Kept audios < 4s as these have better transcriptions.

README.md

@@ -0,0 +1,27 @@
+---
+dataset_info:
+  features:
+  - name: audio
+    dtype:
+      audio:
+        sampling_rate: 16000
+  - name: transcription
+    dtype: string
+  - name: language
+    dtype: string
+  - name: dataset_source
+    dtype: string
+  - name: duration
+    dtype: float64
+  splits:
+  - name: validation
+    num_bytes: 52361033.2
+    num_examples: 105
+  download_size: 7679166
+  dataset_size: 52361033.2
+configs:
+- config_name: default
+  data_files:
+  - split: validation
+    path: data/validation-*
+---

data/validation-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4b29da206830d0f1534a55ba4906eff4e04319b9a9f56c59034e0f5b402c0fb0
+size 7679166