Vosk Speech to text

[Akash Sharma]

Hi, I am working on a speech to text project for converting Hindi Audio calls to transcripts. I am using  ‘vosk-model-hi-0.22’. The model gives good output but due to different accents and background noise, the model apparently hallucinates and gives out random (sometimes rhyming) words. I have used audio pre-processing as well, but it doesn't change the output enough. Need suggestions on training and fine tuning.