2026

transcription tools are either expensive or require beefy hardware. whisper.cpp exists but setup is painful — cmake, CUDA paths, dependency hell.

google colab gives you a free T4 GPU. faster-whisper gives you the fastest whisper implementation out there. why not combine them?

wanted something dead simple: upload audio → run one cell → download .srt files.

• → runs faster-whisper large-v3-turbo on colab free tier T4 GPU
• → batch processes .m4a, .mp3, .wav, .flac, .aac
• → auto-generates .srt subtitle files per audio
• → one-click zip download for all subtitles
• → RTF ~0.23 on free-tier hardware

two tools, one workflow:

gcolab-whisper → audio in, .srt out
srtplayer     → audio + .srt, synced playback

transcribe on colab. play back locally. no cloud uploads, no API keys, no subscriptions.

4 cells in a colab notebook:

• 1. wget — download your audio file
• 2. pip install — faster-whisper
• 3. transcribe — model load, batch process, save .srt
• 4. zip — bundle all subtitles for download

you don't need a $3000 rig to do AI transcription. you don't need to pay for APIs. you don't need to compile whisper.cpp from source.

you need a browser, a google account, and 5 minutes.

• → SYSTRAN/faster-whisper — CTranslate2 implementation of whisper
• → openai/whisper — original model weights
• → Google Colab — free T4 GPU tier

github repo →

open in colab →

srtplayer → // play back your .srt files