I do not know how to program an AI transcription engine, and that is all I utilize Descript for. First of all, I must add that I have tried other engines available, and only one gets close enough to the performance you all provide - yet, the complexity, required training, required hardware/software, and expense is relatively unreasonable compared to what you offer. The only feature they offer is a comprehensive (yet hard to train) editing tool, which with care, experience, and lots of customization, I have nearly replicated just with MS Word. On the positive side, which is all I have today (though I will try to come up with something to request...), whatever the Descript team has done to the transcription engine lately is SUPERB. It is so good that I have to go to my editing macros and readjust them, because the usual corrections they seek no longer exist. Surprisingly, I see phrases being quoted, and in a 600-page document, with maybe 50 quoted phrases and sentences provided by the AI, only about seven of them were misquoted - and it was mostly because the speakers broke the sentence in ways that the AI could not reconcile. I was surprised and delighted - I did not think this was possible. Additionally, names of weird places that I usually have to hazard a guess on a search engine are easily found and properly spelled by Descript a lot better than before. As what I transcribe is legal, I found even case law properly transcribed - a new one for any AI I have worked with; it is not all-encompassing and nowhere near all correct, but positively surprising nonetheless. This is where the AI gets it wrong or not at all: days and years. For some reason, twenty twenty six does not translate as 2026; one May twenty four does not translate as 1 May 24; et cetera. I have macros for it, but thought you would like to know. That one seems easier to fix at a glance than proper quotations! There are many other numerical issues related to translation, but dates seem like the more standardized I can think of that would benefit the most people. I have also noticed positive differences in the capitalization used by Descript; it catches things that I occasionally forget to capitalize. Military acronyms are getting better formatted as well, which is a very huge surprise... (if you all need military court transcripts and examples, I could get you sanitized versions for your AI dictionary and learning model, as well as lists of weird terms and how they are spoken - I used to teach this material). So, good feedback thus far, and thanks for all the work you all do. Oh, I remember trying out the feature that will replace certain things in the translated text for whatever we gave it... I have no idea if it has been worked on since I stopped using it, but when I tried it (only when it first came out), I immediately saw the value in it but that it needed lots of work. I do not know if you all do beta testing of features with enterprising users that can give you detailed feedback to help you, but I volunteer as long as time allows. :-) Thank you all for the work you do. Thanks to Descript, my productivity is at least 250% more AND better than before using this AI tool, which allows me to work fewer hours for the same productivity, or increase my product (but being a 100% disabled vet that needs constant breaks, I need tools that help me increase hourly productivity, since working more hours is occasionally impossible due to health./physical issues). Again, thanks!! ( I would NOT MIND military discount, even if only on the annual pricing... just saying... LoL)
