---
language:
- so

tags:
- somali
- asr
- speech-recognition
- speech-corpus
- audio
- stt
- transcribed-speech
- female-voice
- clean-audio
- noise-reduction
- low-resource-language
- african-language

license: apache-2.0

task_categories:
- automatic-speech-recognition

size_categories:
- 10K<n<100K
---

# Somali ASR Corpus

Somali ASR Corpus is a cleaned Somali speech dataset designed for Automatic Speech Recognition (ASR) and Speech-to-Text (STT) research. The dataset contains transcribed Somali speech recordings processed with noise reduction and audio normalization techniques to improve training quality for speech models.

## Dataset Description

- Creator: maanka2
- Dataset Type: Speech Corpus
- Language: Somali (so)
- Speaker: Female
- Sampling Rate: 16 kHz
- Primary Use Case: Automatic Speech Recognition (ASR)

## Dataset Features

| Feature | Type |
|----------|------|
| text | string |
| audio | Audio |

## Processing

The audio files were processed using:

- Noise Reduction
- Audio Normalization
- Standardized 16 kHz Sampling Rate

These preprocessing steps help reduce background noise while preserving speech quality.

## Intended Uses

This dataset can be used for:

- Automatic Speech Recognition (ASR)
- Speech-to-Text (STT)
- Speech Foundation Models
- Acoustic Model Training
- Somali Language Technology Research
- Low-Resource Language Research

## Dataset Statistics

- Split: Train
- Number of Samples: 24,640
- Language: Somali
- Speaker Gender: Female

## Example Usage

python from datasets import load_dataset  dataset = load_dataset(     "maanka2/somali-asr-corpus",     split="train" )  print(dataset[0]) 

## Limitations

- Single-speaker dataset.
- Female voice recordings only.
- May not represent all Somali accents and dialects.

## Citation

If you use this dataset in research or production systems, please cite the dataset repository.

## Acknowledgements

This dataset was created for advancing Somali Automatic Speech Recognition and speech technology development for low-resource African languages.