写一个语音NLMS自适应滤波的函数

下面是一个语音NLMS自适应滤波的函数，供参考：

import numpy as np

def nlms_filter(input_signal, desired_signal, filter_order, step_size, leak):
    """
    Nonlinear least mean squares (NLMS) adaptive filter for speech signal denoising.

    Parameters:
    input_signal: array-like, 1d
        The noisy speech signal to be filtered.
    desired_signal: array-like, 1d
        The clean speech signal used as reference for the filtering.
    filter_order: int
        The order of the adaptive filter.
    step_size: float
        The step size parameter for the NLMS algorithm.
    leak: float
        The leakage parameter, used to reduce the bias in the filter coefficients.

    Returns:
    output_signal: array-like, 1d
        The filtered speech signal.

    """

    # Initialize filter coefficients and output signal
    filter_coeffs = np.zeros(filter_order)
    output_signal = np.zeros(len(input_signal))

    # Iterate over the input signal, updating the filter coefficients and generating the
    # output signal sample by sample
    for n in range(filter_order, len(input_signal)):
        # Extract the current block of input samples and the current desired signal sample
        current_input = input_signal[n - filter_order : n]
        current_desired = desired_signal[n]

        # Compute the output signal for the current sample
        current_output = np.dot(current_input, filter_coeffs)

        # Compute the error between the current desired signal sample and the current output
        current_error = current_desired - current_output

        # Update the filter coefficients using the NLMS algorithm
        scale_factor = np.dot(current_input, current_input) + 1e-8  # Add a small number to avoid division by zero
        delta_coeffs = step_size * current_error * current_input / scale_factor
        filter_coeffs += delta_coeffs + leak * filter_coeffs

        # Save the current output signal sample
        output_signal[n] = current_output

    return output_signal

请注意，在这里提供的代码仅供参考，您应该理解每个参数和操作的含义，并根据您的具体需求进行调整和修改。此外，该算法也可能存在一些缺陷和局限性，需要仔细评估和验证其效果。