YouSpeak is an app that allows the user to use a local AI model to generate clones of voices from Youtube videos.
This was a really great project and I feel like the end result is one that is functional and actually useful. This was my first time actually developing and releasing an app and I learnt a lot.
This will be the final devlog. I have managed to package the app with pyinstaller and add it to a git release. I have installed it on a different computer and it worked perfectly. If anyone does want to try it, please just be aware that the model takes a really long time to load for the first time. After that it should be saved and it will be much quicker.
About to ship the project. Really happy with how it came out. Since the last log I have mostly improved existing features and fixed bugs. Fixed the shorten audio function which wasn’t working and forcing it to process the whole file each time it generated an output. Added the option to select an output folder, it was previously saving to the project folder. All features are working now.
@James15
on
Scottish Assistant
Took a while but I finally got this project ready to ship! Had a lot of fun making this and learnt a lot. Although I didn’t have much time to dedicate to it, I’ve really enjoyed this program and I hope they continue with these events in future.
@James15
on
Scottish Assistant
2h 39m 42s logged
About to ship the project. if anyone wants to try it please set it upon your computer. Because I have an AMD GPU I haven’t been able to set the LLM up to run on GPU in Docker. You will be able to see the UI and interact with it but you can’t get responses without setting it up locally. I am using Gemma4 but other lightweight LLMs that can accept the context to turn them Scottish should work and you can adjust it easily in the code. This has been a really great project and I’ve learnt so much. Thanks for reading!
Added new ways to add audio samples. New status indicators for processes and slightly different UI layout. You can now add audio by just pasting a youtube link and use the shorten option to cut it down to the first 10 seconds so that it processes faster. You can also add both .mp3 and .wav files.
Finished first iteration of app.
Turns text into speech that is a clone of a voice of the user’s choosing. Very basic interface that allows users to add new voices or choose from a list of previous ones and then input text and have the output generated and added to the project folder. This should allow other people to use the app if they just install the Qwen-TTS model.
I want to improve the interface and add the ability to extract audio from Youtube videos.
@James15
on
Scottish Assistant
3h 5m 20s logged
Fully functional! :D Will need to work on the look of the UI.
@James15
on
Scottish Assistant
4h 46m 11s logged
New text to speech with voice cloning and google search to get more accurate results
@James15
on
Scottish Assistant
1h 2m 56s logged
Created a python script to access a basic LLM ( Gemma4) running with Ollama. Added context to make it Scottish. Trying now to add text-to-speech to have the output read out. It responds quite slowly but is running locally on my computer.
