Cache audio triggers

art/attacks/poisoning/perturbations/audio_perturbations.py

@@ -16,136 +16,110 @@
 # TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 # SOFTWARE.
 """
-Adversarial perturbations designed to work for images.
+Adversarial perturbations designed to work for audio.
+Uses classes, rather than pure functions as in image_perturbations.py,
+because loading the audio trigger from disk (librosa.load()) is very slow
+and should be done only once.
 """
 import numpy as np
 import librosa
 
 
-def insert_tone_trigger(
-    x: np.ndarray,
-    sampling_rate: int = 16000,
-    frequency: int = 440,
-    duration: float = 0.1,
-    random: bool = False,
-    shift: int = 0,
-    scale: float = 0.1,
-) -> np.ndarray:
+class CacheTrigger:
     """
-    Adds a 'tone' with a given frequency to audio example. Works for a single example or a batch of examples.
+    Adds an audio backdoor trigger to a set of audio examples. Works for a single example or a batch of examples.
 
-    :param x: N x L matrix or length L array, where N is number of examples, L is the length in number of samples.
-              X is in range [-1,1].
-    :param sampling_rate: Positive integer denoting the sampling rate for x.
-    :param frequency: Frequency of the tone to be added.
-    :param duration: Duration of the tone to be added.
+    :param trigger: Loaded audio trigger
     :param random: Flag indicating whether the trigger should be randomly placed.
     :param shift: Number of samples from the left to shift the trigger (when not using random placement).
     :param scale: Scaling factor for mixing the trigger.
     :return: Backdoored audio.
     """
-    n_dim = len(x.shape)
-    if n_dim > 2:
-        raise ValueError("Invalid array shape " + str(x.shape))
-
-    if n_dim == 2:
-        return np.array(
-            [
-                insert_tone_trigger(single_audio, sampling_rate, frequency, duration, random, shift, scale)
-                for single_audio in x
-            ]
-        )
-
-    original_dtype = x.dtype
-    audio = np.copy(x)
-    length = audio.shape[0]
-
-    tone_trigger = librosa.tone(frequency, sr=sampling_rate, duration=duration)
-
-    bd_length = tone_trigger.shape[0]
-    if bd_length > length:
-        print("audio shape:", audio.shape)
-        print("trigger shape:", tone_trigger.shape)
-        raise ValueError("Backdoor audio does not fit inside the original audio.")
-
-    if random:
-        shift = np.random.randint(length - bd_length)
 
-    if shift + bd_length > length:
-        raise ValueError("Shift + Backdoor length is greater than audio's length.")
-
-    trigger_shifted = np.zeros_like(audio)
-    trigger_shifted[shift : shift + bd_length] = np.copy(tone_trigger)
-
-    audio += scale * trigger_shifted
-    audio = np.clip(audio, -1.0, 1.0)
-
-    return audio.astype(original_dtype)
-
-
-def insert_audio_trigger(
-    x: np.ndarray,
-    sampling_rate: int = 16000,
-    backdoor_path: str = "../../../utils/data/backdoors/cough_trigger.wav",
-    duration: float = 1.0,
-    random: bool = False,
-    shift: int = 0,
-    scale: float = 0.1,
-) -> np.ndarray:
+    def __init__(
+        self,
+        trigger: np.ndarray,
+        random: bool = False,
+        shift: int = 0,
+        scale: float = 0.1,
+    ):
+        self.trigger = trigger
+        self.scaled_trigger = self.trigger * scale
+        self.random = random
+        self.shift = shift
+        self.scale = scale
+
+    def insert(self, x: np.ndarray) -> np.ndarray:
+        """
+        :param x: N x L matrix or length L array, where N is number of examples, L is the length in number of samples.
+                  X is in range [-1,1].
+        :return: Backdoored audio.
+        """
+        n_dim = len(x.shape)
+        if n_dim == 2:
+            return np.array([self.insert(single_audio) for single_audio in x])
+        if n_dim != 1:
+            raise ValueError("Invalid array shape " + str(x.shape))
+        original_dtype = x.dtype
+        audio = np.copy(x)
+        length = audio.shape[0]
+        bd_length = self.trigger.shape[0]
+        if bd_length > length:
+            raise ValueError("Backdoor audio does not fit inside the original audio.")
+        if self.random:
+            shift = np.random.randint(length - bd_length)
+        else:
+            shift = self.shift
+        if shift + bd_length > length:
+            raise ValueError("Shift + Backdoor length is greater than audio's length.")
+
+        audio[shift : shift + bd_length] += self.scaled_trigger
+        audio = np.clip(audio, -1.0, 1.0)
+        return audio.astype(original_dtype)
+
+
+class CacheAudioTrigger(CacheTrigger):
     """
     Adds an audio backdoor trigger to a set of audio examples. Works for a single example or a batch of examples.
 
