add new metadata

README.md

@@ -12,13 +12,15 @@ dataset_info:
         sampling_rate: 16000
   - name: transcription
     dtype: string
+  - name: num_frames
+    dtype: int32
   config_name: all
   splits:
   - name: train
-    num_bytes: 40431744
+    num_bytes: 41063364
     num_examples: 157905
-  download_size: 5977720839
-  dataset_size: 40431744
+  download_size: 5978669282
+  dataset_size: 41063364
 ---
 
 # Dataset Card for OpenSLR Nepali Large ASR Cleaned
@@ -51,7 +53,7 @@ The data set has been manually quality-checked, but there might still be errors.
 
 The audio files are sampled at a rate of 16KHz, and leading and trailing silences are trimmed using torchaudio's voice activity detection.
 
-For your reference following was the function applied on each of the original openslr utterances.
+For your reference, following was the function applied on each of the original openslr utterances.
 ```python
 import torchaudio
 
@@ -85,11 +87,11 @@ Nepali
     'speaker_id': '09da0',
     'utterance': {
         'path': '/root/.cache/huggingface/datasets/downloads/extracted/e3cf9a618900289ecfd4a65356633d7438317f71c500cbed122960ab908e1e8a/cleaned/asr_nepali/data/e1/e1c4d414df.flac',
-        'array': array([-0.00192261, -0.00204468, -0.00158691, ...,  0.00323486,
-        0.00256348,  0.00262451], dtype=float32),
+        'array': array([-0.00192261, -0.00204468, -0.00158691, ...,  0.00323486, 0.00256348,  0.00262451], dtype=float32),
         'sampling_rate': 16000
     },
-    'transcription': '२००५ मा बिते'
+    'transcription': '२००५ मा बिते',
+    'num_frames': 42300
 }
 ```
 
@@ -102,6 +104,7 @@ Nepali
     - array: numpy array of the utterance
     - sampling_rate: sample rate of the utterance
 - transcription: Nepali text which spoken in the utterance
+- num_frames: length of waveform array
 
 ### Data Splits
 

openslr-nepali-asr-cleaned.py

@@ -80,6 +80,7 @@ class OpenslrNepaliAsrCleaned(datasets.GeneratorBasedBuilder):
                 "speaker_id": datasets.Value("string"),
                 "utterance": datasets.Audio(sampling_rate=16000),
                 "transcription": datasets.Value("string"),
+                "num_frames": datasets.Value(dtype='int32'),
             }
         )
         return datasets.DatasetInfo(
@@ -124,4 +125,5 @@ class OpenslrNepaliAsrCleaned(datasets.GeneratorBasedBuilder):
                     "speaker_id": row['speaker_id'],
                     "utterance": path,
                     "transcription": row['transcription'],
+                    "num_frames": int(row['num_frames']),
                 }