Views
No views yet
model_p3_epoch15_avg8.pt — CURRENT BEST (Phase 3): scaled up on ~337h.model_p1_epoch30_avg8.pt — Phase 1 baseline (~89h).| Eval set | Phase 1 (~89h) | Phase 3 (~337h) |
|---|---|---|
| IndicVoices-hi (Hindi) | 37.2% | 25.7% |
| call-test (internal hinglish calls) | 47.3% | 39.9% |
| Svarah (Indian English) | 50.3% | 44.5% |
num_encoder_layers=2,2,3,4,3,2, encoder_dim=192,256,384,512,384,256,
downsampling_factor=1,2,4,8,4,2, causal, chunk_size=16,32,64,-1, left_context_frames=64,128,256,-1.
Pure CTC head, 80-dim fbank @16kHz, subsampling 4 (~25 Hz). Vocab 103 tokens
(<blk>,<sos/eos>,<unk>,▁=word-boundary, 63 Devanagari + 26 a-z + 10 digits).unsanitized_normalized (code-switch word [English] → English, tags stripped);
all text NFC-normalized, punctuation/foreign scripts dropped, Latin lowercased.--init-ckpt), base_lr 0.02, 15 ep, non-finite-loss batches skipped. Single NVIDIA L4.model_p3_epoch15_avg8.pt / model_p1_epoch30_avg8.pt — averaged weights + arch config + vocab_size.
tokens.txt — vocab. scripts/ — full reproduction pipeline.config, load_state_dict(ckpt["model"]),
extract 80-dim lhotse fbank, run encoder_embed → encoder → ctc_output → ctc_greedy_search,
map ids via tokens.txt, replace ▁→space. See scripts/eval_wer.py and scripts/server_ctc.py.