Fix work on Windows

Author: istupakov

src/onnx_asr/asr.py

@@ -48,7 +48,7 @@ class _AsrWithDecoding(Asr):
 
     def __init__(self, preprocessor_name: str, vocab_path: Path, **kwargs: Any):
         self._preprocessor = Preprocessor(preprocessor_name, **kwargs)
-        with Path(vocab_path).open("rt") as f:
+        with Path(vocab_path).open("rt", encoding="utf-8") as f:
             tokens = {token: int(id) for token, id in (line.strip("\n").split(" ") for line in f.readlines())}
         self._vocab = {id: token for token, id in tokens.items()}
         self._blank_idx = tokens["<blk>"]

src/onnx_asr/models/whisper.py

@@ -31,16 +31,16 @@ def bytes_to_unicode() -> dict[int, str]:
 
 class _Whisper(Asr):
     def __init__(self, model_files: dict[str, Path], **kwargs: typing.Any):
-        with model_files["preprocessor_config"].open() as f:
+        with model_files["preprocessor_config"].open("rt", encoding="utf-8") as f:
             preprocessor_config = json.load(f)
 
         self._input_length = preprocessor_config["n_samples"]
         self._preprocessor = Preprocessor(f"whisper{preprocessor_config['feature_size']}", **kwargs)
 
-        with model_files["vocab"].open() as f:
+        with model_files["vocab"].open("rt", encoding="utf-8") as f:
             self._tokens: dict[str, int] = json.load(f)
 
-        with model_files["added_tokens"].open() as f:
+        with model_files["added_tokens"].open("rt", encoding="utf-8") as f:
             self._tokens |= json.load(f)
 
         self._vocab = {id: token for token, id in self._tokens.items()}

src/onnx_asr/utils.py

@@ -62,7 +62,7 @@ def read_wav_files(waveforms: list[npt.NDArray[np.float32] | str]) -> list[npt.N
 
 def pad_list(arrays: list[npt.NDArray[np.float32]], axis: int = 0) -> tuple[npt.NDArray[np.float32], npt.NDArray[np.int64]]:
     """Pad list of Numpy arrays to common length."""
-    lens = np.array([array.shape[axis] for array in arrays])
+    lens = np.array([array.shape[axis] for array in arrays], dtype=np.int64)
     max_len = lens.max()
 
     def pads(array: npt.NDArray[np.float32]) -> list[tuple[int, int]]: