diff --git a/README.md b/README.md
index 9d79b74..d17f33a 100644
--- a/README.md
+++ b/README.md
@@ -4,7 +4,7 @@
This dataset is designed for multi-modal speaker diarization and lip-speech synchronization in the wild. [Demo](https://x-lance.github.io/MSDWILD)
-* [2024/01/08 Updates] The new frame-by-frame face bounding boxes and their positions in each frame have been uploaded as supplementary materials for our dataset. [Jump to Here](#videos-with-frame-by-frame-face-position-annotation).
+* [2024/01/08 Updates] The new frame-by-frame face bounding boxes and their positions in each frame have been uploaded as supplementary materials for our dataset. [-> Jump <-](#videos-with-frame-by-frame-face-position-annotation).
## Dataset Statistics
@@ -97,6 +97,7 @@ Move the `slider` to preview the positions and ID information of faces on differ
* `DarkLabel` only supports Windows (Win10 or Win11) currently and you may use wine (mentioned in this [issue](https://github.com/darkpgmr/DarkLabel/issues/4)) to run on Mac or Linux.
* The result can `not` directly converted to exactly the same [RTTM](./rttms/all.rttm) as some duration or face ids are adjusted and off-screen speech is not included in this part. By the way, the facial identification in each video is unique and also differs from the identifiers in [RTTM](./rttms/all.rttm) mentioned above.
* Different from the above-mentioned cropped face, the annotation here is for the bounding box of the unprocessed face in the original video.
+* Why are we releasing it now? Our initial experiments were conducted using a training set based on cropped faces. However, we realized that facial tagging is extremely important for multi-modal speaker diarization. Consequently, following the publication of our work, we decided to embark on a frame-by-frame review process. The undertaking is massive, involving the inspection of approximately 120,000 video frames, and ensuring that the IDs remain consistent throughout the video. We also conducted a second round of checks for added accuracy. It is only after this meticulous process that we are now able to release the dataset for public use.
* I suggest that this is merely supplementary material for this dataset. Possible future work we envision includes training an end-to-end multimodal speaker diarization that incorporates facial location information, and an evaluation method for a multimodal speaker diarization that takes into account the human face location.
@@ -121,7 +122,7 @@ No other post-processing methods are used.
## Analysis Result
-You can refer to [URL](https://github.com/liutaocode/DiarizationVisualization) to visualize the dataset or your algorithm result.
+You can refer to [URL](https://github.com/liutaocode/DiarizationVisualization) to visualize the dataset based on your algorithm result.
