office-transcribe

When to use

Meeting recordings, voice memos, interviews,.mp3 .wav .m4a to text, need SRT subtitles.

Process

Place audio in the workspace (or use a user-provided path)
run Run the transcription script, output .txt / .srt / .json
Long audio: by office-meeting generate minutes; short memos: 5-point summary

Recommended: faster-whisper (local)

from faster_whisper import WhisperModel
model = WhisperModel("large-v3", device="cpu", compute_type="int8")
segments, info = model.transcribe("meeting.mp3", language="zh", vad_filter=True)
lines = []
for seg in segments:
    lines.append(f"[{seg.start:.1f}s-{seg.end:.1f}s] {seg.text.strip()}")
open("transcript.txt", "w", encoding="utf-8").write("\n".join(lines))

pip install faster-whisper
# 可选 GPU：device=cuda

Speaker segmentation

Whisper has no native diarization; when additional models such as pyannote are needed, first explain the environment and authorization.

SRT subtitles

def fmt(t):
    h, r = divmod(int(t), 3600); m, s = divmod(r, 60); ms = int((t % 1) * 1000)
    return f"{h:02d}:{m:02d}:{s:02d},{ms:03d}"
# 每 segment 写一条 SRT 块

Prohibited

Do not claim transcription is complete unless the output file exists
For sensitive recordings, remind the user to redact before writing to memory/

Missing environment

No faster-whisper / no run → ask the user to provide a text transcript or external transcription result; see office-env-setup