Virtual Appliance Changelog

3.2.0 (2024-08-20)

Added

• Add language-identification technology
• Enable GPU sharing

Changed

• Deploy nvidia-device-plugin from helm chart
• Deploy nvidia-device-plugin to nvidia-device-plugin namespace instead of gpu namespace

Removed

• Do not include models for speaker-identification and enhanced-speech-to-text-built-on-whisper by default on data disk

3.1.0 (2024-07-30)

Added

• Deploy billing-related components
• Loading the licenses from secrets instead of values file
• Reloader which reloads the deployment every time its secret is edited
• Install yq
• Support running voiceprint-extraction on GPU

Changed

• Edit the way how secrets are loaded in Speech-to-Text-Phonexia and Time-Analysis values
• Add comment explaining why disabling media-conversion is usually not a good idea
• Update the Nvidia drivers to version 550
• Build in AWS instead of local infrastructure
• Use cloud-init for configuration
• Root (/) partition is merged with var partition (/var)

3.0.0 (2024-05-15)

Added

• Enable to configure parallelism in speech-to-text-phonexia and time-analysis

Changed

• Rename speech-to-text-whisper-enhanced to enhanced-speech-to-text-built-on-whisper
• Upgrade Rocky Linux version to 9.4

Fixed

• Fix language code for Levantine Arabic

2.1.0 (2024-04-18)

Added

• Introduce extra environment variables for speech-to-text-whisper-enhanced
• Introduce max upload file size specification for filebrowser

Changed

• Increase inotify limits
• Move prometheus storage to data disk
• Admin-related backends are extracted to separate ingressAdmin configuration
• Time-analysis reconfigured as onDemand instance by default
• Speech engine images bumped from 3.60.1 to 3.61.0
• Load models for speech-to-text-phonexia and time-analysis from data disk by default
• Default value for maxFileSize UI limits set to 5MB instead of 5MiB
• Time-analysis is started as onDemand instance by default

Fixed

• Blacklist nouveau driver

2.0.0 (2024-03-29)

Added

• Allow to change api log level
• Allow to configure UI limits
• Add time-analysis subchart
• Add options to change tmpdir volume for speech-platform API and media-conversion

Changed

• (Breaking change) Rename speech-engine subchart to speech-to-text-phonexia subchart
• Store models on data disk for each microservice separately
• Use free version of media-conversion by default

Fixed

• Start nvidia-persistenced only when nvidia driver is loaded

1.1.0 (2024-02-19)

Added

• Enable startup probe for speech-to-text-whisper-enhanced
• Automatically start nvidia-persistenced daemon
• Add comments about api storage resources

Changed

• Use app frontend as landing page
• Expose administrator console to /admin uri
• Increase max data disk size to 20GB
• Disable GSP firmware in nvidia drivers

Fixed

• Calculate speech-to-text-whisper-enhanced capacity on GPU properly

Removed

• Remove swap partition

1.0.0

• Initial release