At DeepVA, we are constantly striving to improve our software and services to provide the best possible experience for our users. We are excited to share the latest updates and improvements released in the May 2024 update. Our latest changes include a mix of new features, improvements, and fixes aimed at increasing functionality and user satisfaction.
Transcript Editor
Our most important update is the new Transcript Editor, which enables seamless editing and management of transcripts from Deep Media Analyser’s Speech Recognition. This feature was developed to provide more control and flexibility in handling speech-to-text output.
The transcript editor is started via the new button in the speech recognition results window. The passages can be adapted manually section by section along the timecode, while a player is available on the right-hand side to listen to the passages again in case of doubt. In order to provide better assistance under time pressure, ‘Word Confidence Colouring’ can be used to display word by word how confident the system is with which words. This makes it easier to recognise which words may need to be checked again.
Once you have finished editing, you can start the export to common formats directly from the editor — all in one interface. The function will be available to all users in a first version with the current update.
Aiconix Live
A lot is also happening with the Deep Live Hub, formerly Aiconix Live:
With the updates of the last weeks, we can now enable seamless transcription, which allows us to replace the workers during important system updates without interrupting the streams and thus the transcription. This not only enables seamless and uninterrupted transcription and translation but will also make the system much more flexible in the future regarding quick changes or adjustments during a stream.
The new user interface has also been online since last week, which not only enables better control over batch processing and live streams, but also displays thumbnails of current and past streams.
Further innovations in the area of the viewer and our own ATT protocol are planned for the coming week: Partials will be available in quasi-real-time via both channels. This means that words already recognised in the transcription will be delivered immediately. However, it is also possible to replace them directly with potential corrections in order to be even faster for fellow readers. For the time being, however, the partials will only be available for transcriptions — more details on this soon.
The Deep Live Hub will also benefit greatly in its development in the coming months due to the increasing number of users and their feedback on the various broadcasts and live events that are produced using it.
Further innovations
Speaker identification indexing: Since March, our speaker identification has included indexing capabilities that allow users to ‘fingerprint’ the speaker’s voice and index it using a number. In future, this will also make it possible to cluster speakers and label them retrospectively.
Cancel function for ongoing uploads — we have added a new function in the storage uploader dialogue box that allows users to cancel ongoing uploads. This improvement aims to provide a smoother and more engaging user experience.
Enhanced facial recognition capabilities: Release of the ‘celebrities v29’ model for our facial recognition mining module, which includes over 100 new identities in total and is also available for immediate use.
Important corrections & bug fixes
UI and usability improvements: We have fixed various minor UI bugs in the software, including in the storage, on the indexing overview and in the data set clusters.
We also solved a problem when re-running jobs: From now on, the correct model and data set will be preselected when restarting, a small improvement with high added value for processes.
Back in February, an error with data set exports was fixed that caused collisions of class name folders. This now ensures better data management.
All DeepVA changelog updates are available here: https://docs.deepva.com/changelog/
