Downloads · 30 days
0
ako101/parakeet-tdt-0.6b-v3-sherpa-onnx-fp16
parakeet-tdt-0.6b-v3-sherpa-onnx-fp16 is a automatic speech recognition model from ako101. 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.
ONNX export of nvidia/parakeet-tdt-0.6b-v3 packaged for sherpa-onnx. The FastConformer encoder is exported in fp16 (near-fp32 accuracy) and the decoder/joiner are int8, giving a compact on-device transducer suitable f…
Downloads · 30 days
0
Access
Public
Updated Jun 5, 2026
Repo size
1.3 GB
Likes
1
Public
Click a slice to open those files.
.onnx1.3 GB · 100%
From the Hugging Face model README
ONNX export of nvidia/parakeet-tdt-0.6b-v3
packaged for sherpa-onnx. The FastConformer
encoder is exported in fp16 (near-fp32 accuracy) and the decoder/joiner are int8,
giving a compact on-device transducer suitable for mobile.
| File | ~Size | Notes |
|---|---|---|
encoder.fp16.onnx | 1.18 GB | FastConformer encoder, fp16 (float32 IO) |
decoder.int8.onnx | 11 MB | transducer decoder, int8 |
joiner.int8.onnx | 6 MB | transducer joiner, int8 |
tokens.txt | 92 KB | token table |
The fp16 encoder is exported natively from NeMo (post-hoc float16 conversion produces an invalid graph for this encoder's length-arithmetic subgraph). Inputs and outputs are kept float32 via inserted casts, so sherpa-onnx can feed standard float32 features.
import sherpa_onnx
recognizer = sherpa_onnx.OfflineRecognizer.from_transducer(
tokens="tokens.txt",
encoder="encoder.fp16.onnx",
decoder="decoder.int8.onnx",
joiner="joiner.int8.onnx",
model_type="nemo_transducer",
feature_dim=128,
decoding_method="greedy_search",
num_threads=4,
)
import soundfile as sf
samples, sr = sf.read("audio.wav", dtype="float32")
s = recognizer.create_stream()
s.accept_waveform(sr, samples)
recognizer.decode_stream(s)
print(s.result.text)
Audio should be 16 kHz mono. model_type="nemo_transducer" and feature_dim=128
are required for this model.
Derived from NVIDIA's parakeet-tdt-0.6b-v3, released under
CC-BY-4.0. This export is redistributed
under the same license. Please credit NVIDIA NeMo and the original model.