Fact · Speech recognition
20times faster than real time
In our October 2026 benchmark on a Mac with an M2 Pro, Parakeet TDT 0.6B v3 transcribed 100 Romanian FLEURS recordings about 20 times faster than real time.
Measured by us on October 19, 2026, from 100 recordings
Our own measurement
This number comes from our benchmark, not from another publication. The full record (the setup, every recording's checksum, every transcript and timing) is published, so you can recompute it.
- What it measures
- NVIDIA Parakeet TDT 0.6B v3 converted to Core ML (Parakeet Ultra, FluidInference, model revision 95eaa59a39d4), run through FluidAudio 0.17.4 on the CPU and GPU, with the language given. This is the engine the Yap app uses on Mac. Run on macOS 26.6.2, 16 GB of memory; other apps were open, as on any working Mac.
- How it is measured
- Speed = 956.0 seconds of audio / 47.34 seconds of processing = 20.2 times real time, over 100 recordings processed one at a time on a warm engine.
- What it does not show
- Speed depends on the chip, memory and what else the Mac is doing, and short sentences carry a fixed overhead per recording, so long recordings run faster than this.
In the data
A row of Speech Recognition Speed: Real-Time Factor Data, as Parakeet TDT v3, Romanian (M2 Pro): 20 times faster than real time.
- System
- Parakeet TDT 0.6B v3
- Kind
- Open model
- Hardware
- Apple M2 Pro
- Test set
- FLEURS
- Year
- 2026