testing_1

app.py

@@ -1,8 +1,7 @@
 import gradio as gr
 import spaces
 import torch
-#import librosa
-#import numpy as np
+import torchaudio
 from transformers import Wav2Vec2FeatureExtractor, Wav2Vec2ForSequenceClassification
 
 device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
@@ -13,13 +12,14 @@ model = Wav2Vec2ForSequenceClassification.from_pretrained(model_name).to(device)
 
 def preprocess_audio(audio):
     print('hallo')
-    #audio_array, sampling_rate = librosa.load(audio, sr=16000)  # Load and resample to 16kHz
-    #return {'speech': audio_array, 'sampling_rate': sampling_rate}
+    waveform, sampling_rate = torchaudio.load(audio)
+    resampled_waveform = torchaudio.transforms.Resample(orig_freq=sampling_rate, new_freq=16000)(waveform)
+    return {'speech': resampled_waveform.numpy().flatten(), 'sampling_rate': 16000}
 
 @spaces.GPU
 def inference(audio):
     print('hello')
-    '''
+    
     example = preprocess_audio(audio)
     inputs = feature_extractor(example['speech'], sampling_rate=16000, return_tensors="pt", padding=True)
     inputs = inputs.to(device)  # Move inputs to GPU
@@ -27,7 +27,7 @@ def inference(audio):
         logits = model(**inputs).logits
     predicted_ids = torch.argmax(logits, dim=-1)
     return model.config.id2label[predicted_ids.item()], logits, predicted_ids   # Move tensors back to CPU for further processing
-    '''
+    
 
 iface = gr.Interface(fn=inference,
                      inputs=gr.Audio(type="filepath"),