MetalVoice

AI-Powered Noise Suppression for macOS

MetalVoice uses state-of-the-art Deep Learning (DeepFilterNet) to eliminate background noise and room reverb in real-time.
Built exclusively for Apple Silicon, leveraging the Neural Engine for zero-latency performance.

Important

Requirement: This application REQUIRES a Mac with Apple Silicon (M1, M2, M3, or newer). It relies on the Apple Neural Engine and Metal Performance Shaders optimized for these chips. It will not work on Intel Macs.

---

✨ Features

• 🚫 Silence the Noise: Instantly removes fans, typing, air conditioning, and background chatter.
• 🗣️ Studio Clarity: De-reverberation clears up "roomy" echoes, making you sound like you're in a studio.
• ⚡ Zero Latency Feel: Optimized Metal pipeline ensures negligible delay tailored for real-time communication.
• 🔒 Privacy First: 100% On-Device. Your voice is processed locally on your Neural Engine. No data leaves your Mac.
• 🛠️ Universal: Works with any microphone (USB, XLR, Built-in) and outputs to any app via virtual cables.
• 🔄 Automatic Updates: Check for updates and install new releases directly from the App and CLI.

📥 Installation

1. Download: Get the latest version from the Releases Page.
2. Unzip: Extract MetalVoice_v1.2.5.zip. Inside, you'll find the GUI MetalVoice.app and the terminal MetalVoiceCLI.
3. Install: Drag MetalVoice.app to your Applications folder.
4. Open: Right-click and choose Open (necessary for the first launch to verify the developer).
   • Note: If prompted about "Malicious Software", go to System Settings -> Privacy & Security -> Open Anyway.

🚀 Usage Guide

1. Launch: Look for the Waveform Icon 🌊 in your menu bar.
2. Select Input: Click the gear ⚙️ icon. Set Input Device to your physical microphone.
3. Select Output: Set Output Device to a Virtual Audio Cable (Recommended: BlackHole 2ch).
   • This routes the "Clean" audio into the cable.
4. Configure Apps: inside Discord/Zoom/OBS, set your Microphone Input to that same Virtual Cable (e.g., BlackHole 2ch).
5. Enable AI: Toggle the switch ON in the MetalVoice menu. Enjoy crystal clear audio!

💻 Terminal CLI & Interactive Dashboard

MetalVoiceCLI is included for lightweight terminal workflows, automated scripts, or dual-pipeline audio filtering (e.g. mic filtering + incoming call filtering):

Interactive Mode:

Simply run ./MetalVoiceCLI without arguments:

1. Choose your microphone and output device with a single keystroke.
2. An Apple-grade responsive live audio dashboard provides real-time VU level metering without spamming terminal history.

  MetalVoice v1.2.5 • Live Audio Pipeline
  ──────────────────────────────────────────────────────────────────────────────

  🎙️  INPUT       MacBook Pro Microphone [48000Hz, 1ch Planar]
     Level       [■■■■■■■■■■·························] 0.0847 RMS (94 pkts/s)

  ⚡  AI DSP      ● ACTIVE (DeepFilterNet)     Gain: 100%
     Output      [■··································] 0.0031 RMS (94 renders/s)

  📦  BUFFER      1024 frames (0 drops, 0 underruns)

  🔊  OUTPUT      BlackHole 2ch [48000Hz, 2ch Planar]
     Status      ● STREAMING

  ──────────────────────────────────────────────────────────────────────────────
   Q  Quit   A  AI (ON)   T  Tone (OFF)   + / -  Gain (100%)   H  History   U  Upgrade
  ──────────────────────────────────────────────────────────────────────────────

Live Interactive Hotkeys:

• Q $\to$ Quit cleanly
• A $\to$ Toggle DeepFilterNet AI noise reduction ON/OFF
• T $\to$ Toggle synthetic 440Hz diagnostic test tone
• + / - $\to$ Live output volume gain adjustment
• H $\to$ Toggle between the live dashboard and scrolling log history
• U $\to$ Upgrade in-place when a new release is available

Command Line Options & Scripting:

# Direct flag launch (no prompts)
./MetalVoiceCLI --in "MacBook Pro Microphone" --out "BlackHole 2ch" --gain 1.0

# Check for updates
./MetalVoiceCLI --check-update

# Automatically download & install the latest release
./MetalVoiceCLI --update

💻 Tech Stack

• Core: Swift 5, SwiftUI
• Audio: AVFoundation, Accelerate (vDSP)
• AI Engine: CoreML + Metal Performance Shaders (MPS)
• Model: DeepFilterNet3 (Streaming UNet)

🙏 Credits

• DeepFilterNet: Developed by Hendrik Schröter (Rikorose). DeepFilterNet is one of the most efficient noise suppression models available. MetalVoice uses a custom CoreML implementation of this architecture.
• Icons: SF Symbols by Apple.

---

Made with ❤️ for macOS