Fix boolean padding in ClapFeatureExtractor (#23648) (#49960)

Author: rudy-07

src/transformers/models/clap/feature_extraction_clap.py

@@ -263,7 +263,7 @@ def __call__(
         self,
         raw_speech: np.ndarray | list[float] | list[np.ndarray] | list[list[float]],
         truncation: str | None = None,
-        padding: str | None = None,
+        padding: bool | str | None = None,
         max_length: int | None = None,
         sampling_rate: int | None = None,
         return_tensors: str | TensorType | None = None,
@@ -284,8 +284,9 @@ def __call__(
                 If `config.fusion` is set to True, shorter audios also need to return 4 mels, which will just be a
                 copy of the original mel obtained from the padded audio.
                     - `rand_trunc` will select a random crop of the mel spectrogram.
-            padding (`str`, *optional*):
-               Padding pattern for shorter audio inputs. Three patterns were originally implemented:
+            padding (`bool`, `str`, *optional*):
+                Padding pattern for shorter audio inputs. If `True`, defaults to `self.padding`.
+                Available patterns:
                     - `repeatpad`: the audio is repeated, and then padded to fit the `max_length`.
                     - `repeat`: the audio is repeated and then cut to fit the `max_length`
                     - `pad`: the audio is padded.
@@ -299,7 +300,8 @@ def __call__(
                 pipeline.
         """
         truncation = truncation if truncation is not None else self.truncation
-        padding = padding if padding else self.padding
+        if padding is True or padding is None:
+            padding = self.padding
 
         if sampling_rate is not None:
             if sampling_rate != self.sampling_rate:

tests/models/clap/test_feature_extraction_clap.py

@@ -144,6 +144,20 @@ def test_double_precision_pad(self):
             pt_processed = feature_extractor.pad([{"input_features": inputs}], return_tensors="pt")
             self.assertTrue(pt_processed.input_features.dtype == torch.float32)
 
+    def test_padding_boolean(self):
+        feature_extractor = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict())
+        speech_inputs = floats_list((1, 800))[0]
+
+        np.random.seed(0)
+        out_default = feature_extractor(speech_inputs, return_tensors="np").input_features
+        np.random.seed(0)
+        out_bool = feature_extractor(speech_inputs, padding=True, return_tensors="np").input_features
+        np.random.seed(0)
+        out_repeatpad = feature_extractor(speech_inputs, padding="repeatpad", return_tensors="np").input_features
+
+        self.assertTrue(np.allclose(out_bool, out_default, atol=1e-3))
+        self.assertTrue(np.allclose(out_bool, out_repeatpad, atol=1e-3))
+
     # Copied from tests.models.whisper.test_feature_extraction_whisper.WhisperFeatureExtractionTest._load_datasamples
     def _load_datasamples(self, num_samples):
         ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")