Spaces:
Running
Running
| from typing import Callable, Optional | |
| import numpy as np | |
| import librosa | |
| import gradio as gr | |
| from datetime import datetime | |
| def predict_gradio(data, | |
| uniform_lambda, | |
| sklearn_model, | |
| label_transform, | |
| target_sr: int = 22_050): | |
| if data is None: | |
| return | |
| classes = sklearn_model.classes_ | |
| if label_transform is not None: | |
| classes = label_transform.inverse_transform(classes) | |
| y, sr = data[1], data[0] | |
| y_original_signal = load_as_librosa(y, sr, target_sr=target_sr) | |
| y_uniform = uniform_lambda(y_original_signal, target_sr).astype(np.float32) | |
| prediction = sklearn_model.predict_proba(y_uniform.reshape(1, -1)) | |
| result = {str(label): float(confidence) for ( | |
| label, confidence) in zip(classes, prediction.flatten())} | |
| print(f"{datetime.now()}") | |
| return result | |
| def load_as_librosa(y: np.ndarray, sr: int, target_sr: int = 22050) -> np.ndarray: | |
| data_dtype = y.dtype | |
| dtype_min = np.iinfo(data_dtype).min | |
| dtype_max = np.iinfo(data_dtype).max | |
| dtype_range = np.abs(dtype_max-dtype_min) | |
| y_normalize = (y.astype(np.float32)-dtype_min)/dtype_range | |
| y_normalize_resample = librosa.resample(y=y_normalize, | |
| orig_sr=sr, | |
| target_sr=target_sr) | |
| return y_normalize_resample |