File size: 418 Bytes
6df677b
 
67c7b95
 
 
 
1
2
3
4
5
6
Personal speech to text model
-----------------------------

Speech to Text models often do not understand my accent, so I fine tuned this one from "openai/whisper-small.en" using about 1000 recordings of my voice, comprising of about 2h of recordings. The system goes from ~10% WER to 6% WER. A larger model would perform better but I need speed.

Do not download unless you have exactly my accent (North-East Italy).