torchaudio has no attribute 'AudioMetaData' (fix)

from pyannote.audio import Pipeline fails with AttributeError: module 'torchaudio' has no attribute 'AudioMetaData' when pyannote.audio 3.x meets torchaudio 2.9 or newer. Tested fixes: pin torch==2.8.0 and torchaudio==2.8.0, or upgrade to pyannote.audio 4.

1. The error

A fresh environment with pyannote.audio 3.3.2. Its metadata requires only torchaudio>=2.2.0, so pip is free to pick torchaudio 2.9 or newer. We pinned 2.9.0 to show the first failing release:

pip install "pyannote.audio==3.3.2" "torch==2.9.0" "torchaudio==2.9.0"

import torch, torchaudio, pyannote.audio
print("torch", torch.__version__, "| torchaudio", torchaudio.__version__, "| pyannote.audio", pyannote.audio.__version__)
from pyannote.audio import Model, Pipeline
print("import OK")

Traceback (most recent call last):
  ...
    from .core.inference import Inference
  File ".../site-packages/pyannote/audio/core/inference.py", line 35, in <module>
    from pyannote.audio.core.io import AudioFile
  File ".../site-packages/pyannote/audio/core/io.py", line 60, in <module>
    ) -> torchaudio.AudioMetaData:
         ^^^^^^^^^^^^^^^^^^^^^^^^
AttributeError: module 'torchaudio' has no attribute 'AudioMetaData'

pyannote.audio 3.x uses torchaudio.AudioMetaData as a type hint in pyannote/audio/core/io.py, so the import itself fails before any model is loaded. In our test hasattr(torchaudio, "AudioMetaData") is True on torchaudio 2.8.0 and False on 2.9.0.

2. Fix: pin torch and torchaudio to 2.8

Keep pyannote.audio 3.x and pin the pair together (torch and torchaudio versions must match):

pip install "pyannote.audio==3.3.2" "torch==2.8.0" "torchaudio==2.8.0"

import torch, torchaudio, pyannote.audio
print("torch", torch.__version__, "| torchaudio", torchaudio.__version__, "| pyannote.audio", pyannote.audio.__version__)
from pyannote.audio import Model, Pipeline
print("import OK")

.../site-packages/pyannote/audio/core/io.py:212: UserWarning: torchaudio._backend.list_audio_backends has been deprecated. This deprecation is part of a large refactoring effort to transition TorchAudio into a maintenance phase. The decoding and encoding capabilities of PyTorch for both audio and video are being consolidated into TorchCodec. Please see https://github.com/pytorch/audio/issues/3902 for more information. It will be removed from the 2.9 release. 
  torchaudio.list_audio_backends()
torch 2.8.0+cpu | torchaudio 2.8.0+cpu | pyannote.audio 3.3.2
import OK

The import works; the deprecation warning is torchaudio announcing the same refactor.

3. Fix: upgrade to pyannote.audio 4

pyannote.audio 4.x dropped the old torchaudio I/O and depends on torchcodec. On torch 2.9.0:

pip install "pyannote.audio==4.0.7"

import torch, torchaudio, pyannote.audio
print("torch", torch.__version__, "| torchaudio", torchaudio.__version__, "| pyannote.audio", pyannote.audio.__version__)
from pyannote.audio import Model, Pipeline
print("import OK")

.../site-packages/pyannote/audio/core/io.py:48: UserWarning: 
torchcodec is not installed correctly so built-in audio decoding will fail. Solutions are:
* use audio preloaded in-memory as a {'waveform': (channel, time) torch.Tensor, 'sample_rate': int} dictionary;
* fix torchcodec installation. Error message was:

Could not load this library: .../site-packages/torchcodec/libtorchcodec_image.so
  warnings.warn(
torch 2.9.0+cpu | torchaudio 2.9.0+cpu | pyannote.audio 4.0.7
import OK

The import works. The warning is from our test box, which has no FFmpeg installed: TorchCodec needs the FFmpeg libraries to decode audio files. Either install FFmpeg, or pass audio already in memory as {"waveform": tensor, "sample_rate": sr}, as the warning says. pyannote.audio 4 is a major version, so check its release notes before switching a working 3.x pipeline.

Related: torchaudio has no attribute info · Could not load libtorchcodec · torch.load: Weights only load failed · ultralytics YOLO: Weights only load failed · No module named torchaudio.backend · No module named torchaudio.io

Written with AI assistance. Measured on 2026-10-24 on Linux with CPython 3.11.17, CPU wheels of torch/torchaudio 2.8.0 and 2.9.0, pyannote.audio 3.3.2 and 4.0.7. Output pasted unedited except shortened site-packages paths and a cut traceback. PlainTables is our extension; pyannote.audio and PyTorch are free.