Update README.md

README.md

@@ -19,7 +19,7 @@ dataset_info:
     - name: audio
       dtype:
         audio:
-          sampling_rate: 22050
+          sampling_rate: 16000
     - name: lyrics
       dtype: string
     - name: mbti
@@ -32,12 +32,6 @@ dataset_info:
       dtype: string
     - name: tempo
       dtype: string
-    - name: lyrics_url
-      dtype: string
-    - name: video_url
-      dtype: string
-    - name: track_id
-      dtype: string
     - name: file_name
       dtype: string
   splits:
@@ -55,9 +49,9 @@ configs:
 An audio and lyrics dataset curated specifically for Math Rock and Midwest Emo genres. Designed as the primary training data for a multimodal analysis system to classify emotion and MBTI personality using Transformer architectures and audio feature extraction.
 
 ## Dataset Description
-- **Total Tracks:** 2500 tracks.
-- **Audio Format:** Embedded RAW bytes (Mono, 22050 Hz). Extreme duration anomalies have been truncated to the first 5 minutes to optimize memory allocation during training.
-- **Total Size:** ~105.3 GB (distributed across 251 Parquet shards).
+- **Total Tracks:** 2500 full tracks.
+- **Audio Format:** FLAC (Mono, 16000 Hz). Downsampled and compressed natively for WavLM compatibility and optimized PyTorch dataloader performance.
+- **Total Size:** ~14.8 GB (distributed across 50 Parquet shards).
 - **Data Sources:** Metadata, lyrics, and audio files extracted independently.
 
 ## Classification Labels (Ground Truth)
@@ -68,16 +62,16 @@ An audio and lyrics dataset curated specifically for Math Rock and Midwest Emo g
 - **Tempo:** Slow, Moderate, Fast.
 
 ## Usage
-Because the dataset is over 100 GB in size, it is highly recommended to use `streaming=True` during inference or training to prevent local memory overload.
+The dataset is optimized for standard machine learning workflows. You can load it directly into memory or use `streaming=True` if you are working in environments with strict memory constraints.
 
 ```python
 from datasets import load_dataset
 
-# Load dataset in streaming mode
-dataset = load_dataset("anggars/neural-mathrock", streaming=True)
+# Load dataset
+dataset = load_dataset("anggars/neural-mathrock")
 
 # Iterate and fetch a sample
-sample = next(iter(dataset['train']))
+sample = dataset['train'][0]
 
 print(f"Artist : {sample['artist']}")
 print(f"Song   : {sample['song']}")