鲁棒性测试理论案例

public int add(int a,int b){
	//a为被加数，b为加数，return为和
	int result = a + b;
	return int;
}

import org.junit.jupiter.api.Test;
import static org.junit.jupiter.api.Assertions.assertEquals;
public class AddTest {
	@Test
	void testAddPositiveNumbers() {
		assertEquals(5, add(2, 3));
	}
	@Test
	void testAddNegativeNumbers() {
		assertEquals(-5, add(-2, -3));
	}
	@Test
	void testAddMixedNumbers() {
		assertEquals(0, add(-3, 3));
	}
	@Test
	void testAddZero() {
		assertEquals(3, add(0, 3));
		assertEquals(3, add(3, 0));
	}
	private int add(int a, int b) {
		int result = a + b;
		return result;
	}
}

import cv2
import numpy as np
def **add_gaussian_noise**(image_path, output_path, mean=0, std_dev=25):

# 1. 读取图像 (这里以灰度图为例，如需彩色图可修改)
image = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE)
# 2. 生成与图像尺寸相同的高斯噪声
normal(均值, 标准差, 图像形状)
noise = np.random.normal(mean, std_dev, image.shape)
# 3. 将噪声添加到原图像
noisy_image = image + noise
# 4. 将像素值裁剪到 0-255 的有效范围并转换数据类型
noisy_image = np.clip(noisy_image, 0, 255).astype(np.uint8)
# 5. 保存结果图像
cv2.imwrite(output_path, noisy_image)
# 使用示例
add_gaussian_noise(\'1.jpeg\', \'2.jpeg\', mean=0, std_dev=50)

import os
# 关键修复：强制 SpeechBrain 使用 COPY 模式，避免 Windows 权限错误
os.environ\['SB_LOCAL_STRATEGY'] = 'COPY'
import torch
import speechbrain as sb
from speechbrain.inference.speaker import EncoderClassifier
from pydub import AudioSegment
import sys
# 1. 加载预训练的声纹识别模型（ECAPA-TDNN，VoxCeleb数据集训练）
print("正在加载声纹识别模型，请稍候（首次运行可能会下载几百MB的模型文件）...")
classifier = EncoderClassifier.from_hparams(
source="speechbrain/spkrec-ecapa-voxceleb",
savedir="pretrained_models/spkrec-ecapa-voxceleb"
)
# 2. 辅助函数：将 m4a 转为 16kHz 单声道 wav 临时文件
def convert_m4a_to_wav(m4a_path, wav_path):
	try:
		audio = AudioSegment.from_file(m4a_path, format="m4a")
		audio = audio.set_frame_rate(16000).set_channels(1)
		audio.export(wav_path, format="wav")
		return True
	except Exception as e:
		print(f"格式转换失败: {e}")
		return False
# 3. 核心函数：提取声纹特征向量（Embedding）
def extract_embedding(audio_path):
	tmp_wav_path = audio_path.replace(".m4a", "_tmp.wav")
	if audio_path.endswith(\'.m4a\'):
		if not convert_m4a_to_wav(audio_path, tmp_wav_path):
			return None
			audio_file = tmp_wav_path
		else:
			audio_file = audio_path
		try:
			signal = sb.dataio.dataio.read_audio(audio_file)
			batch = signal.unsqueeze(0)
			embedding = classifier.encode_batch(batch, normalize=True)
			return embedding.squeeze()
		except Exception as e:
			print(f"读取音频或提取特征失败: {e}")
			return None
		finally:
			if audio_path.endswith(\'.m4a\') and os.path.exists(tmp_wav_path):
				os.remove(tmp_wav_path)
# 4. 对比两个音频文件
def compare_audio_files(file1, file2):
	print(f\"正在分析: {os.path.basename(file1)} {os.path.basename(file2)}\")
	emb1 = extract_embedding(file1)
	emb2 = extract_embedding(file2)
	if emb1 is None or emb2 is None:
		print(\"分析失败，请检查音频文件是否损坏。\")
		return 
	similarity = torch.nn.functional.cosine_similarity(emb1,emb2,dim=0).item()
	print(\"-\" \* 30)
	print(f\"声纹相似度得分: {similarity:.4f}\")
	print(\"-\" \* 30)
	if similarity \> 0.75:
		print(\"结论：极大概率是同一个人说话 (相似度 \> 0.75)\")
	elif similarity \> 0.60:
		print(\"结论：可能是同一个人，但需要更长时间语音片段确认 (0.600.75)\")
	else:
		print(\"结论：大概率是不同的人说话 (相似度 \< 0.60)\")
# ================== 使用方法 ==================
if __name__ == "__main__":
	\# 检查是否传入了文件参数
	if len(sys.argv) \< 3:
		print(\"请指定两个音频文件的路径。\")
		print(\"用法示例: python compare_voice.py file1.m4a file2.m4a\")
		sys.exit(1)
	audio_file_1 = sys.argv\[1\]
	audio_file_2 = sys.argv\[2\]
	compare_audio_files(audio_file_1, audio_file_2)

public int add(int a,int b){**
//a为被加数，b为加数，return为和
int result = a + b;
	return result;
}

import org.junit.jupiter.api.Test;
import static org.junit.jupiter.api.Assertions.assertEquals;
public class AddTest {
	@Test
	void testAddPositiveNumbers() {
		assertEquals(5, add(2, 3));
	}
	@Test
	void testAddNegativeNumbers() {
		assertEquals(-5, add(-2, -3));
	}
	@Test
	void testAddMixedNumbers() {
		assertEquals(0, add(-3, 3));
	}
	@Test
	void testAddZero() {
		assertEquals(3, add(0, 3));
		assertEquals(3, add(3, 0));
	}
	private int add(int a, int b) {
		int result = a + b;
		return result;
	}
}