👀 Running F5-TTS on a Refurbished Mac Studio M2 Ultra: A Setup Guide

Date Created: 2025-05-08
By: 16BitMiker
[ BACK.. ]

Setting up a high-performance text-to-speech system like F5-TTS on Apple Silicon might seem daunting, but with the right steps, it’s surprisingly smooth. I recently configured F5-TTS on a newly refurbished Mac Studio with the M2 Ultra chip, and I’m here to walk you through the entire installation process—what worked, what didn’t, and how to get everything running beautifully on macOS.

Let’s dive in. 🧠🔊

📋 What Is F5-TTS?

F5-TTS is a state-of-the-art, non-autoregressive text-to-speech system developed by SWivid. It uses flow matching and Diffusion Transformers (DiT) to synthesize high-quality speech quickly and efficiently. It supports real-time inference, multilingual output, and even voice cloning—all in an open-source package.

🏔️ System Prep: Starting Fresh on macOS M2 Ultra

Here’s the software and hardware context:

Machine: Mac Studio (M2 Ultra, 128GB RAM)
OS: macOS Sonoma 14.4+
Python: 3.13 (though 3.10 is recommended)
Package Manager: Homebrew
Virtual Environment: venv

Let’s break it down step by step.

📦 Step-by-Step Installation

✅ Step 1: Install Homebrew and Core Dependencies


x
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
eval "$(/opt/homebrew/bin/brew shellenv)"
echo 'eval "$(/opt/homebrew/bin/brew shellenv)"' >> ~/.zshrc
source ~/.zshrc

# Install Python 3.10 (recommended) or use existing Python 3.13
brew install python@3.10
brew install git ffmpeg wget

✅ Step 2: Clone the Repository


xxxxxxxxxx
mkdir -p $HOME/git
cd $HOME/git
git clone https://github.com/SWivid/F5-TTS.git
cd F5-TTS

✅ Step 3: Set Up a Virtual Environment


xxxxxxxxxx
python3 -m venv f5tts_env
source f5tts_env/bin/activate

✅ Step 4: Install the Package

F5-TTS uses a pyproject.toml, so install it in editable mode:


xxxxxxxxxx
pip install --upgrade pip
pip install -e .

✅ Step 5: Install PyTorch with MPS (Metal Performance Shaders) Support


xxxxxxxxxx
pip uninstall -y torch torchvision torchaudio
pip install torch torchvision torchaudio

Then verify GPU support:


xxxxxxxxxx
python -c "import torch; print('MPS Available:', torch.backends.mps.is_available())"

✅ Output:


xxxxxxxxxx
MPS Available: True

🔽 Step 6: Download Pretrained Model Checkpoints


xxxxxxxxxx
mkdir -p ckpts/F5TTS_v1_Base
wget -P ckpts/F5TTS_v1_Base https://huggingface.co/SWivid/F5-TTS/resolve/main/F5TTS_v1_Base/model_1250000.safetensors

📝 Note: This is a ~1.3 GB file and may take a few seconds depending on your internet speed.

▶️ Step 7: Run the Inference CLI

First, grab a sample reference voice:


xxxxxxxxxx
curl -L https://github.com/SWivid/F5-TTS/raw/main/examples/reference.wav -o reference.wav

Then run the inference:


xxxxxxxxxx
f5-tts_infer-cli \
  --model F5TTS_v1_Base \
  --ref_audio reference.wav \
  --gen_text "This is F5-TTS running on the M2 Ultra Mac." \
  --output_dir results

✅ Result:

The output audio is saved in the results/ folder.
The voice matches the reference sample remarkably well.

🌐 Bonus: Run the Gradio Web UI

Want a user-friendly way to test your voices?


xxxxxxxxxx
f5-tts_infer-gradio --port 7860

Then open your browser and visit:

👉 http://localhost:7860

You’ll get a full-featured web interface with:

Reference audio upload
Text input
Model selection
Output playback

🔧 Troubleshooting and Notes

Python Compatibility

F5-TTS officially recommends Python 3.10 due to some library constraints. I used Python 3.13 without issues, but if you run into errors (especially with librosa or numpy), consider downgrading.

Missing Models

If inference fails with a checkpoint error, double-check your directory:


xxxxxxxxxx
ls ckpts/F5TTS_v1_Base/
# Should show: model_1250000.safetensors

Missing CLI Commands

If f5-tts_infer-cli or f5-tts_infer-gradio doesn't work, try running the Python modules directly:


xxxxxxxxxx
python -m f5_tts.infer.examples.cli
python -m f5_tts.infer.examples.gradio_app

🔁 Relaunching F5-TTS After a System Restart

If you’ve rebooted your Mac or closed your terminal and want to relaunch F5-TTS, just follow these quick steps:

✅ Open Terminal and navigate to your project:


xxxxxxxxxx
cd $HOME/git/F5-TTS

✅ Reactivate your virtual environment:


xxxxxxxxxx
source f5tts_env/bin/activate

✅ (Optional) Verify access to CLI tools:


xxxxxxxxxx
f5-tts_infer-cli --help

✅ Run Inference Again:


xxxxxxxxxx
f5-tts_infer-cli \
  --model F5TTS_v1_Base \
  --ref_audio reference.wav \
  --gen_text "Welcome back! This is F5-TTS after a reboot." \
  --output_dir results

✅ Or launch the Gradio interface:


xxxxxxxxxx
f5-tts_infer-gradio --port 7860

🧠 Pro Tip: If you frequently reboot, consider adding an alias to your .zshrc:


xxxxxxxxxx
echo 'alias f5tts="cd ~/git/F5-TTS && source f5tts_env/bin/activate"' >> ~/.zshrc

Then you can just type:


xxxxxxxxxx
f5tts

And you're ready to roll. 🚀

👥 What Worked Well

✅ PyTorch MPS support was plug-and-play
✅ CLI tools launched without issue
✅ Model and reference audio integration was seamless
✅ The Gradio UI worked out-of-the-box
✅ Voice cloning was fast and expressive

🚀 Final Thoughts

Setting up F5-TTS on an Apple Silicon Mac Studio was not only possible—it was enjoyable. Performance was smooth, and the installation process was relatively painless thanks to modern Python packaging and MPS support.

If you’re looking to explore real-time voice synthesis, prototyping audiobooks, or building a personalized voice assistant, F5-TTS is a solid open-source starting point.