Full-context Nepali ASR — the strongest open model we measured on real Nepali telephone audio (see the NepTel leaderboard in the repo).
Try it: demo Space ·
Everything else: github.com/Ampixa/nepaliconformer
(NepTel benchmark, per-system outputs, full honest results)
121.3M-parameter 17-layer Conformer (d=512, striding ×4, 40 ms frames), hybrid TDT/CTC decoder,
1,024-piece Devanagari SentencePiece. Full self-attention, offline decoding.
~1,655 h of mostly conversational Nepali (YouTube podcasts/interviews) with Google Chirp 2
pseudo-labels + 105 h human-labeled read speech; telephony codec, noise, reverb and tempo
augmentation. Label-noise ceiling and every measured limitation (English, sung speech, slow
speech, end-of-turn) are documented in the
repo's RESULTS.md.
1from nemo.collections.asr.models import EncDecHybridRNNTCTCBPEModel
2m = EncDecHybridRNNTCTCBPEModel.restore_from("nepali_conformer_offline.nemo")
3print(m.transcribe(["audio.wav"])[0].text)
License: CC-BY-NC-4.0 (weights). Code in the repo: MIT.
