diff --git a/demos/audio_searching/README.md b/demos/audio_searching/README.md
index 010fb15cb6fa8d962ff41104e7a5b5811a59fe14..a2d86df160171fe6fa88db7c53e06b60c68dfece 100644
--- a/demos/audio_searching/README.md
+++ b/demos/audio_searching/README.md
@@ -13,7 +13,7 @@ In this demo, you will learn how to build an audio retrieval system to retrieve
 
 ![Workflow of an audio searching system](./img/audio_searching.png)
 
-Note：this demo uses the [CN-Celeb](http://openslr.org/82/) dataset of at least 650,000 audio entries and 3000 speakers to build the audio vector library, which is then retrieved using a preset distance calculation. The dataset can also use other,  Adjust as needed, e.g. Librispeech, VoxCeleb, UrbanSound, etc
+Note：this demo uses the [CN-Celeb](http://openslr.org/82/) dataset of at least 650,000 audio entries and 3000 speakers to build the audio vector library, which is then retrieved using a preset distance calculation. The dataset can also use other,  Adjust as needed, e.g. Librispeech, VoxCeleb, UrbanSound, GloVe, MNIST, etc
 
 ## Usage
 ### 1. Prepare MySQL and Milvus services by docker-compose
@@ -146,14 +146,14 @@ Then to start the system server, and it provides HTTP backend services.
 - memory：132G
 
 dataset：
-- CN-Celeb, train size 650,000, test size 10,000, dimention 256, distance L2
+- CN-Celeb, train size 650,000, test size 10,000, dimention 192, distance L2
 
 recall and elapsed time statistics are shown in the following figure：
 
   ![](./img/result.png)
 
 
-Compared with other algorithms, the retrieval framework based on Milvus ranks in the middle in terms of speed and performance. Under the premise of 90% recall rate, the retrieval time is about 2.9 milliseconds, which can meet most application scenarios
+The retrieval framework based on Milvus takes about 2.9 milliseconds to retrieve on the premise of 90% recall rate, and it takes about 500 milliseconds for feature extraction (testing audio takes about 5 seconds), that is, a single audio test takes about 503 milliseconds in total, which can meet most application scenarios
 
 ### 5.Pretrained Models
 
diff --git a/demos/audio_searching/README_cn.md b/demos/audio_searching/README_cn.md
index 8da53d44e598f784902ab0b5898dce9149fc0699..ba5cb31db0f88ee505ce4ff0c800e4cb2313e89a 100644
--- a/demos/audio_searching/README_cn.md
+++ b/demos/audio_searching/README_cn.md
@@ -148,13 +148,13 @@ ffce340b3790  minio/minio:RELEASE.2020-12-03T00-03-10Z  "/usr/bin/docker-ent…"
 - 内存：132G
 
 数据集：
-- CN-Celeb, 训练集 65万, 测试集 1万，向量维度 256，距离 L2
+- CN-Celeb, 训练集 65万, 测试集 1万，向量维度 192，距离计算方式 L2
 
 召回和耗时统计如下图：
 
   ![](./img/result.png)
 
-和其他算法比较，基于 milvus 的检索框架在速度与性能排名居中，在召回率 90% 的前提下，检索耗时约 2.9 毫秒，可以满足大多数应用场景
+基于 milvus 的检索框架在召回率 90% 的前提下，检索耗时约 2.9 毫秒，加上特征提取(Embedding)耗时约 500毫秒(测试音频时长约 5秒)，即单条音频测试总共耗时约 503 毫秒，可以满足大多数应用场景
 
 ### 5. 预训练模型
 
diff --git a/demos/audio_searching/img/result.png b/demos/audio_searching/img/result.png
index 63a2ed2342941054d379d4f4cf0b6846212b7deb..c4efc0c7f295870b4d6d4cb7ff8ebcd80b94ef38 100644
Binary files a/demos/audio_searching/img/result.png and b/demos/audio_searching/img/result.png differ