feat: add support for file-like objects (#640)

This commit is contained in:
Harry Coultas Blum authored and GitHub committed 2021-04-12 11:31:13 +02:00
1 parent 37a6760738
commit 3c4d0a68a3
2 files changed
+119 -95

No files matched your search

+95 -81
View File
@@ -30,10 +30,12 @@ We should switch torchaudio resampling as well at some point...
import math
import warnings
from io import IOBase
from pathlib import Path
from typing import Mapping, Optional, Text, Tuple, Union
import librosa
import numpy as np
import torch
import torchaudio
from torch import Tensor
@@ -42,25 +44,22 @@ from pyannote.core import Segment
from pyannote.core.utils.types import Alignment
from pyannote.database import ProtocolFile
AudioFile = Union[Path, Text, ProtocolFile, dict]
torchaudio.set_audio_backend("soundfile")
"""
AudioFile = Union[Text, Path, IOBase, Mapping]
AudioFileDocString = """
Audio files can be provided to the Audio class using different types:
- a "str" instance: "/path/to/audio.wav"
- a "Path" instance: Path("/path/to/audio.wav")
- a ProtocolFile (or regular dict) with an "audio" key:
{"audio": Path("/path/to/audio.wav")}
- a ProtocolFile (or regular dict) with both "waveform" and "sample_rate" key:
- a "str" or "Path" instance: "audio.wav" or Path("audio.wav")
- a "IOBase" instance with "read" and "seek" support: open("audio.wav", "rb")
- a "Mapping" with any of the above as "audio" key: {"audio": ...}
- a "Mapping" with both "waveform" and "sample_rate" key:
{"waveform": (channel, time) numpy.ndarray or torch.Tensor, "sample_rate": 44100}
For last two options, an additional "channel" key can be provided as a zero-indexed
integer to load a specific channel:
{"audio": Path("/path/to/stereo.wav"), "channel": 0}
integer to load a specific channel: {"audio": "stereo.wav", "channel": 0}
"""
torchaudio.USE_SOUNDFILE_LEGACY_INTERFACE = False
torchaudio.set_audio_backend("soundfile")
class Audio:
"""Audio IO
@@ -105,92 +104,69 @@ class Audio:
return (waveform.t() / (rms + 1e-8)).t()
@staticmethod
def get_duration(file: AudioFile) -> float:
"""Get audio file duration in seconds
def validate_file(file: AudioFile) -> Mapping:
"""Validate file for use with the other Audio methods
Parameters
----------
file : AudioFile
Audio file.
Parameter
---------
file: AudioFile
Returns
-------
duration : float
Duration in seconds.
validated_file : Mapping
{"audio": str, "uri": str, ...}
{"waveform": array or tensor, "sample_rate": int, "uri": str, ...}
{"audio": file, "uri": "stream"} if `file` is an IOBase instance
Raises
------
ValueError if file format is not valid or file does not exist.
"""
if isinstance(file, (ProtocolFile, dict)):
audio = file["audio"]
else:
audio = file
if isinstance(audio, Path):
audio = str(audio)
info = torchaudio.info(audio)
return info.num_frames / info.sample_rate
@staticmethod
def validate_file(file: AudioFile) -> Union[Mapping, ProtocolFile]:
# TODO: add support for file-like object
# see https://github.com/pyannote/pyannote-audio/issues/564
if isinstance(file, str):
file = Path(file)
if isinstance(file, Path):
file = {"audio": file}
if isinstance(file, Mapping):
pass
if "waveform" in file:
elif isinstance(file, (str, Path)):
file = {"audio": str(file), "uri": Path(file).stem}
waveform = file["waveform"]
if len(waveform.shape) != 2 or waveform.shape[0] > waveform.shape[1]:
raise ValueError(
"'waveform' must be provided as a (channel, time) torch Tensor."
)
elif isinstance(file, IOBase):
return {"audio": file, "uri": "stream"}
sample_rate = file.get("sample_rate", None)
if sample_rate is None:
raise ValueError(
"'waveform' must be provided with their 'sample_rate'."
)
else:
raise ValueError(AudioFileDocString)
if "uri" not in file:
file["uri"] = "waveform"
if "waveform" in file:
waveform: Union[np.ndarray, Tensor] = file["waveform"]
if len(waveform.shape) != 2 or waveform.shape[0] > waveform.shape[1]:
raise ValueError(
"'waveform' must be provided as a (channel, time) torch Tensor."
)
