Update README.md

README.md

@@ -1,27 +1,16 @@
----
-dataset_info:
-  features:
-  - name: audio
-    dtype:
-      audio:
-        sampling_rate: 16000
-  - name: dataset
-    dtype: string
-  - name: text
-    dtype: string
-  - name: id
-    dtype: string
-  - name: duration_ms
-    dtype: float64
-  splits:
-  - name: validation
-    num_bytes: 109125775.62304688
-    num_examples: 511
-  download_size: 107921497
-  dataset_size: 109125775.62304688
-configs:
-- config_name: default
-  data_files:
-  - split: validation
-    path: data/validation-*
----
+A filtered (<=30s duration) slice (512 samples) of the Earnings22 dataset.
+
+```python
+def add_duration(sample):
+    y, sr = sample['audio']["array"], sample['audio']["sampling_rate"]
+    sample['duration_ms']=librosa.get_duration(y=y, sr=sr) * 1000
+    return sample
+
+tedlium = load_dataset("esb/datasets", "earnings22", split='validation', trust_remote_code=True)
+# compute duration to filter
+tedlium = tedlium.map(add_duration)
+tedlium = tedlium.select(range(512))
+
+# Whisper max supported duration
+tedlium = tedlium.filter(lambda example: example['duration_ms'] < 30000)
+```