wishcribe 1.0.0

Multi-speaker audio/video transcription — Whisper large + pyannote.audio (offline)

wishcribe ✍️

Multi-speaker audio/video transcription — Whisper large + pyannote.audio, fully offline after first run.

[SPEAKER_00] 00:00:01
  Selamat datang di rapat hari ini.

[SPEAKER_01] 00:00:05
  Terima kasih. Mari kita mulai.

[SPEAKER_00] 00:00:10
  Baik, topik pertama adalah anggaran kuartal ini.

---

Installation

pip install wishcribe

ffmpeg is also required (one-time system install):

brew install ffmpeg        # macOS
sudo apt install ffmpeg    # Ubuntu/Debian

---

Quick start

Step 1 — download all models (run once)

wishcribe download --hf-token hf_xxx

This downloads and caches:

• Whisper large (~2.9 GB) → ~/.cache/whisper/large.pt
• pyannote diarization (~1 GB) → ~/.cache/huggingface/hub/...

Output:

📦  WISHCRIBE — MODEL DOWNLOADER
══════════════════════════════════════════
  Whisper model : large
  Diarization   : HuggingFace download (token provided)
══════════════════════════════════════════

📥 Downloading Whisper 'large' model (2.9 GB)...
✅ Whisper 'large' downloaded and cached  (2.9 GB)

📥 Downloading pyannote diarization model (~1 GB)...
✅ Diarization model downloaded and cached

🎉 All models cached! wishcribe now works fully offline.
   Run transcription with:
   wishcribe --video meeting.mp4

Step 2 — transcribe (fully offline, forever)

wishcribe --video meeting.mp4

That's it. No token, no internet, no extra flags.

---

Usage — CLI

Download command

# Download default model (large)
wishcribe download --hf-token hf_xxx

# Download a specific model size
wishcribe download --hf-token hf_xxx --model medium

# Use a local pyannote model folder (no HuggingFace needed)
wishcribe download --model-path /path/to/pyannote-model

Run / transcribe command

# Basic (Whisper large by default)
wishcribe --video meeting.mp4
wishcribe run --video meeting.mp4    # same thing

# With language + speaker count
wishcribe --video meeting.mp4 --bahasa id --speakers 3

# Override Whisper model
wishcribe --video meeting.mp4 --model medium
wishcribe --video meeting.mp4 --model small

# Use OpenAI API for transcription (diarization still offline)
wishcribe --video meeting.mp4 --use-api --api-key sk-xxx

# Custom output folder + save JSON
wishcribe --video meeting.mp4 --output ./results --json

All run options

Argument	Description	Default
--video	Path to video or audio file (required)	—
--hf-token	HuggingFace token — first-time only	—
--model-path	Path to local pyannote model folder	—
--model	tiny/base/small/medium/large	large
--bahasa	Language code e.g. id, en	auto-detect
--speakers	Number of speakers (optional)	auto
--output	Output folder	same as input
--use-api	Use OpenAI Whisper API	False
--api-key	OpenAI API key (with --use-api)	—
--json	Also save .json	False
--no-txt	Skip .txt output	False
--no-srt	Skip .srt output	False

---

Usage — Python

from wishcribe import download, transcribe

# Step 1 — download models once
download(hf_token="hf_xxx")

# Step 2 — transcribe offline
segments = transcribe("meeting.mp4")

# With options
segments = transcribe(
    "meeting.mp4",
    model="large",     # default — best accuracy
    language="id",
    num_speakers=3,
    output_dir="./out",
)

for seg in segments:
    print(f"[{seg.speaker}] {seg.start:.1f}s  {seg.text}")

---

How offline mode works

Cache location	What's stored
~/.cache/whisper/large.pt	Whisper large model weights (2.9 GB)
~/.cache/huggingface/hub/models--pyannote--...	Diarization model (~1 GB)

Once cached, both load instantly from disk — no internet ever needed.

---

Whisper model guide

Model	Size	Speed	Accuracy
tiny	75 MB	Very fast	Fair
base	139 MB	Fast	Good
small	461 MB	Moderate	Better
medium	1.4 GB	Slow	Very good
large	2.9 GB	Slowest	Best ⭐ (default)

---

HuggingFace setup (for download command)

1. Sign up at https://huggingface.co
2. Accept the license: https://huggingface.co/pyannote/speaker-diarization-3.1
3. Create a Read token: https://huggingface.co/settings/tokens

Only needed once for wishcribe download.

---

Output files

File	Description
<n>_transcript.txt	Plain text grouped by speaker
<n>_transcript.srt	SRT subtitles with speaker labels
<n>_transcript.json	Raw JSON array (opt-in)

---

Publishing

make build      # build dist/
make publish    # upload to PyPI → pip install wishcribe

---

License

MIT