Feature Requests

For the audio effect 'Ducking' it would be essential to be able to put in decimal numbers. Currently, only integers can be put in as a parameter.

I am using minimax for audio dubbing and I am getting very good results and that will improve descript as well.

Right now, I have to add double spaces (i.e., "Hello world, my name is Descript.) between words to get the "effect", but with the amount of videos I'm making, it gets cumbersome.

It appears to be difficult to find functional solutions for this (transliteration but also pronounciation and text-to-speech generation). We knew, that a directive dating back some years existed, to explore and envolve these localizations into services for Speech recognition and Speech synthesis. Thankful for any contactation on these issues. Many thanks in advance ; be well, all good wishes to you.

Some of the default lengths for transitions are not what I would choose. For example I think .08 seconds is usually a bit too long for my crossfades. I'd like to set the default to .05. Ideally, I'd like to set different defaults for different types of transitions, or at least have newly added transitions use my most recently set length so I don't have to change it every time I add one.

Gain is lowered right at the beginning of recording and changes after. Better if volume adjusted consistently

Please add the ability to adjust the angle of a video in 0.5 or 1.0-degree increments. It would be helpful to be able to change the rotation rather than just flipping the video between 90, 180, 270, or 360 degrees

The transcription function is pretty frustratingly inaccurate, especially when there's ElevenLabs Speech-to-text and OpenAI Whisper that do transcription so much better. Is there an option where we choose different models for transcription (even using our own API keys from those platforms)? My use can is transcribing long-form meetings, and the descript tool is the best for editing those meetings for clarity (snipping out tangents and removing filler words), but I'll find that I'll then need to export the audio file, and then take it to something like ElevenLabs for a more accurate transcription. Thank you for your consideration.

After the update, the image generation function is significantly less intelligent than before, can only generate extremely simple text and lines, and the layout and color are extremely monotone.