Clarify standard FM Complex evaluation and STEER data preparation workflows

README.md

@@ -175,10 +175,12 @@ Every folder contains `metadata.parquet`, `audio.zip` and `manifest.json` with S
 
 ## Continue training
 
-Fine-tune the released [SteerSound checkpoint](https://huggingface.co/GeWu-Lab/SteerSound) for downstream audio editing with either workflow:
+**Pretraining and standard FM fine-tuning:** use the [original audio datasets used in our paper](https://github.com/GeWu-Lab/SteerSound/blob/main/pretrain_and_standard_FM_finetune/dataset/SOURCES.md) or your own audio–caption data. Obtain raw audio from its original datasets and follow the [format and processing guide](https://github.com/GeWu-Lab/SteerSound/blob/main/pretrain_and_standard_FM_finetune/dataset/README.md). Editing examples are constructed online. Use the [standard training guide](https://github.com/GeWu-Lab/SteerSound/tree/main/pretrain_and_standard_FM_finetune) to pretrain or fine-tune, then evaluate on Complex val/test.
 
-1. **Your own audio–caption pairs:** follow the [audio–caption preparation guide](https://github.com/GeWu-Lab/SteerSound/blob/main/pretrain_and_standard_FM_finetune/dataset/README.md). The loader constructs editing instructions and input/target audio online. Fine-tune using [Emphasis](https://github.com/GeWu-Lab/SteerSound/tree/main/STEER_post_training/emphasis/complex) or [Recovery](https://github.com/GeWu-Lab/SteerSound/tree/main/STEER_post_training/recovery/complex), then validate and evaluate on the Complex scene dataset (`complex_val` / `complex_test`).
-2. **Our released Selective dataset:** download `selective_train`, `selective_val`, and `selective_test`. Follow [Emphasis / Selective](https://github.com/GeWu-Lab/SteerSound/tree/main/STEER_post_training/emphasis/selective) or [Recovery / Selective](https://github.com/GeWu-Lab/SteerSound/tree/main/STEER_post_training/recovery/selective) to fine-tune on the supplied editing pairs and evaluate on the corresponding splits.
+**STEER post-training:** start from the released [SteerSound checkpoint](https://huggingface.co/GeWu-Lab/SteerSound) and choose one of these settings:
+
+1. **Complex audio editing data:** prepare the original audio datasets used in our paper, or your own audio–caption data, using the same guide. Apply [Emphasis](https://github.com/GeWu-Lab/SteerSound/tree/main/STEER_post_training/emphasis/complex) or [Recovery](https://github.com/GeWu-Lab/SteerSound/tree/main/STEER_post_training/recovery/complex) with online editing-example construction; validate and evaluate on `complex_val` / `complex_test`.
+2. **Selective dataset:** download the released `selective_train`, `selective_val`, and `selective_test` from the [🤗 dataset](https://huggingface.co/datasets/GeWu-Lab/SteerSound). Follow [Emphasis / Selective](https://github.com/GeWu-Lab/SteerSound/tree/main/STEER_post_training/emphasis/selective) or [Recovery / Selective](https://github.com/GeWu-Lab/SteerSound/tree/main/STEER_post_training/recovery/selective) to post-train on the supplied editing pairs and evaluate on the corresponding splits.
 
 ## Read metadata