|
Download README.md from baker76/speech-models: direct link, hf CLI and curl.
- Browser
- Download file 2.94 kB
-
https://huggingface.co/baker76/speech-models/resolve/main/README.md
- Command line
-
hf download hf://baker76/speech-models/README.md
-
curl -L -o README.md https://huggingface.co/baker76/speech-models/resolve/main/README.md
2.94 kB
| library_name: onnx | |
| pipeline_tag: automatic-speech-recognition | |
| license: other | |
| tags: | |
| - onnx | |
| - sherpa-onnx | |
| - speech-recognition | |
| - streaming-asr | |
| # Streaming speech models, as ONNX | |
| The speech recognition models used by [Rhyme Station](https://github.com/benbaker76/RhymeStation), | |
| a freestyle rap trainer, to turn the microphone into words as they are spoken. The app downloads | |
| the one it is set to use from here, as do its `tools/fetch_model.sh` and `tools/fetch_model.cmd`. | |
| **Nothing here was trained or converted by this organization.** The files are unmodified copies, | |
| kept here so the app has a stable place to download from, and each model stays under the | |
| license its authors gave it: see the file in its folder. | |
| | Folder | Size | Model | ONNX export | License | | |
| | --- | --- | --- | --- | --- | | |
| | `sherpa-onnx-nemotron-3.5-asr-streaming-0.6b-560ms-int8-2026-06-11` | 682 MB | [nvidia/nemotron-3.5-asr-streaming-0.6b](https://huggingface.co/nvidia/nemotron-3.5-asr-streaming-0.6b) | [csukuangfj2](https://huggingface.co/csukuangfj2/sherpa-onnx-nemotron-3.5-asr-streaming-0.6b-560ms-int8-2026-06-11) | [OpenMDW-1.1](https://openmdw.ai/license/1-1/) | | |
| | `sherpa-onnx-streaming-zipformer-en-kroko-2025-08-06` | 71 MB | [Banafo/Kroko-ASR](https://huggingface.co/Banafo/Kroko-ASR), English community model | [csukuangfj](https://huggingface.co/csukuangfj/sherpa-onnx-streaming-zipformer-en-kroko-2025-08-06) | CC BY-SA | | |
| Both are streaming transducers in the layout [sherpa-onnx](https://github.com/k2-fsa/sherpa-onnx) | |
| expects: an encoder, a decoder, a joiner and `tokens.txt`, with 8-bit quantized weights. | |
| - **Nemotron** is the more accurate, most of all on casual speech, and understands many | |
| languages (the language is set per stream, for example `en`, `ja` or `auto`). Its chunk size | |
| is 560 ms. | |
| - **Kroko** is English only, a tenth of the size and several times faster, which suits a phone. | |
| | File | SHA-256 | | |
| | --- | --- | | |
| | `sherpa-onnx-nemotron-…/encoder.int8.onnx` | `012e9321373af99021415e0b0eb3ec827b4be3153be6f30d9b448fe65e896e68` | | |
| | `sherpa-onnx-nemotron-…/decoder.int8.onnx` | `19f9c98fc6d0a2c33a65a43b36fdb2e914c26c0aa9764be3aebc502a1e982fb0` | | |
| | `sherpa-onnx-nemotron-…/joiner.int8.onnx` | `4101c7c679a0bc30483794b27a059e34e79232aa2068d78d51231a22c8b0d7ce` | | |
| | `sherpa-onnx-nemotron-…/tokens.txt` | `729cc103155bafa785f9cd45746cd41cabe97eab7182fc04d594129587958f8a` | | |
| | `sherpa-onnx-streaming-zipformer-en-kroko-…/encoder.onnx` | `d4881c57449d581e0770fd53fa66c2fdc6cd167d92ece7c715e603defc96d9d4` | | |
| | `sherpa-onnx-streaming-zipformer-en-kroko-…/decoder.onnx` | `455ba38466fce8d5a57e7db68a323b684079ca4d9e1dd93a740d9b2429aae3b1` | | |
| | `sherpa-onnx-streaming-zipformer-en-kroko-…/joiner.onnx` | `d406f616736350e2a7df3e39398b78eb2fc1a2ca6973a19d3853fa3227e25b52` | | |
| | `sherpa-onnx-streaming-zipformer-en-kroko-…/tokens.txt` | `396dbeb5f4858875690716084f54e90d339679d0ba3e6b5b584f3d7589254d2d` | | |