DeepVA changelog march 2025

Changelog March 2025: Audio Enhancement, Content Moder­ation and more Languages

We have updated our composite AI platform DeepVA and are excited to introduce the latest improve­ments and two new features. This update enhances usability, expands language support, and strengthens the platform’s overall flexi­bility and relia­bility.

Deep Media Analyzer: New Features & Enhance­ments

AI-based Audio Enhancement by AI Coustics

The Audio Enhancement Module is a new addition to the Deep Media Analyzer, designed to signif­i­cantly improve the quality of audio recordings. By utilizing AI-driven noise reduction and frequency restoration, it enhances speech clarity, making spoken content more intel­li­gible and natural.

This module is partic­u­larly useful for processing low-quality audio sources, removing background noise, and restoring sound balance.

The module offers two specialized model archi­tec­tures: 

  • FINCH

    Focuses on noise suppression by elimi­nating background sounds and reducing unwanted audio artifacts. This model is ideal for scenarios like podcasts, inter­views, and field recordings where a clean audio output is essential.

  • LARK

    Aims at frequency restoration and enhancing audio clarity by recov­ering missing frequencies, thereby adding depth and richness to the sound. It’s partic­u­larly suitable for music remas­tering, restoring archival materials, and improving original audio tracks.

This innovation is developed by our partner AI-Coustics, demon­strating the strength of collab­o­ra­tions between start-ups, research insti­tu­tions, and our own devel­opment team. By integrating solutions from specialized AI partners, we help bring cutting-edge technology into production workflows, ensuring that smart innova­tions transition from research to real-world appli­ca­tions.

Looking ahead, this module will also play a key role in increasing transcription quality for bad audio recordings, by enhancing audio clarity this leads to more accurate speech-to-text results, reducing errors in automated transcription workflows.

New Feature — Content Moder­ation

The Content Moder­ation Module in the Deep Media Analyzer intro­duces an automated way to analyze and rate video content by segmenting individual shots and assigning ratings based on the ESRB Content Descriptors. This AI-powered feature evaluates visual elements in a video or image and catego­rizing content into main rating categories such as Violence, Nudity, and Substance Use, along with detailed second-level descriptors like Mild Violence or Drug Reference.

This function­ality is partic­u­larly useful for automating rating proce­dures for large content libraries and providing rating recom­men­da­tions for human reviewers. By pre-processing vast amounts of video content, it helps broad­casters, streaming platforms, and media archives ensure compliance with content guide­lines while saving time and reducing manual workload.

Additionally, this module enhances consis­tency in content classi­fi­cation and can be integrated into workflow automation.

Improved User Interface

Many small improve­ments in terms of usability were made:

  • We improved the timeline chart in our results window in order to give a better visibilty and differ­en­ci­ation between the single lines of the chart.
  • The Threat Detection Module received a vertical timeline chart.
  • We added a copy ID function for transcripts, making navigation and refer­encing easier.

Bug Fixes: More Stability and Consis­tency

  • Model Parameter Fix: Resolved an issue where parameters were not correctly restored when rerunning a job.

Deep Live Hub: New Features & Improve­ments

More languages

Our Deep Live Hub now offers even more possi­bil­ities and supports five new languages: Persian (fa), Irish (ga), Hebrew (he), Maltese (mt) and Cantonese (yue). Whether you’re looking to facil­itate inter­na­tional events or make live content acces­sible to a wider audience, this extension will break down language barriers even more effec­tively.

Improved Help & Navigation

The live editor now supports keyboard shortcuts to make live subti­tling faster and more intuitive. These are also displayed via the infobox when hovering via a button.
The following hotkeys are now usable:

Function Hotkey
First paragraph
Alt + Up Arrow
Second paragraph
Alt + Left Arrow
Highlighted paragraph
Shift
Next paragraph
Alt + Right Arrow
Last paragraph
Alt + Down Arrow
Pause auto-scroll
Ctrl + Spacebar

This makes it easier to edit longer transcrip­tions with less mouse inter­action and keeps the focus on the keyboard and text. In addition, we have integrated a help icon directly into the user interface that provides quick access to support materials.

Improved Help & Navigation

  1. SRT Formatting: The formatting of SRT files has been corrected, and SRT files can now be manually deleted from past jobs, in addition to the previous automatic deletion.
  2. Language List: The drop-down list is now sorted correctly and contains the new languages.
  3. Live Viewer Improve­ments: Fixed the issue of non-erasable Live Viewer entries.
  4. Streaming relia­bility: Fixed time zone differ­ences in the streaming job table.
  5. Timer bug fix: Fixed a recurring issue where the timer reset unexpectedly.

In the coming months, we’ll be preparing new AI-powered workflows for transcription, export creation, and metadata gener­ation to further optimize your media processing.

All DeepVA changelog updates are available here: https://docs.deepva.com/changelog/

Share

Email
LinkedIn
Facebook
Twitter
Search

Table of Contents

latest AI news

Subscribe to our newsletter

Don’t worry, we reserve our newsletter for important news, so we only send a few updates once in a while. No spam!