small optimization

    test_sig_norm = test_sig / -np.iinfo(np.int16).min

    # check if diff of signal is zero already, then SNR is infinite, since no noise

    diff_norm = ref_sig_norm - test_sig_norm

    if np.all(diff_norm == 0):

    diff_sig_norm = ref_sig_norm - test_sig_norm

    if np.all(diff_sig_norm == 0):

        return np.asarray([np.inf] * ref_sig_norm.shape[1])

    denom_add = 10**-13 * len_seg

    segment_counter = np.zeros(ref_sig.shape[1])

    for ref_seg, test_seg in zip(

    for ref_seg, diff_seg in zip(

        get_framewise(ref_sig_norm, len_seg, zero_pad=True),

        get_framewise(test_sig_norm, len_seg, zero_pad=True),

        get_framewise(diff_sig_norm, len_seg, zero_pad=True),

    ):

        nrg_ref = np.sum(ref_seg**2, axis=0)

        diff_seg = ref_seg - test_seg

        nrg_diff = np.sum(diff_seg**2, axis=0)

        ss_seg = np.log10(1 + nrg_ref / (denom_add + nrg_diff))

    ssnr[segment_counter == 0] = -np.inf

    # set to zero for channels with no diff (this handles e.g. the corner-case of an all-zero channel in both ref and dut)

    zero_diff_mask = np.asarray([np.all(diff_norm[:, c] == 0) for c in range(ref_sig.shape[1])])

    zero_diff_mask = np.asarray([np.all(diff_sig_norm[:, c] == 0) for c in range(ref_sig.shape[1])])

    ssnr[zero_diff_mask] = 0

    return ssnr