.0"
customAsrModels:
  - name: "asr_encoder_hidef_model_en"
    version: "7.0.0"

Changes made to allow clients to dynamically switch between standard and High-Definition models.

• A custom Apple Language and Acoustic model is released.
• New Neural TTS released featuring improved quality voices and functionality (all neural voices to be used as version 8)
  • 19 new high quality US English voices released including 2 new Capacity feature voices (Caspian & Aurora).
  • New TTS voices released for Catalan, Galician and Valencian.
  • TTS now offers over 100 voices across 30 languages & dialects
• CPA & AMD updated to detect Apple call screening — for more information see our API Documentation.
• The Neuron service powering intent & slot detection has been released, along with a testing tool in the Deployment Portal.
  • This may be enabled in the helm charts as shown here, where the languages and models can be defined:

enableNeuron: true
neuronLanguages:
  - name: "en"
neuronModels:
  - name: "banking"
    language: "en_us"

• A Beta version of the new Post Processing Profiles feature is released into the Deployment Portal for users to add regex rules for ASR post processing.
• Admin and Deployment Portals have been restyled and rebranded to Capacity.
• The Admin Portal now provides language support in Spanish, French, German and Portuguese.
• Admin settings per deployment to limit access to custom models has been added to the Admin Portal
• Grammar testing tool has been updated to supply a decision parse tree and provide better grammar error feedback. Built-in grammars or grammars loaded via URL can also be tested.
• File Store functionality has been added into the Container stack to assist clients with local storing and retrieval of grammars, phrase lists, SSML, Lexicons etc. This has been added to the Deployment Portal.
  • We recommend that customers enforce whitelist access to all ingress services:
    • admin-portal-ingress
    • deployment-portal-ingress
    • file-store-ingress
    • lumenvox-api-ingress-grpc
    • management-api-ingress
    • reporting-api-ingress-grpc
  • The following helm chart change enables this service (enabled by default):

enableFilestore: true

• Sentiment Analysis services were moved out of the NLU pod to ITN thus no longer requiring GPUs to run
• Language Identification now allows for configuration to limit the scope of languages detected — InteractionCreateLanguageIdRequest:target_languages
• Istio service mesh support has been added. Linkerd remains the default service mesh. The following helm chart change enables this support:

serviceMesh:
  type: "istio"

• Persistent volume was refactored to simplify the storage structure
• Various vulnerability scan issues resolved
• The API now supports 2-channel raw audio (ULAW, ALAW, PCM16). The AudioFormat message now has an additional field channel_count to specify channel count for RAW audio input. See: Protocol Documentation
• The Grammar manager has been refactored and includes improved error handling & messaging.

• Update Neural TTS, Resource, License and Grammar services to apply http proxy environment variables when connecting to a web server. The following helm chart changes enable HTTP proxy settings:

httpProxy:
  enabled: true
  proxyUrl: "http://proxy.capacity.com:3128"
  noProxy: ".local,.lumenvox"

• Redis & rabbit were optimized to improve performance when scaling.

• Introduced an updated ECMAScript 2023 (ES14) engine used when processing SISR grammars, moving from the previous ECMAScript 2020 (ES11) engine in version 6. Please refer to the online ECMAScript documentation for details. There should be no breaking changes for most users in this upgrade since ECMAScript is relatively stable and is designed to maintain backward compatibility.
• Helm chart changes were made to allow for auto scaling and resource reservations (for nodes & pods) and limitations for all services.
  • A new helm chart configuration allows clients to specify a default deployment id to fully allow installation automation:

enableDefaultDeployment: true
defaultDeployment:
  customDeploymentId: "b75b8b1d-fcbf-4695-9bd3-20f9870df815"
  description: "test-description"
  name: "test-name"

