README.md

---
language:
- tw
license: cc-by-4.0
task_categories:
- automatic-speech-recognition
- text-to-speech
task_ids:
- keyword-spotting
multilinguality:
- monolingual
size_categories:
- 1K<n<10K
modalities:
- audio
- text
dataset_info:
  features:
  - name: audio
    dtype: audio
  - name: text
    dtype: string
  config_name: default
  splits:
  - name: train
    num_bytes: 0
    num_examples: 413463
  download_size: 0
  dataset_size: 0
tags:
- speech
- twi
- akan
- ghana
- african-languages
- low-resource
- parallel-corpus
pretty_name: Twi Words Speech-Text Parallel Dataset
---

# Twi Words Speech-Text Parallel Dataset

## Dataset Description

This dataset contains 413463 parallel speech-text pairs for Twi (Akan), a language spoken primarily in Ghana. The dataset consists of audio recordings paired with their corresponding text transcriptions, making it suitable for automatic speech recognition (ASR) and text-to-speech (TTS) tasks.

### Dataset Summary

- **Language**: Twi (Akan) - `tw`
- **Task**: Speech Recognition, Text-to-Speech
- **Size**: 413463 audio files > 1KB (small/corrupted files filtered out)
- **Format**: WAV audio files with corresponding text labels
- **Modalities**: Audio + Text

### Supported Tasks

- **Automatic Speech Recognition (ASR)**: Train models to convert Twi speech to text
- **Text-to-Speech (TTS)**: Use parallel data for TTS model development
- **Keyword Spotting**: Identify specific Twi words in audio
- **Phonetic Analysis**: Study Twi pronunciation patterns

## Dataset Structure

### Data Fields

- `audio`: Audio file in WAV format
- `text`: Corresponding text transcription

### Data Splits

The dataset contains a single training split with 413463 filtered audio files.

### File Structure

Each audio segment is stored as a numbered pair:
- `NNNN.wav`: Audio file (e.g., `0001.wav`)
- `NNNN.txt`: Corresponding text file (e.g., `0001.txt`)

This structure ensures clean organization and easy pairing of audio-text data.

## Dataset Creation

### Source Data

The audio data has been sourced ethically from consenting contributors. To protect the privacy of the original authors and speakers, specific source information cannot be shared publicly.

### Data Processing

1. Audio files were processed using forced alignment techniques
2. Word-level segmentation was performed with padding to prevent abrupt cuts
3. Audio segments were filtered based on:
   - Minimum duration requirements
   - Volume/vocal content thresholds
   - File size validation (> 1KB)
4. Each valid segment was saved as a numbered audio-text pair
5. Audio processing used the [MMS-300M-1130 Forced Aligner](https://huggingface.co/MahmoudAshraf/mms-300m-1130-forced-aligner) tool for alignment and quality assurance

### Quality Control

- Empty or silent audio segments were automatically filtered out
- Very short segments (< 200ms) were excluded
- Low-volume segments were removed to ensure vocal content
- Audio padding (100ms) was added to prevent abrupt word cuts

### Annotations

Text annotations are stored in separate `.txt` files corresponding to each audio file, representing the exact spoken content in each audio segment.

## Considerations for Using the Data

### Social Impact of Dataset

This dataset contributes to the preservation and digital representation of Twi, supporting:
- Language technology development for underrepresented languages
- Educational resources for Twi language learning
- Cultural preservation through digital archives

### Discussion of Biases

- The dataset may reflect the pronunciation patterns and dialects of specific regions or speakers
- Audio quality and recording conditions may vary across samples
- The vocabulary is limited to the words present in the collected samples

### Other Known Limitations

- Limited vocabulary scope (word-level rather than sentence-level)
- Potential audio quality variations
- Regional dialect representation may be uneven
- Automatic filtering may have removed some valid segments

## Additional Information

### Licensing Information

This dataset is released under the Creative Commons Attribution 4.0 International License (CC BY 4.0).

### Acknowledgments

- Audio processing and alignment performed using [MMS-300M-1130 Forced Aligner](https://huggingface.co/MahmoudAshraf/mms-300m-1130-forced-aligner)
- The original audio is produced by The Ghana Institute of Linguistics, Literacy and Bible Translation in partnership with Davar Partners
- Automated quality filtering and padding applied to ensure high-quality audio segments

### Citation Information

If you use this dataset in your research, please cite:

```
@dataset{twi_words_parallel_2025,
  title={Twi Words Speech-Text Parallel Dataset},
  year={2025},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/datasets/michsethowusu/twi-words-speech-text-parallel}}
}
```

### Contact

For questions or concerns about this dataset, please open an issue in the dataset repository.

## Usage Example

```python
from datasets import load_dataset

# Load the dataset
dataset = load_dataset("michsethowusu/twi-words-speech-text-parallel")

# Access audio and text pairs
for example in dataset["train"]:
    audio = example["audio"]
    text = example["text"]
    print(f"Text: {text}")
    print(f"Audio sample rate: {audio['sampling_rate']}")
```