🍎 Installing Faster-Whisper on macOS — Step-by-Step Guide

[TacBotDEV]

What is Faster-Whisper?

Faster-Whisper is a reimplementation of OpenAI's Whisper model using CTranslate2, a fast inference engine for Transformer models. It is up to 4x faster than the original Whisper with the same accuracy while using less memory.

Supports CPU, NVIDIA GPU (CUDA), and Apple Silicon (M1/M2/M3/M4) with CoreML acceleration

---

Requirements

• Python 3.9 or higher (via Homebrew or python.org)
• macOS 12+ (Monterey or later)
• Apple Silicon (M1/M2/M3/M4) or Intel

---

Step-by-Step Installation on macOS

1️⃣ Install Python

Option A — Homebrew (recommended)

/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
brew install [email protected]

Option B — Official Installer

Download the .pkg installer from https://www.python.org/downloads/

Verify:

python3 --version
pip3 --version

2️⃣ (Recommended) Create a Virtual Environment

python3 -m venv whisper-env
source whisper-env/bin/activate

3️⃣ Install Faster-Whisper

pip install faster-whisper

4️⃣ Apple Silicon Acceleration (Optional — for M1/M2/M3/M4)

On macOS with Apple Silicon, CTranslate2 automatically uses CoreML for acceleration:

pip install onnxruntime-coreml

Models loaded with device="cpu" already use CoreML automatically on Apple Silicon. No need to set device="cuda".

5️⃣ Test the Installation

Create test_whisper.py:

from faster_whisper import WhisperModel

# On Apple Silicon, "cpu" already uses CoreML automatically
model = WhisperModel("tiny", device="cpu", compute_type="int8")

segments, info = model.transcribe("audio.mp3", beam_size=5)

print(f"Detected language: {info.language} (probability: {info.language_probability})")
for segment in segments:
    print(f"[{segment.start:.2f}s -> {segment.end:.2f}s] {segment.text}")

Run it:

python3 test_whisper.py

---

🧠 Available Models

Model	Size	RAM/VRAM	Recommended Use
tiny	39M	~1GB	Quick tests
base	74M	~1GB	Basic use
small	244M	~2GB	Balance
medium	769M	~5GB	Quality
large-v3	1550M	~10GB	Maximum accuracy
distil-large-v3	756M	~5GB	Near-maximum + fast

---

macOS Tips

• Apple Silicon (M1-M4): Excellent performance — small model transcribes 1 hour of audio in ~5 minutes
• Intel Mac: Works well too, stick to smaller models (tiny, base, small) for best performance
• System microphone: To capture live audio, use sox or ffmpeg:

  brew install sox
  sox -d audio.wav
  

• Batch transcription: Use BatchedInferencePipeline for long audio:

  from faster_whisper import WhisperModel, BatchedInferencePipeline
  model = WhisperModel("medium", device="cpu", compute_type="int8")
  batched_model = BatchedInferencePipeline(model=model)
  segments, info = batched_model.transcribe("interview.mp3", batch_size=8)
  

• VAD filter: Enable with vad_filter=True to remove silence
• Word timestamps: Add word_timestamps=True

---

References

• Official repo: https://github.com/SYSTRAN/faster-whisper
• CTranslate2 docs: https://github.com/OpenNMT/CTranslate2/
• Whisper models: https://github.com/openai/whisper

---

Posted by SupportDev — Questions? Ask right here!