• Our Go SDK has been released as v1.0.0

Updates

• Historical pod restart issue resolved in all ASR models
• Resolved ASR issue with Swedish generating excessive logs for a single failed interaction
• Duplicate word issues with High Definition ASR version 5.X models resolved for grammar based recognition
• Resolved ITN issue with "|" characters appearing in final results
• Resolved ITN issue with unexpected character between product names like “iPhone ? 17” and “iPhone .17”
• ITN issue where email addresses, names and addresses are fully capitalized resolved
• Licensing pod updated to resolve issues if a client inadvertently scales this incorrectly (by nominating a leader and slave pod)
• Issue when performing helm upgrades and configuration service fails to load all configurations for all deployments — causing certain deployments to become unusable — resolved
• Grammar pod memory leak resolved
• Client reported MRCP garbage rule returns “NO-MATCH” resolved
• Resolved issues using built in grammars with es-MX language code
• Recognition of currency amount issues resolved for es-US using the built-in currency (es, es-MX and es-CO grammars impacted)
• Added port level grammar-based stats for MRCP transactions
• Updates made to licensing service and licensing metrics
• Infosec vulnerability — out of bounds error resolved for the simple mrcp client
• Infosec vulnerability — gnupg2 vulnerability in ubuntu:24 resolved
• Infosec vulnerability — MongoDB updated to v8.0.17 as a fix for CVE-2025-14847. Customers advised to update to 7.0.29 or 8.0.17.
• Resolved AMD false “beep” detected when no beep exists in the audio
• Admin portal button added to copy the deployment id in the deployment details page
• Admin portal column sorting issues resolved
• Admin portal issues with importing deployments resolved
• Admin & Deployment Portals and Management API refactored to improve deployment & deployment meta data handling
• Admin & Deployment Portal pagination issues resolved
• TTS issue when adding punctuation after the </say-as> causing the synthesis to be incorrect resolved
• TTS Synthesizer failure -1 error with inline Spanish SSML resolved
• TTS SSML say-as interpret-as="vxml:currency" not working as expected resolved
• TTS SSML say-as interpret-as="address" not working as expected resolved
• Improvements made to Brazilian Portuguese TTS voices
• Resolved issues with log messages being printed — unable to create licensing report for AMD/CPA interaction
• New English ASR models provide improved performance with repetition of Alpha numerics

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

If using TTS, we recommend adding 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"
      - name: "aurora"

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

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

Key installation guide changes:

Capacity recommends a minimum of Kubernetes version 1.33.

The following versions of required software are supported:

• PostgreSQL: 17.5 (recommended)
• MongoDB: 8.2 (minimum) — alternatively 7.0.29 or 8.0.17
• Redis: 8.2 (minimum)
• RabbitMQ: 4.1.8 (recommended)
• NGINX: 1.14.1 (Please note that NGINX are deprecating their ingress support; Capacity are investigating alternatives)
• Linkerd: edge 25.8.4

See the External Dependency Support Matrix article for more details.

Upgrade procedures

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

If you are performing an upgrade, you need to ensure that your NGINX versions are updated to 1.14.1

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

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

Updated API guide

API information for all speech products can be found in our API Documentation

Model versions as part of the release

• ASR updated to remove dependencies flagged in vulnerability scans.
  • All existing ASR models repackaged as 7.0. Previous ASR models are not compatible with Capacity Private Cloud version 7.0.0
  • Higher definition models are tagged as hidef (for higher definition accuracy models — speak to Capacity before you use these as they will require additional ASR pod scaling)
  • New High-Definition ASR models released for English, Spanish, French, Italian, Japanese and Portuguese.
• TTS — Version 8.X (Neural TTS) sample rate 16 — can be down sampled to 8kHz. Legacy TTS models will still run under version 1.0. Further voice enhancements are currently being released in TTS voices version 8.1.0 so Capacity recommends that clients cater for 8.X.
  • New TTS voices released for Catalan, Galician and Valencian.
  • 19 new en-US voices including Caspian and Aurora released under TTS version 8.1.X
• VB — 2.1.15
• VB incorporates Selene 2.4.3 which was integrated into the Container stack

Model version changes

• ASR acoustic & language models were released using a 7 major version designation, including the new hidef variants
• All TTS voices were released using an 8 major version designation