whisper-small-fsc
openai/whisper-small finetuned for Filipino
(Tagalog/Taglish) on
sapinsapin/filipinospeechcorpus.
Trained for 2000 steps on 10000 clips (batch 8×2, lr 1e-05, fp16 + gradient checkpointing). WER/CER are on the held-out split, lowercased; CER is the model-selection metric (Taglish orthography varies at the word level).
| metric | value |
|---|---|
| loss | 0.1793 |
| wer | 0.1591 |
| cer | 0.0706 |
Trained with finetune_asr.py from the
halohalo pipeline; the dataset
adapter normalizes each corpus to (audio@16k, text, speaker_id) so corpora
are swappable with a --dataset flag.