Cassini model downloads

Speech recognition models for Cassini. Each file is compressed with seekable Zstandard and has a permanent SHA-256 address.

Downloads support resuming. Decompress each .zst file before loading the model; no archive extraction is needed.

Download manifest · SHA-256 checksums · Attribution and licenses

Parakeet TDT-CTC 110M — int8

Download: 95.41 MiB · Installed size: 125.56 MiB

FileDownloadInstalled
model.int8.onnx.zst95.40 MiB125.55 MiB
tokens.txt.zst4.92 KiB9.72 KiB

Parakeet TDT 0.6B v3 — int8

Download: 443.27 MiB · Installed size: 639.42 MiB

FileDownloadInstalled
encoder.int8.onnx.zst431.27 MiB621.97 MiB
decoder.int8.onnx.zst8.41 MiB11.30 MiB
joiner.int8.onnx.zst3.54 MiB6.06 MiB
tokens.txt.zst44.76 KiB91.74 KiB

Parakeet TDT 0.6B v3 — fp32

Download: 2.20 GiB · Installed size: 2.37 GiB (including optional files)

FileDownloadInstalled
encoder.onnx.zst35.43 MiB39.90 MiB
encoder.weights.zst2.11 GiB2.27 GiB
decoder.onnx.zst41.70 MiB45.05 MiB
joiner.onnx.zst22.23 MiB24.11 MiB
tokens.txt.zst44.76 KiB91.74 KiB
bpe.vocab.zst (optional)46.21 KiB114.66 KiB

Silero VAD

Download: 458.35 KiB · Installed size: 628.76 KiB

FileDownloadInstalled
silero_vad.onnx.zst458.35 KiB628.76 KiB

Parakeet models by NVIDIA, ONNX exports distributed for sherpa-onnx, under CC BY 4.0. Silero VAD by Silero Team, under the MIT license. Decompressed model contents are unchanged.