README.md

metadata

dataset_info:
  config_name: swa
  features:
    - name: id
      dtype: string
    - name: audio
      dtype: audio
    - name: audio_language
      dtype: string
    - name: text
      dtype: string
    - name: prompt
      dtype: string
    - name: duration
      dtype: float64
    - name: speaker_id
      dtype: string
  splits:
    - name: train
      num_bytes: 10279990402
      num_examples: 60216
    - name: dev
      num_bytes: 515571820
      num_examples: 3258
    - name: dev_test
      num_bytes: 511609431
      num_examples: 3117
  download_size: 11296884933
  dataset_size: 11307171653
configs:
  - config_name: swa
    data_files:
      - split: train
        path: swa/train-*
      - split: dev
        path: swa/dev-*
      - split: dev_test
        path: swa/dev_test-*

Swahili Speech Dataset – Final Version

Overview

This dataset contains Swahili speech audio paired with transcriptions.
It is split into training, development, and test sets with no speaker overlap.

Dataset Splits

Split	Samples	Hours
train	60,216	359.98
dev	3,258	19.30
dev_test	3,117	18.62
Total	66,591	397.89

Features

• id – unique sample ID
• audio – audio file
• text – transcription
• prompt – prompt text
• duration – audio length in seconds
• speaker_id – anonymized speaker ID