Skip to content

Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.

允许 model.generate 使用bytes io, 以便不写入文件,节省io时间 (#2343)

maliubiao172a3152b42a
diff --git a/funasr/utils/load_utils.py b/funasr/utils/load_utils.py
index 1d80fcf36..072b578ef 100644
--- a/funasr/utils/load_utils.py
+++ b/funasr/utils/load_utils.py
@@ -86,8 +86,10 @@ def load_audio_text_image_video(
): # download url to local file
data_or_path_or_list = download_from_url(data_or_path_or_list)
- if isinstance(data_or_path_or_list, str) and os.path.exists(data_or_path_or_list): # local file
+ if (isinstance(data_or_path_or_list, str) and os.path.exists(data_or_path_or_list)) or hasattr(data_or_path_or_list, 'read'): # local file or bytes io
if data_type is None or data_type == "sound":
+ if hasattr(data_or_path_or_list, "read") and hasattr(data_or_path_or_list, "seek"):
+ data_or_path_or_list.seek(0)
# if use_ffmpeg:
# data_or_path_or_list = _load_audio_ffmpeg(data_or_path_or_list, sr=fs)
# data_or_path_or_list = torch.from_numpy(data_or_path_or_list).squeeze() # [n_samples,]