---
language:
- en
pretty_name: imda-
---

# IMDA National Speech Corpus (NSC) Speech-to-Text

Originally from https://www.imda.gov.sg/how-we-can-help/national-speech-corpus, this repository simply a mirror. This dataset associated with Singapore Open Data Licence, https://www.sla.gov.sg/newsroom/statistics/singapore-open-data-licence

We uploaded mp3 files and compressed using 7z,

```bash
7za x part1-mp3.7z.001
```

All notebooks at https://github.com/mesolitica/malaya-speech/tree/malaysian-speech/data/imda

## total lengths

1. part 1, 1117.9866586978865 hours
2. part 2, 1052.1777120312486 hours
3. part 3, 2162.4968734548597 hours
4. part 4, 2133.685638097089 hours

## Why no HuggingFace dataset format?

We had bad experiences with HuggingFace dataset format to load huge dataset. Reading mp3 files during iteration is much more faster and efficient.