@mjdxp I feel a deep ambivalence toward this topic. modern machine learning systems capable of media-to-text transcription are a big step up from the frankly miserable quality of automated transcription we had before, and there's an additional degree of agency which can be granted through those capabilities, but ultimately it still leaves disabled people relying on unpredictable and unreliable systems in the face of sheer apathy from human creators who could and should be captioning their media.
@mjdxp ironically one of the best usages of modern ML ASR systems is for forced alignment, yet that seems to be one of the least talked-about use-cases.
The MP4 Format does allow for subtitles, https://en.wikipedia.org/wiki/MPEG-4_Part_17
@mjdxp We helped the tech companies do that on this network by attacking anyone who used a machine learning model for any reason.
I am convinced the culture of fedi alone prevented many people who could use machine learning applications for reasonable and decent use cases from doing so.