byteknotlabs/AksharSTT-V4
Automatic Speech Recognition • Updated • 1
Offline speech to text for Hindi, English and Hinglish. Open weights, Apache 2.0. akshar.byteknotlabs.com
Note Latest. 29.5 MB, mean WER 20.78 on six public test sets
Note 20.4 MB, 2-bit with a 4-bit encoder, mean WER 21.71
Note The first release, mean WER 22.28