this post was submitted on 28 Jun 2026
27 points (100.0% liked)
Books
594 readers
4 users here now
For all books - fiction and non-fiction.
founded 4 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Check out my speech to text suite, if you're not using tools it will speedrun your efforts for your friends. https://codeberg.org/CritBase111/speech-to-text
If you have yt-dlp installed, you can just do:
This will pull the video (not just from YouTube), transcribe to a txt and SRT files (subtitles), and translate to Spanish, all locally. Translation has some varying results.
If you have a lot of videos you can pre download them, then run stt with a --batch command on the entire folder containing all your videos.
Or there is also a simple GUI that you launch with 'stt gui'
Unfortunately codeberg had a server outage and may not be entirely back yet so I'm not sure if its accessible right now. If it's not back let me know and I'll upload the files somewhere else temporarily.
my mama taught me to not click links from strangers but I will check it out, securely....in a secure way...
EDIT: ALSO HELL YEAH, pip <3
pls pls dont fry my brain, tears are running down my face rn what WHAT DO I DO HOW
run the install command for your OS in a terminal window! They are located here: https://codeberg.org/CritBase111/speech-to-text#quick-install. So if you are on Windows for example it's the iwr command that's followed by a link.
After you have installed stt correctly (the installer will tell you "stt installed, Open a NEW terminal window and type: stt --help"), exit the terminal window and open a new one. In the new windowe, run 'stt gui' from your terminal (powershell if on windows). It contains all the options the tool offers. By 'run from the terminal' I mean literally just open a terminal window and type the command + enter). It opens a window that looks like this:
So for example, since you are translating videos, you can paste the youtube URL to the video in the input field, then change the Target field in 'Model & Language' to the language you are translating to. If you scroll down in the graphical interface, you will see output formats, and you can select 'txt' to make a simple text file, or also check 'srt' for a subtitles file - a subtitles file can be immediately loaded into the video.
Then press the Transcribe button in the bottom left. It can take a while, but it will work. It will save the files on your computer, transcribed and translated!
It's still a few instructions but I tried to make the app as user-friendly as I could π the hardest part is installing it - when it's running it can't destroy anything on your computer, at most it creates files (the transcriptions) that you can delete.
if you want to use the terminal version, do 'stt transcribe --help' and it will show you the commands you can use. But the graphical interface is the exact same as the terminal version, you're not missing out on anything if you use the interface.
π₯Ί π₯Ί π₯Ί π₯Ί π₯Ί ths is like quantum physics to me, but i will try my best to do as instructed
which operating system are you on? I can provide more specific instructions.
The difficult part is installing the program, using it will come naturally once you've done it once or twice.
uuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuh i believe it is windows 11
Okay, so, I had to look at the windows flow a bit more in-depth. Winget was a good tool but it's actually unmaintained and people have been reporting some possible malicious packages. So I redid the entire install script to not use winget anymore. It's much easier to install stuff on Linux, there the install script would have just done everything for you π but no worries, it is what it is.
What you will need:
Then you can run the install script provided on my codeberg repo. In Powershell (not cmd):
iwr https://codeberg.org/CritBase111/speech-to-text/raw/branch/main/install.ps1 | iexThe install script is very simple; it just sets up a folder called stt in C:\Users\YourUsername. Everything that STT sets up by itself stays within that stt\ folder. When stt is installed correctly, the installer script will tell you with the words "stt installed. Open a NEW terminal window and type: stt --help"
Open a new cmd or powershell, and simply run the command 'stt gui'. It will open the user interface in a new window, which is probably easier to use. I find that when I was on Windows I preferred user interfaces, and on Linux I prefer command-line tools lol.
The first time you actually start transcribing or translating a video, stt will automatically download the models. This happens automatically, you just have to wait. But it means the first transcription will take a bit more time to finish.
If you ever want to uninstall stt from your computer, run this in PowerShell:
ffmpeg, yt-dlp and pandoc are pretty nice tools to have on any computer. ffmpeg can convert and compress any video file. yt-dlp allows you to download videos from any website by just running 'yt-dlp https://link.com/' in the CMD, and pandoc can convert any text format to any other text format, so you can convert . So they are dependencies, but they are generally good tools to have. Python as well is pretty nice to have as a lot of open-source software uses python. You only install it once then most open-source software is available to you.
π₯Ί π₯Ί π₯Ί π₯Ί π₯Ί π₯Ί π₯Ί π₯Ί i have python and i have ytdl...the trouble here is....should i trust u or not π
Why not WSL?
wsl --installI can't recall if there's any set-up besides this, but I think that using WSL is better than rawdogging EXEs in Windows.
That's actually very smart, for people who aren't afraid of using WSL. Then you would just
sudo apt installthe dependencies (and I assume since it's an ubuntu subsystem it already has python 3.12), and then run the installer for Linux through WSL:curl -fsSL https://codeberg.org/CritBase111/speech-to-text/raw/branch/main/install.sh | bashI also heard that your C:\ drive mounts in the WSL through /mnt/c/, so you can pass any audio file to stt by simply specifiying 'stt transcribe "/mnt/c/Documents/my_video.mp4"'
but I assume you will have to run stt through the WSL cli from then on?
Yes it would be the CLI. Thing is that it's simpler since the Linux filesystem gets exposed in the File Manager (it's literally a "Linux" tab next to "This computer")
So, you can just drag and drop a file from the Videos folder in Windows into Linux and vice-versa.
So you could drag the video into Linux, run
stt transcribe /home/user/my_video.mp4and then grab it back to Windows Videos folder (or just use the Linux folder since it's treated as a regular directory by Windows)It's easier for a beginner than messing with mounting and
mv. It's what I do in my work computer whenever I need to do something with yt-dlp.It's very, very convenient. I wonder if WSL is the reason why winget was abandoned
Wait
Wait wait wait...
This does translation from English into Spanish?!
For subtitles in a video, correct?
Hopefully, I am following this convo along well enough.
Anyway, a translation tool, preferably one I don't really have to pay for, is what I want for my YouTube and TikTok channels, and one that can put subtitles I the video and put the subtitles or lines basically where they should be as the video goes along lol
yep, it does everything.
stt transcribe 'file.mp4' --target-lang es,fr,pt,jp --format srtwill transcribe the audio from the video (auto-detects language), then translate the result into their own .srt (subtitles) files. You can chain arguments together such as es,fr,pt,jp -> it will create a different file for each combination. so one .srt for spanish, one .srt for french, etc.stt translate 'file.txt' --source-lang en --target-lang --format txt,srtwill only translate a previously existing text file. you can chain --format (or any argument, including the input files) to output the translation as plain text file, and another .srt fileif you have pandoc and yt-dlp installed, the app integrates with them to provide more input and output formats with pandoc, and yt-dlp to download the video automatically from a web page (so instead of 'file.mp4' you can provide a youtube link). It automatically uses them, you don't have to do anything special.
it also needs python 3.10 or higher and ffmpeg to work. It downloads the models on first use, expect about 5-10 gigabytes of download. Needs I would say around 4GB of Vram at the minimum to work.
translation can be hit or miss in terms of quality, some language pairs work better than others. If it's bad, just open deepseek web, switch to expert, paste the original subtitles, and tell it to 'translate these subtitles to [language], output within three backticks for code formatting'.