Merge pull request #1207 from saumya0303/master

add resampling audio input transformation defense in torch

defenses/torch/audio/input_tranformation/resampling.py

@@ -0,0 +1,27 @@
+import torchaudio
+import librosa
+
+# There exist a limitation of this defense that it may lead to the problem of aliasing, and we can use the narrowband sample rate
+# rather than downsampling followed by upsampling.
+# resampling reference https://core.ac.uk/download/pdf/228298313.pdf
+# resampling input transformation defense for audio
+
+T = torchaudio.transforms
+
+# Read audio file
+audio_data = librosa.load(files, sr=16000)[0][-19456:]
+
+audio_data = torch.tensor(audio_data).float().to(device)
+
+# Discarding samples from a waveform during downsampling could remove a significant portion of the adversarial perturbation, thereby prevents an adversarial attack.
+
+# resample the audio files to 8kHz from 16kHz
+sample = T.Resample(16000, 8000, resampling_method="sinc_interpolation")
+
+audio_resample_1 = sample(audio_data)
+
+# resample the audio back to 16kHz
+sample = T.Resample(8000, 16000, resampling_method="sinc_interpolation")
+
+# Give audio_resample_2 as input to the asr model
+audio_resample_2 = sample(audio_resample_1)