AI Meeting Notes App in 2026: faster-whisper (25k Stars) + WhisperX Local Pipeline - Free Transcription with Speaker Labels

Stop paying for Otter.ai. faster-whisper (24,747 stars) transcribes meetings locally, WhisperX (23,436) adds speaker diarization, and an LLM writes the summary and action items - $0 per month.

💡 What You Will Learn

Stop paying for Otter.ai. faster-whisper (24,747 stars) transcribes meetings locally, WhisperX (23,436) adds speaker diarization, and an LLM writes the summary and action items - $0 per month.

📜 Table of Contents

• The short answer
• The pipeline
• What you get
• Real numbers
• FAQ

The short answer

A free local meeting-notes pipeline: faster-whisper (24,747 stars, MIT) transcribes audio 4x faster than original Whisper, WhisperX (23,436 stars, BSD-2) adds word-level timestamps and speaker diarization, and any LLM converts the transcript into minutes, decisions and action items. Everything runs on your laptop.

The pipeline

pip install faster-whisper openai-whisper  # for the CLI

Step 1 - Transcribe with faster-whisper

from faster_whisper import WhisperModel
model = WhisperModel("medium", device="cuda", compute_type="float16")
segments, info = model.transcribe("meeting.m4a", language="en")
text = "".join(s.text for s in segments)

Step 2 - Add speaker labels with WhisperX

import whisperx
model = whisperx.load_model("medium", device="cuda")
audio = whisperx.load_audio("meeting.m4a")
result = model.transcribe(audio)
diarize = whisperx.DiarizationPipeline().assign(speaker_labels=result["segments"])

Step 3 - Generate the minutes with an LLM

prompt = ("From this transcript write: 1) summary 2) decisions 3) action items "
          "with owners. Transcript:
" + transcript)

What you get

• Speaker-labeled transcript with timestamps
• 200-word executive summary
• Decisions list
• Action items with owners (the LLM infers from "I will...", "you should...")

Real numbers

• A 60-minute meeting transcribes in ~10-15 min on a consumer GPU, ~30-40 min on CPU.
• faster-whisper uses CTranslate2 - 4x faster than Whisper (106,660 stars) with same accuracy.
• Paid tools charge $10-25/month; this stack costs $0 plus your laptop's electricity.

FAQ

Q: How accurate is speaker diarization? A: Good for 2-6 speakers in clean audio; degrades with crosstalk or poor mics. Use a boundary microphone for best results.

Q: Can it handle Chinese meetings? A: Yes - Whisper supports 90+ languages including Chinese; set language="zh".

Q: What about virtual meeting audio? A: Use a virtual audio cable or the meeting app's "record audio only" export - any audio file works.

❓ FAQ

How accurate is speaker diarization?

Good for 2-6 speakers in clean audio; degrades with crosstalk or poor mics. Use a boundary microphone for best results.

Can it handle Chinese meetings?

Yes - Whisper supports 90+ languages including Chinese; set `language="zh"`.

What about virtual meeting audio?

Use a virtual audio cable or the meeting app's "record audio only" export - any audio file works.