-    :param x: N x L matrix or length L array, where N is number of examples, L is the length in number of samples.
-              X is in range [-1,1].
     :param sampling_rate: Positive integer denoting the sampling rate for x.
     :param backdoor_path: The path to the audio to insert as a trigger.
     :param duration: Duration of the trigger in seconds. Default `None` if full trigger is to be used.
-    :param random: Flag indicating whether the trigger should be randomly placed.
-    :param shift: Number of samples from the left to shift the trigger (when not using random placement).
-    :param scale: Scaling factor for mixing the trigger.
-    :return: Backdoored audio.
     """
-    n_dim = len(x.shape)
-    if n_dim > 2:
-        raise ValueError("Invalid array shape " + str(x.shape))
-
-    if n_dim == 2:
-        return np.array(
-            [
-                insert_audio_trigger(single_audio, sampling_rate, backdoor_path, duration, random, shift, scale)
-                for single_audio in x
-            ]
-        )
-
-    original_dtype = x.dtype
-    audio = np.copy(x)
-
-    length = audio.shape[0]
-
-    trigger, bd_sampling_rate = librosa.load(backdoor_path, mono=True, sr=None, duration=duration)
-
-    if sampling_rate != bd_sampling_rate:
-        print(
-            "Backdoor sampling rate does not match with the sampling rate provided. "
-            "Resampling the backdoor to match the sampling rate."
-        )
-        trigger, _ = librosa.load(backdoor_path, mono=True, sr=sampling_rate, duration=duration)
-
-    bd_length = trigger.shape[0]
-
-    if bd_length > length:
-        raise ValueError("Backdoor audio does not fit inside the original audio.")
 
-    if random:
-        shift = np.random.randint(length - bd_length)
-
-    if shift + bd_length > length:
-        raise ValueError("Shift + Backdoor length is greater than audio's length.")
-
-    trigger_shifted = np.zeros_like(audio)
-    trigger_shifted[shift : shift + bd_length] = np.copy(trigger)
+    def __init__(
+        self,
+        sampling_rate: int = 16000,
+        backdoor_path: str = "../../../utils/data/backdoors/cough_trigger.wav",
+        duration: float = None,
+        **kwargs,
+    ):
+        trigger, bd_sampling_rate = librosa.load(backdoor_path, mono=True, sr=None, duration=duration)
+
+        if sampling_rate != bd_sampling_rate:
+            print(
+                f"Backdoor sampling rate {bd_sampling_rate} does not match with the sampling rate provided."
+                "Resampling the backdoor to match the sampling rate."
+            )
+            trigger, _ = librosa.load(backdoor_path, mono=True, sr=sampling_rate, duration=duration)
+        super().__init__(trigger, **kwargs)
+
+
+class CacheToneTrigger(CacheTrigger):
+    """
+    Adds an audio backdoor trigger to a set of audio examples. Works for a single example or a batch of examples.
 
-    audio += scale * trigger_shifted
-    audio = np.clip(audio, -1.0, 1.0)
+    :param sampling_rate: Positive integer denoting the sampling rate for x.
+    :param frequency: Frequency of the tone to be added.
+    :param duration: Duration of the tone to be added.
+    """
 
-    return audio.astype(original_dtype)
+    def __init__(
+        self,
+        sampling_rate: int = 16000,
+        frequency: int = 440,
+        duration: float = 0.1,
+        **kwargs,
+    ):
+        trigger = librosa.tone(frequency, sr=sampling_rate, duration=duration)
+        super().__init__(trigger, **kwargs)

tests/attacks/poison/test_audio_perturbations.py

@@ -22,7 +22,7 @@
 import pytest
 import os
 
-from art.attacks.poisoning.perturbations.audio_perturbations import insert_tone_trigger, insert_audio_trigger
+from art.attacks.poisoning.perturbations.audio_perturbations import CacheToneTrigger, CacheAudioTrigger
 
 from tests.utils import ARTTestException
 
@@ -33,39 +33,45 @@
 def test_insert_tone_trigger(art_warning):
     try:
         # test single example
-        audio = insert_tone_trigger(x=np.zeros(3200), sampling_rate=16000)
+        trigger = CacheToneTrigger(sampling_rate=16000)
+        audio = trigger.insert(x=np.zeros(3200))
         assert audio.shape == (3200,)
         assert np.max(audio) != 0
         assert np.max(np.abs(audio)) <= 1.0
 
         # test single example with differet duration, frequency, and scale
-        audio = insert_tone_trigger(x=np.zeros(3200), sampling_rate=16000, frequency=16000, duration=0.2, scale=0.5)
+        trigger = CacheToneTrigger(sampling_rate=16000, frequency=16000, duration=0.2, scale=0.5)
+        audio = trigger.insert(x=np.zeros(3200))
         assert audio.shape == (3200,)
         assert np.max(audio) != 0
 
