Update README.md

README.md

@@ -33,3 +33,12 @@ configs:
   - split: test
     path: data/test-*
 ---
+
+
+Pomak is an endangered South East Slavic language variety spoken in Northern Greece. It belongs to the South Slavic dialect continuum and is closely related to Bulgarian, while also showing significant influence from Greek and Turkish due to long-term language contact. Pomak is characterized by substantial phonological and lexical variation across speaker communities and lacks a standardized orthography. Despite its linguistic significance, Pomak remains severely under-documented, and language technology resources for the variety are extremely limited.
+
+The speech corpus used to train this model was collected in a controlled studio setting at the ILSP audio-visual studio in Xanthi, Greece. Four native Pomak speakers (two female and two male) recorded readings of Pomak texts, resulting in approximately 14 hours of speech data. All recordings were carried out with the informed consent of the participants. For ASR training, long recordings were segmented into shorter utterances of up to 25 seconds, a process that removed most long pauses and yielded a final training dataset of 11 hours and 8 minutes of segmented speech.
+
+This dataset was used to fine-tune a pre-trained Slavic wav2vec2 model (classla/wav2vec2-large-slavic-parlaspeech-hr) using the Hugging Face Transformers library. The resulting model, wav2vec2-xls-r-slavic-pomak, constitutes the first automatic speech recognition system developed for Pomak. Evaluation on a held-out test set comprising 10% of the data shows a substantial improvement over the pre-trained baseline, with Word Error Rate (WER) reduced from 31.47% to 3.12% and Character Error Rate (CER) reduced from 87.31% to 9.06%.
+
+This model and its associated training pipeline were presented in the paper “ASR pipeline for low-resourced languages: A case study on Pomak”. The work demonstrates that fine-tuning large multilingual and family-specific ASR models can yield high-quality speech recognition performance even in extremely low-resource settings, and aims to support future linguistic research, corpus creation, and language documentation efforts for Pomak.