Downloads · 30 days
0
jeswinjestin/sherpa-onnx-nemo-ctc-indicconformer-malayalam
sherpa-onnx-nemo-ctc-indicconformer-malayalam is a automatic speech recognition model from jeswinjestin. Use it when you need speech turned into text. It is set up for sherpa-onnx. The card lists the license as cc-by-4.0.
A ready-to-use sherpa-onnx offline NeMo-CTC model for Malayalam speech-to-text. Runs fully offline on CPU. Drop-in for any sherpa-onnx app (desktop, Android, iOS, Raspberry Pi).
Downloads · 30 days
0
Access
Public
Updated Aug 8, 2026
Repo size
493 MB
Likes
2
Public
Click a slice to open those files.
.onnx493 MB · 100%
From the Hugging Face model README
A ready-to-use sherpa-onnx offline NeMo-CTC model for Malayalam speech-to-text. Runs fully offline on CPU. Drop-in for any sherpa-onnx app (desktop, Android, iOS, Raspberry Pi).
model.onnx — the CTC model (with sherpa-onnx metadata added).tokens.txt — the token table.import sherpa_onnx
rec = sherpa_onnx.OfflineRecognizer.from_nemo_ctc(
model="model.onnx", tokens="tokens.txt", num_threads=2, provider="cpu")
s = rec.create_stream()
s.accept_waveform(16000, samples) # 16 kHz mono float32
rec.decode_stream(s)
print(s.result.text)
The weights are NVIDIA/AI4Bharat IndicConformer (Conformer hybrid CTC/RNNT, CTC branch),
converted to ONNX by the community, then packaged for sherpa-onnx by adding the metadata its
NeMo-CTC loader needs (vocab_size, normalize_type=per_feature, subsampling_factor=8,
model_type=EncDecHybridRNNTCTCBPEModel) and generating tokens.txt from the vocabulary.
sulabhkatiyar/indicconformer-120m-onnx / trysem/indicconformer-120m-onnx.