# tinyconformer-ctc-ls100

Sub-1M conformer + char-level CTC trained on LibriSpeech train.clean.100 (pilot).

- **test.clean**: WER 68.45%, CER 26.97% (52576 words)
- **test.other**: WER 82.50%, CER 39.68% (52343 words)

Params: 978,461. Greedy decoding, no language model, trained within a 2 USD compute budget.