Verbatim speech-to-text with word timestamps on a 4 GB GPU: CrisperWhisper 2.0 Turbo (CT2 int8) + wav2vec2 CTC alignment. Turkish-first, Linux/WSL2.
Controllable Transcription. Verbatim ( every, filler, pause, stutter, vocal sound) , or intended ( what the speaker meant to say, optimized for readability) with word-level timestamps.
Word and phoneme level speech timing for spontanoues speech. Fast, accurate and robust.