LocalWhisper •
LocalWhisper has used two interfaces. The first was a Python and Flet desktop app; I later built a Telegram bot for exchanging audio and text.
The challenge
I needed transcription, text review, and speech synthesis to run on my computer without freezing the interface. I could not build the experience I wanted with Flet.
Decisions and implementation
I chose Faster-Whisper, Llama 3.2 3B, and Kokoro TTS to fit my computer's resources. I separated recording, transcription, review, and speech playback. In a later iteration, I moved audio and text exchange to a Telegram bot.
Result
In the desktop version, I connected the three models in a working flow. The linked repository contains that implementation. I later built a Telegram bot for sending and receiving audio and text; the portfolio does not show that version yet.
What I would improve next
I would update the demo and repository to show the Telegram bot, then explain which parts of its processing run on my computer.
