Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| import torch | |
| import librosa | |
| import numpy as np | |
| import spaces | |
| from transformers import ASTFeatureExtractor, ASTForAudioClassification | |
| # ============================================================ | |
| # LOAD MODEL | |
| # ============================================================ | |
| MODEL_PATH = "." | |
| print("Loading AST model...") | |
| feature_extractor = ASTFeatureExtractor.from_pretrained(MODEL_PATH) | |
| model = ASTForAudioClassification.from_pretrained(MODEL_PATH) | |
| model.eval() | |
| print("Model loaded successfully!") | |
| print(model.config.id2label) | |
| # ============================================================ | |
| # KONSTANTA | |
| # ============================================================ | |
| SAMPLE_RATE = 16000 | |
| CLIP_DURATION = 3 | |
| MAX_SAMPLES = SAMPLE_RATE * CLIP_DURATION | |
| # ============================================================ | |
| # PREPROCESSING | |
| # ============================================================ | |
| def preprocess_audio(audio_path): | |
| audio, sr = librosa.load(audio_path, sr=SAMPLE_RATE, mono=True) | |
| # Normalisasi | |
| max_val = np.max(np.abs(audio)) | |
| if max_val > 1e-8: | |
| audio = audio / max_val * 0.95 | |
| # Potong 3 detik | |
| audio = audio[:MAX_SAMPLES] | |
| # Padding jika kurang | |
| if len(audio) < MAX_SAMPLES: | |
| audio = np.pad(audio, (0, MAX_SAMPLES - len(audio))) | |
| return audio | |
| # ============================================================ | |
| # PREDIKSI | |
| # ============================================================ | |
| def prediksi(audio_path): | |
| if audio_path is None: | |
| return {"Tidak ada audio": 1.0} | |
| try: | |
| audio = preprocess_audio(audio_path) | |
| inputs = feature_extractor( | |
| audio, | |
| sampling_rate=SAMPLE_RATE, | |
| return_tensors="pt" | |
| ) | |
| device = "cuda" if torch.cuda.is_available() else "cpu" | |
| model.to(device) | |
| inputs = {k: v.to(device) for k, v in inputs.items()} | |
| with torch.no_grad(): | |
| outputs = model(**inputs) | |
| probs = torch.softmax(outputs.logits, dim=-1)[0] | |
| hasil = { | |
| model.config.id2label[i]: float(probs[i]) | |
| for i in range(len(probs)) | |
| } | |
| return hasil | |
| except Exception as e: | |
| return {f"Error: {str(e)}": 1.0} | |
| # ============================================================ | |
| # UI | |
| # ============================================================ | |
| with gr.Blocks() as demo: | |
| gr.Markdown("# ⚙️ Sistem Deteksi Kerusakan Motor Matik") | |
| gr.Markdown( | |
| """ | |
| Upload atau rekam suara mesin. | |
| Audio akan otomatis diproses: mono, 16 kHz, 3 detik, lalu diklasifikasi menggunakan model AST. | |
| """ | |
| ) | |
| with gr.Row(): | |
| input_audio = gr.Audio( | |
| sources=["upload", "microphone"], | |
| type="filepath", | |
| label="Audio Mesin" | |
| ) | |
| output = gr.Label( | |
| label="Hasil Prediksi", | |
| num_top_classes=3 | |
| ) | |
| btn = gr.Button("Analisis", variant="primary") | |
| btn.click(fn=prediksi, inputs=input_audio, outputs=output) | |
| demo.queue() | |
| demo.launch() | |