Tail-word retention
The benchmark harness includes a synthetic regression for short push-to-talk clips where the final word is easy to lose. It compares raw Unified, padded Unified, and production v3 before any candidate model is exposed in the app.
Benchmarks
How fast is Presspeech?
Presspeech keeps its latency claims tied to a reproducible benchmark in experiments/swift-bench/. Numbers below are p50 across five trials per backend, first inference excluded.
Setup
Environment
MachineMac mini M4, 10 cores, 16 GB.
OS and toolchainmacOS 26.4.1, Xcode/Swift 6.3.
TrialsFive per backend per clip. First inference after model load excluded from p50.
AudioGenerated 16 kHz mono Float32 WAV clips, all tested against the same backends.
p50 latency
Results
| Clip | Duration | FluidAudio on ANE | Prior GPU path | Speedup |
|---|---|---|---|---|
short-clean |
2.50 s | 1.57× | ||
medium-clean |
3.99 s | 1.83× | ||
disfluent |
5.31 s | 1.97× | ||
longer-technical |
9.49 s | 1.97× |
Machine-readable results: benchmarks/results.json. Reproduction docs: experiments/swift-bench.
Candidate models
Candidate model evaluation
Private corpus
For candidate-model decisions, maintainers can run a local-only v3-vs-Unified comparison over private real-dictation clips. Reports keep transcript text and filenames redacted by default while showing WER, final-word failures, and latency.
Release gate
run-release-asr-checks.sh checks production v3 by default. Pass --include-candidate-models to add Unified tail-word and v3-vs-Unified comparisons.
| 2026-06-23 public smoke backend | Avg WER | Worst WER | Final-word failures | Avg p50 |
|---|---|---|---|---|
v3 |
1.4% | 7.0% | 0 | 100.7 ms |
apple |
11.7% | 18.2% | 0 | 384.8 ms |
nemotron-en |
5.3% | 11.3% | 0 | 1032.7 ms |
Apple SpeechAnalyzer and Nemotron English both stayed behind Parakeet TDT v3 on this smoke corpus, so neither is a production replacement.
Reading the numbers
Interpretation
Latency
The ANE path is consistently faster than the prior GPU path on these clips, with p50 results around 92-152 ms.
Accuracy
Both backends produced essentially identical transcripts on the synthetic test audio, including the same TTS artifacts.
Power
The benchmark measures compute latency, not energy. Power measurement remains a future bench improvement.