Audio Cleaner for Apple Silicon: M1, M2, M3, M4
Updated: May 2026
TL;DR: Aulio Studio ships as a native Apple Silicon build. The AI inference pipeline uses the GPU and the Neural Engine, which is why a 2-minute clip cleans in about 20 seconds on M1 and roughly 8 seconds on M4. Intel Macs are supported too, just slower.
Why Apple Silicon matters for AI audio processing
AI audio cleaning is computationally intensive. A neural model processes every millisecond of your audio file through thousands of learned parameters. On older hardware or a generic x86 build, this takes time. On Apple Silicon, with its unified memory architecture and dedicated Neural Engine, the same computation runs dramatically faster.
Aulio Studio is compiled as a native ARM binary, not an Intel app running through Rosetta 2. Native compilation means the app can use the full instruction set of the M-series chips and schedule work across the performance cores, efficiency cores, GPU, and Neural Engine as appropriate. The AI inference work is offloaded to the hardware that does it best.
Processing times by chip generation
These are typical times for a 2-minute mono voice recording, noise reduction only:
- M1 / M1 Pro: approximately 20 seconds
- M2 / M2 Pro: approximately 14 seconds
- M3 / M3 Pro: approximately 11 seconds
- M4 / M4 Pro: approximately 8 seconds
- Intel Mac (x86): approximately 60–90 seconds depending on CPU generation
Times scale roughly linearly with recording length. A 30-minute episode on M1 takes around 3–4 minutes; on M4, around 2 minutes. Batch jobs run sequentially, so a 10-episode batch on M4 completes in about 20 minutes of unattended processing.
The role of the Neural Engine
Apple Silicon chips include a dedicated Neural Engine block specifically designed for matrix multiply operations, which are the core computation in neural networks. The Neural Engine handles this type of work without competing for GPU bandwidth or CPU cycles. This is why AI inference on Apple Silicon is fast even while other work is happening on the machine: the audio processing is using hardware that most other tasks do not touch.
Aulio Studio uses Apple's Core ML framework to schedule inference work, which automatically routes computation to the most efficient available hardware on your specific chip.
Memory and battery impact
The AI model in Aulio Studio loads into memory once at startup and stays resident. On M1 with 8 GB unified memory, the model footprint is around 80 MB. This is the same regardless of file size: a 2-minute file and a 60-minute file use the same model, just for longer.
Battery impact during processing is noticeable but brief. A 2-minute clip causes a short burst of GPU/Neural Engine activity that completes in seconds. Processing a 30-minute episode draws down roughly 3–5% of battery on an M2 MacBook Air. Batch processing is the heavier scenario: a 10-episode overnight batch on battery is not ideal, but on mains power it runs fine.
Intel Mac support
Aulio Studio runs on Intel Macs running macOS 14 or later. The app falls back to CPU-based inference, which is slower but produces identical results. If you are on an Intel Mac and processing times feel slow, the answer is running the job overnight or as a background task rather than waiting in front of it.
Try it on your Mac
The 14-day free trial from the Mac App Store is the full app on your hardware. You can benchmark how fast it processes your typical file lengths before deciding to purchase. No credit card required.
Aulio Studio home · AI audio cleaner for Mac · Local AI audio on Mac · Free trial download