# DL Sprint 4.0|Bengali Long-form Speech Recognition — Data

来源：https://www.kaggle.com/competitions/dl-sprint-4-0-bengali-long-form-speech-recognition/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description


This is a Bangla Long-Form ASR (Automatic Speech Recognition)  dataset. The task is to transcribe long Bangla audio recordings into Bangla text. Training data includes audio files with their corresponding reference transcriptions, test data includes audio files only. Submissions require one csv file containing transcriptions for test files. Punctuation is  not required  in the transcriptions.
## Files

### Training Data
- **`train/audio/*.wav`** - Training audio recordings
- **`train/annotation/*.txt`** - transcription for each audio

### Test Data
- **`test/audio/*.wav`** - Test audio recordings to predict

### Submission
- **`sample_submission.csv`** - Example submission in the required format

### Submission Format (`sample_submission.csv`)

| Column | Description |
|--------|-------------|
| `filename` | Test file name |
| `transcript` | transcription text |