AudioEffector Usages¶

This tutorial shows how to use torchaudio.io.AudioEffector to apply various effects and codecs to waveform tensor.

Note

This tutorial requires FFmpeg libraries. Please refer to FFmpeg dependency for the detail.

Overview¶

AudioEffector combines in-memory encoding, decoding and filtering that are provided by StreamWriter and StreamReader.

The following figure illustrates the process.

https://download.pytorch.org/torchaudio/tutorial-assets/AudioEffector.png

import torch
import torchaudio

print(torch.__version__)
print(torchaudio.__version__)

2.4.0
2.4.0

from torchaudio.io import AudioEffector, CodecConfig

import matplotlib.pyplot as plt
from IPython.display import Audio

for k, v in torchaudio.utils.ffmpeg_utils.get_versions().items():
    print(k, v)

libavcodec (60, 3, 100)
libavdevice (60, 1, 100)
libavfilter (9, 3, 100)
libavformat (60, 3, 100)
libavutil (58, 2, 100)

Usage¶

To use AudioEffector, instantiate it with effect and format, then either pass the waveform to apply() or stream() method.

effector = AudioEffector(effect=..., format=...,)

# Apply at once
applied = effector.apply(waveform, sample_rate)

apply method applies effect and codec to the entire waveform at once. So if the input waveform is long, and memory consumption is an issue, one can use stream method to process chunk by chunk.

# Apply chunk by chunk
for applied_chunk = effector.stream(waveform, sample_rate):
    ...

Example¶

src = torchaudio.utils.download_asset("tutorial-assets/Lab41-SRI-VOiCES-src-sp0307-ch127535-sg0042.wav")
waveform, sr = torchaudio.load(src, channels_first=False)

Gallery¶

def show(effect, *, stereo=False):
    wf = torch.cat([waveform] * 2, dim=1) if stereo else waveform
    figsize = (6.4, 2.1 if stereo else 1.2)

    effector = AudioEffector(effect=effect, pad_end=False)
    result = effector.apply(wf, int(sr))

    num_channels = result.size(1)
    f, ax = plt.subplots(num_channels, 1, squeeze=False, figsize=figsize, sharex=True)
    for i in range(num_channels):
        ax[i][0].specgram(result[:, i], Fs=sr)
    f.set_tight_layout(True)

    return Audio(result.numpy().T, rate=sr)