Model

Wav2Vec2 XLS-R · Turkish

Purpose

This model turns Turkish speech into text with a 300-million-parameter XLS-R speech encoder fine-tuned on Turkish recordings from Mozilla Common Voice 8.0. A decoding step called CTC merges repeated predictions and removes blanks to produce a transcript. The model card reports a word error rate of 0.2869 on its evaluation set, or about 28.7%.

Base model or method
Meta Wav2Vec2 XLS-R 300M

Sources

Wav2Vec2 XLS-R · Turkish

Sequence Modeling with CTC