diff --git a/README.md b/README.md
index 3851abd..e666788 100644
--- a/README.md
+++ b/README.md
@@ -8,7 +8,7 @@ is a Fréchet distance; lower is better.
 
 All models are evaluated with the same SLIP protocol. Lower is better.
 
-| VAE | Frame rate (Hz) | Latent dim | SeedTTS-Eval | LibriSpeech-PC |
+| VAE | Frame rate (Hz) | Latent dim | SeedTTS-Eval | LibriSpeech-PC test-clean |
 | --- | ---: | ---: | ---: | ---: |
 | [Semantic-VAE (dim 16)](https://github.com/ZhikangNiu/Semantic-VAE) | 40 | 16 | **0.2025** | **0.1778** |
 | [KVAE-Audio](https://github.com/kandinskylab/kvae-audio) | 50 | 64 | 0.2066 | 0.2428 |
@@ -30,8 +30,9 @@ All models are evaluated with the same SLIP protocol. Lower is better.
 | [SAME-S](https://github.com/Stability-AI/stable-audio-3) | 10.77 | 256 | 1.5220 | 1.7352 |
 | [AudioLDM](https://github.com/haoheliu/AudioLDM) | 25 | 128 | 1.8676 | 2.2150 |
 
-SeedTTS-Eval contains 3,108 evaluated utterances and LibriSpeech-PC contains 1,127. Exact values and source links are
-available in [`leaderboard.csv`](leaderboard.csv).
+SeedTTS-Eval contains 3,108 evaluated utterances. LibriSpeech-PC test-clean contains the 1,127 target utterances from
+the [F5-TTS cross-sentence evaluation list](https://github.com/SWivid/F5-TTS/blob/main/data/librispeech_pc_test_clean_cross_sentence.lst).
+Exact values and model links are available in [`leaderboard.csv`](leaderboard.csv).
 
 ## Installation
 
@@ -53,7 +54,9 @@ slip datasets prepare-seedtts \
   --output data/seedtts-eval-v1
 ```
 
-Or download [LibriSpeech-PC](https://huggingface.co/datasets/AudioLLMs/librispeech_test_clean):
+Or prepare LibriSpeech-PC test-clean. The target set follows the
+[F5-TTS cross-sentence list](https://github.com/SWivid/F5-TTS/blob/main/data/librispeech_pc_test_clean_cross_sentence.lst),
+with audio obtained from [AudioLLMs/librispeech_test_clean](https://huggingface.co/datasets/AudioLLMs/librispeech_test_clean):
 
 ```bash
 slip datasets download-librispeech-pc \