如何在torchaudio中加载bytes类型的WAV音频对象？

DevOps 小满

2026-8-10

问题

我尝试将名为audio的bytes类型对象加载为torchaudio对象，代码如下：

def convert_audio(audio, target_sr: int = 16000): 
    wav, sr = torchaudio.load(audio) 
    #(...) 其他代码

我没找到torchaudio加载bytes音频对象的相关文档，它似乎只支持路径字符串。但我的应用需要节省IO操作，不能写入再读取.wav文件，只能直接处理音频对象。请问这种情况有什么解决办法吗？

直接使用audio时，出现如下错误：

Exception has occurred: AttributeError       (note: full exception trace is shown but execution is paused at: _run_module_as_main)
'bytes' object has no attribute 'seek'. You can only torch.load from a file that is seekable. Please pre-load the data into a buffer like io.BytesIO and try to load from it instead.
  File "/home/felipe/.local/lib/python3.10/site-packages/torch/serialization.py", line 348, in _check_seekable
    f.seek(f.tell())

使用BytesIO时：

Exception has occurred: UnpicklingError       (note: full exception trace is shown but execution is paused at: _run_module_as_main)
invalid load key, '\x00'.
  File "/home/felipe/.local/lib/python3.10/site-packages/torch/serialization.py", line 1002, in _legacy_load
    magic_number = pickle_module.load(f, **pickle_load_args)
  File "/home/felipe/.local/lib/python3.10/site-packages/torch/serialization.py", line 795, in load
    return _legacy_load(opened_file, map_location, pickle_module, **pickle_load_args)
  File "/home/felipe/Coding projects/silero/stt.py", line 35, in convert_audio
    wav,sr = torch.load(io.BytesIO(audio))
  File "/home/felipe/Coding projects/silero/stt.py", line 60, in transcribe
    input = prepare_model_input(convert_audio(audio),
  File "/home/felipe/Coding projects/silero/psgui.py", line 97, in <module>
    transcripton = stt.transcribe('en',audio)
  File "/usr/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/usr/lib/python3.10/runpy.py", line 196, in _run_module_as_main (Current frame)
    return _run_code(code, main_globals, None,

解决方案

核心问题是用错了方法：torch.load()是用来加载PyTorch序列化的模型或张量文件的，不是处理音频数据的工具。正确做法是用torchaudio.load()配合io.BytesIO包装bytes对象。
修正后的代码示例：

import io
import torchaudio

def convert_audio(audio, target_sr: int = 16000): 
    # 将bytes数据包装成可seek的类文件对象
    audio_buffer = io.BytesIO(audio)
    # 直接用torchaudio加载buffer
    wav, sr = torchaudio.load(audio_buffer) 
    # 按需重采样到目标采样率
    if sr != target_sr:
        wav = torchaudio.functional.resample(wav, sr, target_sr)
    return wav, target_sr

原理：torchaudio.load()本身支持传入可seek的类文件对象，把bytes包装成BytesIO后，就满足了读取要求，完全不需要写入本地文件，能节省IO开销。

内容的提问来源于stack exchange，提问作者plshelpmeout