# Audio Perturbators

![Audio Perterbators](../images/insynth_audio_perturbators.png)
All audio-specific perturbators derive from the `BlackboxAudioPerturbator` class. They take a numpy array (audio signal)
and the sample rate as input and return a numpy array (perturbed audio signal).

## AudioBackgroundWhiteNoisePerturbator

The AudioBackgroundWhiteNoisePerturbator adds white noise to the input signal.

The probability distribution passed to the class during instantiation is used to determine the amount of noise added to the signal.

### Usage

```python
from insynth.perturbators.audio import AudioBackgroundWhiteNoisePerturbator
import librosa
from scipy.stats import norm

perturbator = AudioBackgroundWhiteNoisePerturbator(noise_prob=norm, noise_prob_args={'loc': 0.2, 'scale': 0.2})
audio, sample_rate = librosa.load('audio.wav')
perturbed_im = perturbator.apply((audio, sample_rate))
```

### Result

[Background White Noise](../audio/background_noise.wav)

## AudioCompressionPerturbator

The AudioCompressionPerturbator adds compression artifacts to the input signal.

The probability distribution passed to the class during instantiation is used to determine the amount of compression artifacts added to the signal.

### Usage

```python
from insynth.perturbators.audio import AudioCompressionPerturbator
import librosa
from scipy.stats import norm

perturbator = AudioCompressionPerturbator(compression_prob=norm, compression_prob_args={'loc': 80, 'scale': 40})
audio, sample_rate = librosa.load('audio.wav')
perturbed_im = perturbator.apply((audio, sample_rate))
```

### Result

[Compression](../audio/compression.wav)

## AudioPitchPerturbator

The AudioPitchPerturbator changes the pitch of the input signal.

The probability distribution passed to the class during instantiation is used to determine how much the pitch is changed.

### Usage

```python
from insynth.perturbators.audio import AudioPitchPerturbator
import librosa
from scipy.stats import norm

perturbator = AudioPitchPerturbator(pitch_prob=norm, pitch_prob_args={'loc': 0, 'scale': 8})
audio, sample_rate = librosa.load('audio.wav')
perturbed_im = perturbator.apply((audio, sample_rate))
```

### Result

[Pitch](../audio/pitch.wav)

## AudioClippingPerturbator

The AudioClippingPerturbator clips the input signal.

The probability distribution passed to the class during instantiation is used to determine how much the signal is clipped.

### Usage

```python
from insynth.perturbators.audio import AudioClippingPerturbator
import librosa
from scipy.stats import norm

perturbator = AudioClippingPerturbator(clipping_prob=norm, clipping_prob_args={'loc': 20, 'scale': 30})
audio, sample_rate = librosa.load('audio.wav')
perturbed_im = perturbator.apply((audio, sample_rate))
```

### Result

[Clipping](../audio/clipping.wav)

## AudioVolumePerturbator

The AudioVolumePerturbator changes the volume of the input signal.

The probability distribution passed to the class during instantiation is used to determine how much the volume is changed.

### Usage

```python
from insynth.perturbators.audio import AudioVolumePerturbator
import librosa
from scipy.stats import norm

perturbator = AudioVolumePerturbator(volume_prob=norm, volume_prob_args={'loc': 0, 'scale': 10})
audio, sample_rate = librosa.load('audio.wav')
perturbed_im = perturbator.apply((audio, sample_rate))
```

### Result

[Volume](../audio/volume.wav)

## AudioEchoPerturbator

The AudioEchoPerturbator adds echo to the input signal.

The probability distribution passed to the class during instantiation is used to determine how large the delay of the added echo is.

### Usage

```python
from insynth.perturbators.audio import AudioEchoPerturbator
import librosa
from scipy.stats import norm

perturbator = AudioEchoPerturbator(echo_prob=norm, echo_prob_args={'loc': 0.3, 'scale': 0.1})
audio, sample_rate = librosa.load('audio.wav')
perturbed_im = perturbator.apply((audio, sample_rate))
```

### Result

[Echo](../audio/echo.wav)

## AudioShortNoisePerturbator

The AudioShortNoisePerturbator adds short background noise to the input signal.

The `noise_types` parameter passed to the class during instantiation determines which noise types are added to the signal (chosen randomly).
They are used as the directory names to search noise files from under `data/audio/background_noise/`.

### Usage

```python
from insynth.perturbators.audio import AudioShortNoisePerturbator
import librosa

perturbator = AudioShortNoisePerturbator()
audio, sample_rate = librosa.load('audio.wav')
perturbed_im = perturbator.apply((audio, sample_rate))
```

### Result

[Short Noise](../audio/short_noise.wav)

## AudioBackgroundNoisePerturbator

The AudioBackgroundNoisePerturbator adds background noise to the input signal.

The `noise_types` parameter passed to the class during instantiation determines which noise types are added to the signal (chosen randomly).
They are used as the directory names to search noise files from under `data/audio/background_noise/`.

### Usage

```python
from insynth.perturbators.audio import AudioBackgroundNoisePerturbator
import librosa

perturbator = AudioBackgroundNoisePerturbator()
audio, sample_rate = librosa.load('audio.wav')
perturbed_im = perturbator.apply((audio, sample_rate))
```

### Result

[Background Noise](../audio/background_noise.wav)

## AudioImpulseResponsePerturbator

The AudioImpulseResponsePerturbator convolves the input signal with an impulse response.

The `impulse_types` parameter passed to the class during instantiation determines which impulse response types the signal is convolved with.
They are used as the directory names to search noise files from under `data/audio/pulse_response/`.

### Usage

```python
from insynth.perturbators.audio import AudioImpulseResponsePerturbator
import librosa

perturbator = AudioImpulseResponsePerturbator()
audio, sample_rate = librosa.load('audio.wav')
perturbed_im = perturbator.apply((audio, sample_rate))
```

### Result

[Impulse Response](../audio/impulse_response.wav)