Upload README.md with huggingface_hub

README.md

@@ -20,20 +20,25 @@ This dataset contains Malayalam speech audio files and their corresponding trans
 
 ## Dataset Description
 
-The dataset consists of approximately 4,126 Malayalam audio clips in `.wav` format, each accompanied by a `.json` file containing the transcription and segment information.
+The dataset consists of approximately 4,126 Malayalam audio clips, split into training and testing sets with a 90/10 ratio, stratified by speaker gender.
 
 - **Total Audio Files**: 4,126
+- **Train Split**: 3,712 samples
+- **Test Split**: 414 samples
 - **Total Duration**: ~5-6 hours (estimated)
 - **Format**: 16kHz WAV
-- **Annotation**: JSON segments and `metadata.jsonl` entry for Hugging Face `datasets` library compatibility.
+- **Annotation**: JSON segments and split-specific `metadata.jsonl` entries.
 
 ## Dataset Structure
 
-Each sample in the dataset is named with a consistent pattern (e.g., `female_mlf_01130_00015565294`) and has two associated files:
-1. `[id].wav`: The audio file.
-2. `[id].json`: Segment information and transcription.
+The dataset is organized into `train` and `test` directories:
+1. `train/`: Contains 3,712 audio-JSON pairs and `metadata.jsonl`.
+2. `test/`: Contains 414 audio-JSON pairs and `metadata.jsonl`.
 
-Additionally, a `metadata.jsonl` file is provided at the root for easy loading via the `datasets` library.
+Each split contains:
+- `[id].wav`: The audio file.
+- `[id].json`: Segment information and transcription.
+- `metadata.jsonl`: Mapping of audio files to transcriptions.
 
 ### metadata.jsonl format
 ```json