         # test a batch of examples
-        audio = insert_tone_trigger(x=np.zeros((10, 3200)), sampling_rate=16000)
+        audio = trigger.insert(x=np.zeros((10, 3200)))
         assert audio.shape == (10, 3200)
         assert np.max(audio) != 0
 
         # test single example with shift
-        audio = insert_tone_trigger(x=np.zeros(3200), sampling_rate=16000, shift=10)
+        trigger = CacheToneTrigger(sampling_rate=16000, shift=10)
+        audio = trigger.insert(x=np.zeros(3200))
         assert audio.shape == (3200,)
         assert np.max(audio) != 0
         assert np.sum(audio[:10]) == 0
 
         # test a batch of examples with random shift
-        audio = insert_tone_trigger(x=np.zeros((10, 3200)), sampling_rate=16000, random=True)
+        trigger = CacheToneTrigger(sampling_rate=16000, random=True)
+        audio = trigger.insert(x=np.zeros((10, 3200)))
         assert audio.shape == (10, 3200)
         assert np.max(audio) != 0
 
         # test when length of backdoor is larger than that of audio signal
         with pytest.raises(ValueError):
-            _ = insert_tone_trigger(x=np.zeros(3200), sampling_rate=16000, duration=0.3)
+            trigger = CacheToneTrigger(sampling_rate=16000, duration=0.3)
+            _ = trigger.insert(x=np.zeros(3200))
 
         # test when shift + backdoor is larger than that of audio signal
         with pytest.raises(ValueError):
-            _ = insert_tone_trigger(x=np.zeros(3200), sampling_rate=16000, duration=0.2, shift=5)
+            trigger = CacheToneTrigger(sampling_rate=16000, duration=0.2, shift=5)
+            _ = trigger.insert(x=np.zeros(3200))
 
     except ARTTestException as e:
         art_warning(e)
@@ -76,56 +82,57 @@ def test_insert_audio_trigger(art_warning):
     file_path = os.path.join(os.getcwd(), "utils/data/backdoors/cough_trigger.wav")
     try:
         # test single example
-        audio = insert_audio_trigger(x=np.zeros(32000), sampling_rate=16000, backdoor_path=file_path)
+        trigger = CacheAudioTrigger(sampling_rate=16000, backdoor_path=file_path)
+        audio = trigger.insert(x=np.zeros(32000))
         assert audio.shape == (32000,)
         assert np.max(audio) != 0
         assert np.max(np.abs(audio)) <= 1.0
 
         # test single example with differet duration and scale
-        audio = insert_audio_trigger(
-            x=np.zeros(32000),
+        trigger = CacheAudioTrigger(
             sampling_rate=16000,
             backdoor_path=file_path,
             duration=0.8,
             scale=0.5,
         )
+        audio = trigger.insert(x=np.zeros(32000))
         assert audio.shape == (32000,)
         assert np.max(audio) != 0
 
         # test a batch of examples
-        audio = insert_audio_trigger(x=np.zeros((10, 16000)), sampling_rate=16000, backdoor_path=file_path)
+        trigger = CacheAudioTrigger(sampling_rate=16000, backdoor_path=file_path)
+        audio = trigger.insert(x=np.zeros((10, 16000)))
+
         assert audio.shape == (10, 16000)
         assert np.max(audio) != 0
 
         # test single example with shift
-        audio = insert_audio_trigger(x=np.zeros(32000), sampling_rate=16000, backdoor_path=file_path, shift=10)
+        trigger = CacheAudioTrigger(sampling_rate=16000, backdoor_path=file_path, shift=10)
+        audio = trigger.insert(x=np.zeros(32000))
         assert audio.shape == (32000,)
         assert np.max(audio) != 0
         assert np.sum(audio[:10]) == 0
 
         # test a batch of examples with random shift
-        audio = insert_audio_trigger(
-            x=np.zeros((10, 32000)),
-            sampling_rate=16000,
-            backdoor_path=file_path,
-            random=True,
-        )
+        trigger = CacheAudioTrigger(sampling_rate=16000, backdoor_path=file_path, random=True)
+        audio = trigger.insert(x=np.zeros((10, 32000)))
         assert audio.shape == (10, 32000)
         assert np.max(audio) != 0
 
         # test when length of backdoor is larger than that of audio signal
         with pytest.raises(ValueError):
-            _ = insert_audio_trigger(x=np.zeros(15000), sampling_rate=16000, backdoor_path=file_path)
+            trigger = CacheAudioTrigger(sampling_rate=16000, backdoor_path=file_path)
+            _ = trigger.insert(x=np.zeros(15000))
 
         # test when shift + backdoor is larger than that of audio signal
         with pytest.raises(ValueError):
-            _ = insert_audio_trigger(
-                x=np.zeros(16000),
+            trigger = CacheAudioTrigger(
                 sampling_rate=16000,
                 backdoor_path=file_path,
                 duration=1,
                 shift=5,
             )
+            _ = trigger.insert(x=np.zeros(16000))
 
     except ARTTestException as e:
         art_warning(e)