fax1le/kazakh-omni-asr-ctc-300m-v2
1
Kazakh OmniASR CTC 300M v2 Fine-tuned
Fine-tuned version of omniASR_CTC_300M_v2 for Kazakh ASR.
Training data
- ISSAI KSC2 Kazakh speech data
- FLEURS Kazakh (
kk_kz) - augmented KSC2 training data
Training details
- Base model:
omniASR_CTC_300M_v2 - Tokenizer:
omniASR_tokenizer_written_v2 - Steps: 20,000
- Best checkpoint: step 20,000
- Validation WER: 23.1052
Files
checkpoint/sdp_00.pt— fine-tuned checkpointkazakh_ctc_300m_v2_20k.yaml— Fairseq2/OmniASR asset cardtraining_config.yaml— training config
