Upload Cnn8RnnSoundEventDetection

config.json

@@ -7,6 +7,7 @@
     "AutoModel": "hf_model.Cnn8RnnSoundEventDetection"
   },
   "classes_num": 447,
+  "sample_rate": 32000,
   "torch_dtype": "float32",
   "transformers_version": "4.30.2"
 }

hf_model.py

@@ -118,9 +118,11 @@ class Cnn8RnnConfig(PretrainedConfig):
     def __init__(
         self,
         classes_num: int = 447,
+        sample_rate: int = 32000,
         **kwargs
     ):
         self.classes_num = classes_num
+        self.sample_rate = sample_rate
         super().__init__(**kwargs)
 
 
@@ -136,7 +138,7 @@ class Cnn8RnnSoundEventDetection(PreTrainedModel):
 
         # Logmel spectrogram extractor
         self.melspec_extractor = transforms.MelSpectrogram(
-            sample_rate=32000,
+            sample_rate=config.sample_rate,
             n_fft=1024,
             win_length=1024,
             hop_length=320,