Whistle against Whisper and Moonshine
word error rate %, lower is better
Whistle
Whisper base
Moonshine tiny v2
10
20
30
4.31
LibriSpeech
test-clean
10.49
LibriSpeech
test-other
7.65
SPGISpeech
19.01
Earnings-22
26.07
AMI
22.87
AMI
cleaned
7.61
TED-LIUM
21.4
FLEURS
24.9
MLS
Size
MB, smaller is better
16.9
145.3
41.9
Time to first token
ms on a 10 s clip, smaller is better
11.1
73.2
22.8
Decode
tokens/s on a 10 s clip, larger is better
1,319
266
262
Whistle's figures are labelled; every bar is to the same scale. Whisper and Moonshine word error rates
as published; Whisper's AMI figure is AMI-IHM. FLEURS and MLS average the seven languages, MLS the six
without English. Speed is 10 s of audio on an Apple M4 Pro, at each engine's defaults.