req and app initial

app.py

@@ -1,14 +1,37 @@
 import gradio as gr
 import spaces
 import torch
+import librosa
+import numpy as np
+from transformers import Wav2Vec2FeatureExtractor, Wav2Vec2ForSequenceClassification
 
-zero = torch.Tensor([0]).cuda()
-print(zero.device) # <-- 'cpu' 🤔
+
+
+device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
+
+model_name = "Hemg/human-emotion-detection"
+feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(model_name).to(device)
+model = Wav2Vec2ForSequenceClassification.from_pretrained(model_name).to(device) 
+
+def preprocess_audio(example):
+    audio_array, sampling_rate = librosa.load(audio_file_path, sr=16000)  # Load and resample to 16kHz
+    return {'speech': audio_array, 'sampling_rate': sampling_rate}
 
 @spaces.GPU
-def greet(n):
-    print(zero.device) # <-- 'cuda:0' 🤗
-    return f"Hello {zero + n} Tensor"
+def inference(audio):
+    example = preprocess_audio(audio_file_path)
+    inputs = feature_extractor(example['speech'], sampling_rate=16000, return_tensors="pt", padding=True)
+    inputs = inputs.to(device)  # Move inputs to GPU
+    with torch.no_grad():
+        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=predict_sentiment,
+                     inputs=gr.inputs.Audio(source="microphone", type="filepath"),
+                     outputs="text",
+                     title="Audio Sentiment Analysis",
+                     description="Upload an audio file or record one to analyze sentiment.")
+
 
-demo = gr.Interface(fn=greet, inputs=gr.Number(), outputs=gr.Text())
-demo.launch()
+iface.launch()

requirements.txt

@@ -0,0 +1,5 @@
+gradio
+torch
+transformers
+librosa
+numpy