#!/usr/bin/env python3 """Self-hosted replacement for OpenAI whisper-1 file transcription, using faster-whisper (MIT). Writes plain text and an .srt file with segment timestamps (whisper-1's srt/vtt use case). Usage: python faster_whisper_transcribe.py audio.flac [model_size] # model_size default: small pip install faster-whisper (CPU works; a GPU is much faster for long files)""" import sys from faster_whisper import WhisperModel def ts(sec): ms = int(round(sec * 1000)); h, ms = divmod(ms, 3600000); m, ms = divmod(ms, 60000); s, ms = divmod(ms, 1000) return f"{h:02d}:{m:02d}:{s:02d},{ms:03d}" audio = sys.argv[1] size = sys.argv[2] if len(sys.argv) > 2 else "small" model = WhisperModel(size, device="cpu", compute_type="int8") segments, info = model.transcribe(audio, vad_filter=True) print(f"language={info.language} p={info.language_probability:.2f} duration={info.duration:.1f}s") base = audio.rsplit(".", 1)[0] text, srt = [], [] for i, seg in enumerate(segments, 1): text.append(seg.text.strip()) srt.append(f"{i}\n{ts(seg.start)} --> {ts(seg.end)}\n{seg.text.strip()}\n") open(base + ".txt", "w").write(" ".join(text) + "\n") open(base + ".srt", "w").write("\n".join(srt)) print(" ".join(text)) print("wrote", base + ".txt", "and", base + ".srt")