Update README.

README.md

@@ -26,16 +26,37 @@ dataset_info:
 ---
 
 # AudioSet data
-This repository contains the balanced training set and evaluation set
-of the [AudioSet data](
+This repository contains the balanced training set and evaluation set of
+the [AudioSet data](
 https://research.google.com/audioset/dataset/index.html). The YouTube
-videos were downloaded in March 2023, and so not all of the original
-audios are available.
+videos were downloaded in March 2023, so not all of the original audios
+are available.
 
-Extracting the `*.tar` files will place audio clips into the `audio/`
-directory. The distribuion of audio clips is as follows:
-- `audio/bal_train`: 18685 audio clips out of 22160 originally.
-- `audio/eval`: 17142 audio clips out of 20371 originally.
+The distribuion of audio clips is as follows. In parentheses is the dict
+key used for HugginFace `datasets`:
+- `bal_train` (`train`): 18685 audio clips out of 22160 originally.
+- `eval` (`test`): 17142 audio clips out of 20371 originally.
+
+You can use the `datasets` library to load this dataset, in which case
+the raw audio will be returned along with a sequence of one or more
+labels. Note that the raw audio is returned without further processing,
+so you will need to decode and possibly downsample the audio for model
+training.
+
+Example instance from the `train` subset:
+```python
+{
+ 'video_id': '--PJHxphWEs',
+ 'audio': {
+  'path': 'audio/bal_train/--PJHxphWEs.flac',
+  'array': array([-0.04364824, -0.05268681, -0.0568949 , ...,  0.11446512,
+          0.14912748,  0.13409865]),
+  'sampling_rate': 48000
+ },
+ 'labels': ['/m/09x0r', '/t/dd00088'],
+ 'human_labels': ['Speech', 'Gush']
+}
+```
 
 Most audio is sampled at 48 kHz 24 bit, but about 10% is sampled at
 44.1 kHz 24 bit. Audio files are stored in the FLAC format.