Fact · Speech recognition
0.10percentage points of word error rate
In our October 2026 test on English readings of 3 to 4 minutes handed over cut at pauses into pieces of 10 to 20 seconds, Parakeet TDT 0.6B v3's word error rate was 0.10 percentage points higher than Whisper large-v3-turbo's (not a measured difference: its interval, allowing for the test's engine comparisons, includes zero, so the two engines count as level here).
Measured by us on October 25, 2026, from 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
- Both engines transcribed the same audio (Parakeet TDT v3: 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. Whisper large-v3-turbo: OpenAI Whisper large-v3-turbo, 5-bit quantized (ggml-large-v3-turbo-q5_0.bin), run through whisper.cpp 1.9.4 with Metal on the GPU, with the language given.) Whisper ran with greedy decoding and timestamps on, which is how whisper.cpp moves through audio longer than its 30-second window; whisper.cpp's beam search was not tested. 84 examples of 3 to 4 minutes read by 40 speakers from public-domain audiobooks (LibriSpeech-Long test-clean, Google DeepMind, from LibriSpeech), clean recordings of read English.
- How it is measured
- Difference of the two engines' pooled word error rates on the same readings. Interval: 5,000 paired bootstrap draws of speakers (seed 7); 95% for this comparison alone -0.10 to +0.31, and allowing for all 6 engine comparisons (Bonferroni) -0.16 to +0.39 points, the interval the verdict stands on.
- What it does not show
- LibriSpeech's transcripts spell numbers as words, so an engine that writes digits loses a little against one that spells them, in both conditions; compare the length cost of each engine (its own two conditions) for what the length of a recording does.