# Multi-source data fine-tuning

This is an automatic speech recognition model fine-tuned on approximately 1000 hours of Kazakh speech data based on the Whisper large-v3-turbo model, with a character error rate of 9.16% on the test set.

Finnish automatic speech recognition (ASR) model fine-tuned based on Facebook's wav2vec2-xls-r-300m, trained with 275.6 hours of Finnish data, supports decoding with KenLM language model

Catalan speech recognition model fine-tuned from facebook/wav2vec2-large-xlsr-53, trained on Common Voice and parliamentary speech datasets

A fine-tuned version of Facebook's wav2vec2-xls-r-1b model for Finnish automatic speech recognition (ASR), trained with 259.57 hours of annotated Finnish speech data