Whistle
by Cactus Compute · Model · Voice & audio
unproven
Tiny CPU-only speech-to-text model; speed and accuracy claims are the maker's own.
16.9 MB speech-to-text model runs on CPU only, 30 s per pass, auto language detection, word timestamps
AccessOpen weightsDownloadabout 17 MBLicenseApache 2.0
Why we say so
AI Search confirmed the specs: 16.9 MB, CPU only, 30-second passes, word timestamps and automatic language detection. The claim of decoding about six times faster than Whisper Base with lower error on some benchmarks comes only from Cactus Compute. No independent tester has checked it.
Evidence
- Oct 4Decodes about six times faster than Whisper Base with lower error rate on some benchmarks; installs via pip