diff --git a/src/f5_tts/infer/README.md b/src/f5_tts/infer/README.md
index cba4eb2..fe48a78 100644
--- a/src/f5_tts/infer/README.md
+++ b/src/f5_tts/infer/README.md
@@ -12,6 +12,8 @@ To avoid possible inference failures, make sure you have seen through the follow
 - Uppercased letters will be uttered letter by letter, so use lowercased letters for normal words. 
 - Add some spaces (blank: " ") or punctuations (e.g. "," ".") to explicitly introduce some pauses.
 - Preprocess numbers to Chinese letters if you want to have them read in Chinese, otherwise in English.
+- If the generation output is blank (pure silence), check for ffmpeg installation (various tutorials online, blogs, videos, etc.).
+- Try turn off use_ema if using an early-stage finetuned checkpoint (which goes just few updates).
 
 
 ## Gradio App
diff --git a/src/f5_tts/train/README.md b/src/f5_tts/train/README.md
index 90d7c85..d269449 100644
--- a/src/f5_tts/train/README.md
+++ b/src/f5_tts/train/README.md
@@ -48,6 +48,8 @@ Discussion board for Finetuning [#57](https://github.com/SWivid/F5-TTS/discussio
 
 Gradio UI training/finetuning with `src/f5_tts/train/finetune_gradio.py` see [#143](https://github.com/SWivid/F5-TTS/discussions/143).
 
+The `use_ema = True` is harmful for early-stage finetuned checkpoints (which goes just few updates, thus ema weights still dominated by pretrained ones), try turn it off and see if provide better results.
+
 ### 3. Wandb Logging
 
 The `wandb/` dir will be created under path you run training/finetuning scripts.