Dulu Malayalam CTC β€” combined 778-hour training

Private research checkpoint: epoch 9, step 44874, selected by equal-weight mean Kathbath and IndicVoices development CER (11.4771%).

Training data

319,063 clips, 778.258 hours: Kathbath 45,104; IndicVoices 211,032; IMaSC 34,443; AIR 28,484.

Selected checkpoint evaluation

Split CER % WER %
Kathbath development 6.05 36.18
IndicVoices development 16.90 53.13
IndicVoices test 15.24 52.52
FLEURS validation 8.42 29.71

Single seed. Evaluation sets have prior development exposure. Exact audio/text overlap was removed, but near-duplicate news content may remain. Metrics are not evidence of production readiness. Full precision values are in evaluation.json.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for whiletruelearn/dulu-malayalam-ctc-778h

Finetuned
(178)
this model