Hi, i wanna transplant the api mechanism + file splitting mechanism out and move to a python (kivy?) android app, later.

[ccchan234]

hi, buzz is good that i throw large files to it, it auto split the file by ffmpeg, send to api and merge the .srt.

i myself have wrote some python script that do similar but buzz's mechanism is more robust.

next month when i have more time, i would like to ask cursor to try just to grab this mechanism out and move to a android python GUI, likely using kivy.

if any one interested, even a vibe coder like me, is wellcome to join.

% of success is great, as i already have some python script that do similar; and cursor+AI have helped me write some android kivy pythons ( i just didn't compile them into .apk).

let's see.

[raivisdejus]

Good luck with your experiment. Feel free to share the results :)

[ccchan234]

Good luck with your experiment. Feel free to share the results :)

hey, i just remember that buzz itself is also write in python.
can we simply run it on android? thanks

[raivisdejus]

Buzz (the logic of it) is in python, that part would work on Android. UI elements are in Qt, quick glance seems to indicate that Qt also has Android version. The speech recognition is done with engines like whisper.cpp that also has Android support.

So there seem top be no unsolvable limitations for this and with some work in how the compilation and packaging should be possible. Not necessarily easy, but possible

[ccchan234]

hi, cursor 2 now seems more smooth.

i made a 7 step plan to implement it.

phase -1 and phase 0 successed.

here are the phases:
buzz-cut Development Phases (Simplified)
Phase -1: Proxy + API Validation (~2-3h) - Verify proxy & Groq API work on both Windows + Android ✅
Phase 0: CLI Proof of Concept (~2-3h) - Simple script to transcribe small MP3s (< 25MB), text-only output ✅
Phase 1: Core Engine + SRT Export (~9-11h) - Full API client, FFmpeg integration, generate SRT subtitle files
Phase 2: Large File Chunking (~8-10h) - Handle files > 25MB with smart audio splitting and timestamp merging
Phase 3: Basic Mobile UI (~10-13h) - KivyMD interface, file picker, progress bar, cross-platform
Phase 4: UI Polish (~12-15h) - Multi-screen navigation, settings persistence, Material Design theme
Phase 5: Android Native Integration (~14-18h) - Android file picker (SAF), permissions, APK packaging
Phase 6: Optional Features (~varies) - Batch processing, history, multiple export formats, etc.
Total: ~60-73 hours (baseline to Phase 5)
Current Progress: Phase 0 complete ✅ (Windows + Android validated)

i was born in evil axis country so i have to use proxy.
will somehow share it after finish.

i am shocked that i could not found an android app, that take in the whisper API key, and transcribe a .mp3 file.

wtf.

[raivisdejus]

Keep us posted. I would like to see an Android app that can transcribe some audio.
Good luck!

[ccchan234]

oh NO,
the script can now transcribe .mp3 < 25MB.

but to chunk it; it need ffmpeg.

however

i am not sure how a script inside pydroid3 could reach the ffmpeg inside termux, as android apps are isolated in different sandbox.
in pydroid 3 there are some packages like ffmpeg (which only have codes or some files but no binary) and ffmpegbin (do have a binary like file but not directly executable, seems).
so for me i can only reach this phase.
vibe coding is good, but android is crazy bad hard. ppl already complaining it's file system (SAF) is like running an arm race with google.

so could be used as a script for .mp3 < 25MB, but not over.
will tidy up later.

[ccchan234]

honeslty android is a fxxking weird monster, i dont know why it sucess.
https://stackoverflow.com/questions/74908940/how-do-i-use-ffmpeg-library-in-python-on-android-with-pydroid
at this moment, ios have perfect backup solutions, and android still dont.

[ccchan234]

ok, now using termux, every function worked. the .mp3 could be chunked into custom size e.g. a 30MB file chunked into 30x1MB successfully sent to groq and turned into .srt and .txt.

but termux dont support GUI well.

may be just make a web GUI interface or put aside... sigh.

[ccchan234]

ok, now it's termux + python + ffmpeg;
with web server "flask";
then use browser for GUI:

let me tidy up the code and see if someone interested.

now, on the go, i could bookmark utube/bili videos, then just dnld as audio;
transcribe. ask LLMs to summary. only really dnld+watch USEFUL videos -- used up too much time.