数字语音处理课设 基于python两种去噪算法传统的维纳滤波和改进的谱减法+维纳滤波源码.zip

import librosa import numpy as np from config import DefaultConfig class Wiener: def __init__(self): self.config = DefaultConfig.weiner_config self.file_config = DefaultConfig.file_config self.filter = None def genFilter(self, noise, data): """ :param noise: a series of noise data or a single noise data (Must have the same length as data) :param data: a series of clean data or a single clean data (Must have the same length as noise) :return: """ if isinstance(data, list) and isinstance(noise, list): Px, Pn = [], [] for clean_data, noise_data in zip(data, noise): S_clean = librosa.stft(clean_data, n_fft=self.config.fft_length, hop_length=self.config.frame_shift_length, win_length=self.config.window_length) S_noise = librosa.stft(noise_data, n_fft=self.config.fft_length, hop_length=self.config.frame_shift_length, win_length=self.config.window_length) Pxx = np.mean((np.abs(S_clean)) ** 2, axis=1, keepdims=True) # Dx1 Pnn = np.mean((np.abs(S_noise)) ** 2, axis=1, keepdims=True) Px.append(Pxx) Pn.append(Pnn) train_Pxx = np.mean(np.concatenate(Px, axis=1), axis=1, keepdims=True) train_Pnn = np.mean(np.concatenate(Pn, axis=1), axis=1, keepdims=True) self.filter = (train_Pxx / (train_Pxx + self.config.alpha * train_Pnn)) ** self.config.beta elif isinstance(data, np.ndarray) and isinstance(noise, np.ndarray): S_clean = librosa.stft(data, n_fft=self.config.fft_length, hop_length=self.config.frame_shift_length, win_length=self.config.window_length) S_noise = librosa.stft(noise, n_fft=self.config.fft_length, hop_length=self.config.frame_shift_length, win_length=self.config.window_length) Pxx = np.mean((np.abs(S_clean)) ** 2, axis=1, keepdims=True) Pnn = np.mean((np.abs(S_noise)) ** 2, axis=1, keepdims=True) self.filter = (Pxx / (Pxx + self.config.alpha * Pnn)) ** self.config.beta def waveFiltering(self, data, file_name="filtered.wav"): """ :param file_name: file name for saved filtered data :param data: data with noise :return: data filtered by wiener filter """ file_name = self.file_config.file_output_path + file_name S_data = librosa.stft(data, n_fft=self.config.fft_length, hop_length=self.config.frame_shift_length, win_length=self.config.window_length) S_filtered = S_data * self.filter filtered_data = librosa.istft(S_filtered, hop_length=self.config.frame_shift_length, win_length=self.config.window_length) return filtered_data