#!/usr/bin/env python3 import os import re from pathlib import Path from typing import List BASE_URL = "https://huggingface.co/csukuangfj/sherpa-onnx-apk/resolve/main/" from dataclasses import dataclass @dataclass class APK: major: int minor: int patch: int arch: str short_name: str def __init__(self, s): # sherpa-onnx-1.9.23-x86_64-asr-en-zipformer2.apk # sherpa-onnx-1.9.23-arm64-v8a-asr-zh-zipformer2.apk s = str(s)[len("asr/") :] split = s.split("-") self.major, self.minor, self.patch = list(map(int, split[2].split("."))) self.arch = split[3] self.lang = split[5] self.short_name = split[6] if "arm" in s: self.arch += "-" + split[4] self.lang = split[6] self.short_name = split[7] if "armeabi" in self.arch: self.arch = "y" + self.arch if "arm64" in self.arch: self.arch = "z" + self.arch if "small" in self.short_name: self.short_name = "zzz" + self.short_name def sort_by_apk(x): x = APK(x) return (x.major, x.minor, x.patch, x.arch, x.lang, x.short_name) def generate_url(files: List[str]) -> List[str]: ans = [] base = BASE_URL for f in files: ans.append(base + str(f)) return ans def get_all_files(d: str, suffix: str) -> List[str]: ans = sorted(Path(d).glob(suffix), key=sort_by_apk, reverse=True) return list(map(lambda x: BASE_URL + str(x), ans)) def to_file(filename: str, files: List[str]): content = r"""

APKs for streaming speech recognition

This page lists the streaming speech recognition APKs for sherpa-onnx, one of the deployment frameworks of the Next-gen Kaldi project.
The name of an APK has the following rule: where
You can download all supported models from https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models

APK Comment Model
sherpa-onnx-x.y.z-arm64-v8a-asr-bilingual_zh_en-zipformer.apk It supports both English and Chinese. sherpa-onnx-streaming-zipformer-bilingual-zh-en-2023-02-20.tar.bz2
sherpa-onnx-x.y.z-arm64-v8a-asr-en-nemo_ctc_80ms.apk It supports only English. It is converted from STT En FastConformer Hybrid Transducer-CTC Large Streaming 80ms from NVIDIA/NeMo. Note that only the CTC branch is used. sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-80ms.tar.bz2
sherpa-onnx-x.y.z-arm64-v8a-asr-en-nemo_ctc_480ms.apk It supports only English. It is converted from STT En FastConformer Hybrid Transducer-CTC Large Streaming 480ms from NVIDIA/NeMo. Note that only the CTC branch is used. sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-480ms.tar.bz2
sherpa-onnx-x.y.z-arm64-v8a-asr-en-nemo_ctc_1040ms.apk It supports only English. It is converted from STT En FastConformer Hybrid Transducer-CTC Large Streaming 1040ms from NVIDIA/NeMo. Note that only the CTC branch is used. sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-1040ms.tar.bz2
sherpa-onnx-x.y.z-arm64-v8a-asr-fr-zipformer.apk It supports only French. sherpa-onnx-streaming-zipformer-fr-2023-04-14.tar.bz2
sherpa-onnx-x.y.z-arm64-v8a-asr-zh-zipformer2.apk It supports only Chinese. icefall-asr-zipformer-streaming-wenetspeech-20230615.tar.bz2
sherpa-onnx-x.y.zz-arm64-v8a-asr-zh-small_zipformer.apk It supports only Chinese.
It uses the smallest zipformer and runs
super fast, though its accuracy is not that good.		 sherpa-onnx-streaming-zipformer-zh-14M-2023-02-23.tar.bz2
sherpa-onnx-x.y.z-arm64-v8a-asr-en-small_zipformer.apk It suppors only English.
It uses a very small zipformer and runs
super fast, though its accuracy is not that good.		 sherpa-onnx-streaming-zipformer-en-20M-2023-02-17.tar.bz2


""" if "-cn" not in filename: content += """ For Chinese users, please visit this address, which replaces huggingface.co with hf-mirror.com

中国用户, 请访问这个地址

""" with open(filename, "w") as f: print(content, file=f) for x in files: name = x.rsplit("/", maxsplit=1)[-1] print(f'{name}
', file=f) def main(): apk = get_all_files("asr", suffix="*.apk") to_file("./apk-asr.html", apk) # for Chinese users apk2 = [] for a in apk: a = a.replace("huggingface.co", "hf-mirror.com") a = a.replace("resolve", "blob") apk2.append(a) to_file("./apk-asr-cn.html", apk2) if __name__ == "__main__": main()