Whistle: speech to text in a 16.9MB file

Cactus Whistle is an ASR model designed for ultra-small devices, offering a 16.9MB file size, which is 9x smaller than Whisper base, and operates 6x faster. It supports English, German, French, Spanish, Italian, Dutch, and Polish. Whistle achieves a 4.31 WER on LibriSpeech test-clean and 10.49 on test-other, outperforming Whisper base's 4.9 and 11.0. It also scores 21.4 on the FLEURS average against 24.5, 7.65 on SPGISpeech, and 19.01 on Earnings-22.

This report details Whistle's performance metrics, showing it mostly beats Whisper base with 9x less file size and 6x speed, unlike other general descriptions.