Late December I have an idea for a kinda-funny Web Art project I could make. It would require me to be able to "tokenize" a 4 hour video by word, that is, I'd need timestamps of the beginning of each word.
I assume this means using a text-to-speech library.
However, I don't want to use software that runs not-on-my-computer, or which is made by methods I consider unethical. This means using only old discontinued libraries, since all new ones are cloud-based. (1/3)
In January, I try using "PocketSphinx". This is an old, discontinued speech-recognition library by CMU. It runs for six hours and forty minutes and produces extremely low quality output. https://cohost.org/mcc/post/4147939-hbomberguy I decide it's not good enough for what I'm doing.
Today, I come back to this, and try running "DeepSpeech", an old, discontinued library by Mozilla. After much DLL hell, I get it to run. It runs for seventeen minutes and produces no output. None at all. (2/3)
I wonder which is the worse possibility… that NVidia Omnisphere can't install these libraries for me, or that NVidia Omniworld can install the libraries and I'm over here fighting with the website anyway
CUDA 10 installer is offhandedly mentioning that "overwrites the current display driver". Cannot wait to find out what horrible dll hell consequences I'm setting myself up for in about a year by clicking "OK"
@filenine In the final pane there was a tiny note hidden at the bottom of a long scroll indicating it refused to install the display driver because there was already a newer one installed. Which, good, but why'd they have to scare me like that :(
It said it couldn't find "cudart64_101.dll" so I I installed CUDA 10.2 and that didn't work so I realized "_101" might mean 10.1 so I uninstalled it and installed CUDA 10.1 and it still didn't work and then I closed my cmd.exe window and opened a new one and now it works, and I wonder if CUDA 10.2 would have worked if I'd stuck with that one. On the other hand, CUDA 10.2's installer claimed it had failed to install about 50% of its components and 10.1 gave no errors, so… I know nothing.
@alescratches head I'm on Windows. I thought Docker was Linux. Can a Linux docker image running on a Windows machine actually make use of the video card?
Note: My Python version/packaging problems from earlier actually got worse after my last post on the subject, I just decided to stop posting updates because it was making me too sad
@ohmrun I ran into this problem and I found the solution was to use "pipx" which is a special kind of pip that installs the packages in a weird way that works around Ubuntu's pip hobbling, but also it does not require you to set up a venv. It just works like pip.
Thinking about how Mozilla is "pivoting to AI" but DeepSpeech, one of the very few "AI" products you could possibly find a positive use for (pure-local speech recognition), is not only a Mozilla product but so abandoned that you actually have to downgrade to Python 3.9 to run it