README.md

metadata

license: mit
language:
  - en
task_categories:
  - automatic-speech-recognition
configs:
  - config_name: default
    data_files:
      - split: test
        path: data/test-*
dataset_info:
  features:
    - name: audio
      dtype:
        audio:
          sampling_rate: 16000
    - name: transcript
      dtype: string
  splits:
    - name: test
      num_bytes: 1780718996.498
      num_examples: 3747
  download_size: 1618385128
  dataset_size: 1780718996.498

Mirror note (majentik, 2026-08-26): full mirror of mjwong/SASRBench-v1 at pinned revision 175811c37cfb2d1893f304d62b9c0249bbb4b2f6 (MIT). 3,747 clean Singlish ASR samples derived from IMDA NSC — clean-speech complement to the majentik/WildASR OOD robustness mirror.

SASRBench-v1: Singlish ASR Benchmark V1

Dataset Summary

SASRBench-v1 is a benchmark dataset for evaluating Automatic Speech Recognition (ASR) performance on Singlish. It is derived exclusively from the Part 3 Same Room Environment Close-talk Mic recordings of IMDA's NSC Corpus.

Dataset Derivation

From the Part 3 Same Room Environment Close-talk Mic recordings, audio segments were extracted with the following criteria:

• Minimum Word Count: 10 words
• Maximum Duration: 20 seconds
• Sampling Rate: All audio segments are down-sampled to 16kHz.

A random subset of these segments was then selected to form SASRBench-v1. Each selected audio clip is paired with its corresponding transcript.

Recording Environments in Part 3 (Context)

For context, the original Part 3 of the National Speech Corpus consists of about 1000 hours of conversational data recorded from about 1000 local English speakers, split into pairs. The data includes conversations covering daily life and of speakers playing games provided. Their recordings were split into 2 environments:

• Same Room Environment:
  Speakers were recorded together in one room using two microphones:

  • Close-talk Mic (source for SASRBench-v1)
  • Boundary Mic

• Separate Room Environment:
  Speakers were recorded in individual rooms using two microphones per room:

  • Standing Mic
  • Telephone (IVR)

Dataset Fields

Field	Type	Description
audio	audio	The audio segment extracted from the original recordings.
transcript	string	The transcript corresponding to the audio segment.

Quick Start

To load the test split of SASRBench-v1 using the datasets library, run:

from datasets import load_dataset
data = load_dataset('mjwong/SASRBench-v1', split='test')