Downloads · 30 days
31
13% of all-time downloads
Reza2kn/canary-180m-persian-semiclean31-fast-asr
canary-180m-persian-semiclean31-fast-asr is a automatic speech recognition model from Reza2kn. Use it when you need speech turned into text. It is set up for nemo. The card lists the license as cc-by-4.0.
Experimental full-parameter ASR-only fine-tunes of nvidia/canary-180m-flash with a Persian-heavy tokenizer.
Downloads · 30 days
31
13% of all-time downloads
All-time downloads
230
Public
Repo size
1.5 GB
Likes
0
Public
Click a slice to open those files.
.nemo1.5 GB · 100%
From the Hugging Face model README
Experimental full-parameter ASR-only fine-tunes of nvidia/canary-180m-flash
with a Persian-heavy tokenizer.
Data mix: Reza2kn/persian-asr-semi-clean-31h-awq-wer selector plus a small
English FLEURS retention slice. Filtered/mixed train split: 46,006 rows /
31.742h. Validation split: 938 rows / 0.652h.
Tokenizer: 6144 Persian BPE pieces, 512 English BPE pieces, plus Canary ASR special tokens.
Artifacts:
canary_180m_persian_semiclean31_gpu0_bd900_lr1e4.nemo: 4000 steps,
batch_duration 900, lr 1e-4.canary_180m_persian_semiclean31_gpu1_bd700_lr7e5.nemo: 4000 steps,
batch_duration 700, lr 7e-5.Quick validation on the semiclean mixed held-out split is included under
eval/. This is a small internal split, not a production benchmark:
These are research checkpoints, not production Android exports.