improve(io): switch to torchaudio >= 2.2.0 (pyannote#1674)

fix: fixes pyannote#1576

CHANGELOG.md

@@ -22,6 +22,7 @@
 
 - improve(metric): add support for number of speakers mismatch in `diarization_error_rate` metric
 - improve(pipeline): track both `Model` and `nn.Module` attributes in `Pipeline.to(device)`
+- improve(io): switch to `torchaudio >= 2.2.0`
 
 ## Breaking changes
 

pyannote/audio/core/io.py

@@ -40,8 +40,6 @@
 from pyannote.core import Segment
 from torch import Tensor
 
-torchaudio.set_audio_backend("soundfile")
-
 AudioFile = Union[Text, Path, IOBase, Mapping]
 
 AudioFileDocString = """

pyannote/audio/pipelines/speaker_verification.py

@@ -28,7 +28,6 @@
 import numpy as np
 import torch
 import torch.nn.functional as F
-import torchaudio
 import torchaudio.compliance.kaldi as kaldi
 from huggingface_hub import hf_hub_download
 from huggingface_hub.utils import RepositoryNotFoundError
@@ -40,7 +39,6 @@
 from pyannote.audio.core.model import CACHE_DIR
 from pyannote.audio.pipelines.utils import PipelineModel, get_model
 
-backend = torchaudio.get_audio_backend()
 try:
     from speechbrain.pretrained import (
         EncoderClassifier as SpeechBrain_EncoderClassifier,
@@ -49,8 +47,6 @@
     SPEECHBRAIN_IS_AVAILABLE = True
 except ImportError:
     SPEECHBRAIN_IS_AVAILABLE = False
-finally:
-    torchaudio.set_audio_backend(backend)
 
 try:
     from nemo.collections.asr.models import (

pyannote/audio/tasks/segmentation/mixins.py

@@ -31,7 +31,7 @@
 from pyannote.database.protocol.protocol import Scope, Subset
 from pytorch_lightning.loggers import MLFlowLogger, TensorBoardLogger
 from torch.utils.data._utils.collate import default_collate
-from torchaudio.backend.common import AudioMetaData
+from torchaudio import AudioMetaData
 from torchmetrics import Metric
 from torchmetrics.classification import BinaryAUROC, MulticlassAUROC, MultilabelAUROC
 

requirements.txt

@@ -15,5 +15,5 @@ speechbrain >= 0.5.14
 tensorboardX >= 2.6
 torch >= 2.0.0
 torch_audiomentations >= 0.11.0
-torchaudio >= 2.0.0
+torchaudio >= 2.2.0
 torchmetrics >= 0.11.0