sample_rate: int = file.get("sample_rate", None)
if sample_rate is None:
raise ValueError(
"'waveform' must be provided with their 'sample_rate'."
)
file.setdefault("uri", "waveform")
elif "audio" in file:
if isinstance(file["audio"], IOBase):
return file
if "audio" in file:
path = Path(file["audio"])
if not path.is_file():
raise ValueError(f"File {path} does not exist")
path = Path(file["audio"])
if not path.is_file():
raise ValueError(f"File {path} does not exist")
file.setdefault("uri", path.stem)
if "uri" not in file:
file["uri"] = path.stem
return file
raise ValueError(
"""
Audio files can be provided using different types:
- a "str" instance: "/path/to/audio.wav"
- a "Path" instance: Path("/path/to/audio.wav")
- a ProtocolFile (or regular dict) with an "audio" key:
{"audio": Path("/path/to/audio.wav")}
- a ProtocolFile (or regular dict) with both "waveform" and "sample_rate" key:
{"waveform": (channel, time) numpy.ndarray or torch.Tensor, "sample_rate": 44100}
For last two options, an additional "channel" key can be provided as a zero-indexed
integer to load a specific channel:
{"audio": Path("/path/to/stereo.wav"), "channel": 0}
"""
)
return file
def __init__(self, sample_rate=None, mono=True):
super().__init__()
self.sample_rate = sample_rate
self.mono = mono
@@ -231,8 +207,35 @@ class Audio:
).T
sample_rate = self.sample_rate
waveform = torch.tensor(waveform)
return waveform, sample_rate
def get_duration(self, file: AudioFile) -> float:
"""Get audio file duration in seconds
Parameters
----------
file : AudioFile
Audio file.
Returns
-------
duration : float
Duration in seconds.
"""
file = self.validate_file(file)
if "waveform" in file:
return len(file["waveform"].T / file["sample_rate"])
info = torchaudio.info(file["audio"])
if isinstance(file["audio"], IOBase):
file["audio"].seek(0)
return info.num_frames / info.sample_rate
def __call__(self, file: AudioFile) -> Tuple[Tensor, int]:
"""Obtain waveform
@@ -317,6 +320,9 @@ class Audio:
sample_rate = info.sample_rate
frames = info.num_frames
if isinstance(file["audio"], IOBase):
file["audio"].seek(0)
channel = file.get("channel", None)
# infer which samples to load from sample rate and requested chunk
@@ -368,13 +374,21 @@ class Audio:
data, _ = torchaudio.load(
file["audio"], frame_offset=start_frame, num_frames=num_frames
)
if isinstance(file["audio"], IOBase):
file["audio"].seek(0)
except RuntimeError:
if isinstance(file["audio"], IOBase):
msg = "torchaudio failed to seek-and-read in file-like object."
raise RuntimeError(msg)
msg = (
f"torchaudio failed to seek-and-read in {file['audio']}: "
f"loading the whole file instead."
)
warnings.warn(msg)
waveform, sample_rate = self(file)
waveform, sample_rate = self.__call__(file)
data = waveform[:, start_frame:end_frame]
# storing waveform and sample_rate for next time
+24 -14
View File
@@ -1,11 +1,9 @@
import random
import torch
import torchaudio
from torch import Tensor
from pyannote.audio.core.io import Audio
from pyannote.core import Segment, SlidingWindow
from pyannote.core import Segment
def test_audio_resample():
@@ -63,15 +61,27 @@ def test_can_crop_waveform():
assert sr == 16000
def test_crops_are_correct_shape():
sr = 160001
secs = random.randint(5, 11)
waveform = torch.randn(1, secs * sr)
# File Like Object Tests
def test_can_load_from_file_like():
"Load entire wav of file like"
loader = Audio()
shape = None
for segment in SlidingWindow(end=secs):
wav, sr = loader.crop({"waveform": waveform, "sample_rate": sr}, segment)
if shape is None:
shape = wav.shape
else:
assert shape == wav.shape
with open("tests/data/dev00.wav", "rb") as f:
wav, sr = loader(f)
assert isinstance(wav, Tensor)
assert sr == 16000
def test_can_crop_from_file_like():
"Load cropped sections from file like objects"
loader = Audio()
with open("tests/data/dev00.wav", "rb") as f:
segment = Segment(0.2, 0.7)
wav, sr = loader.crop(f, segment)
assert isinstance(wav, Tensor)
assert sr == 16000
assert wav.shape[1] == 0.5 * 16000