Whisper ASR for Wollo Amharic Dialect
Parameter-efficient fine-tuning (PEFT) of OpenAI Whisper for regional Amharic dialect speech recognition.
Adapting state-of-the-art speech recognition models to regional Amharic dialectal variations (Wollo speech) using parameter-efficient fine-tuning (LoRA), advancing dialect-robust acoustic modeling.
Key Highlights
- Dialect Acoustic Modeling: Capturing regional phonological shifts, lexical variations, and acoustic nuances distinct from standard Addis Ababa Amharic.
- Compute-Efficient Speech PEFT: Fine-tuning Whisper encoder-decoder attention projections while freezing core acoustic representation weights.
- Low-Resource Audio Pipeline: Processing dialect audio recordings and aligning transcriptions using Ethiopic Fidel orthography.