README.md

---
language:
- en
pretty_name: Trump Voice Dataset
tags:
- audio
- speech
- tts
license: mit
task_categories:
- automatic-speech-recognition
- text-to-speech
size_categories:
- 10K<n<100K
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
  - split: combined
    path: data/combined-*
dataset_info:
  features:
  - name: audio
    struct:
    - name: array
      sequence: float64
    - name: path
      dtype: 'null'
    - name: sampling_rate
      dtype: int64
  - name: text
    dtype: string
  - name: speaker
    dtype: string
  - name: pitch_mean
    dtype: float64
  - name: pitch_std
    dtype: float64
  splits:
  - name: train
    num_bytes: 12801793
    num_examples: 10
  - name: test
    num_bytes: 12801823
    num_examples: 10
  - name: combined
    num_bytes: 25629160
    num_examples: 20
  download_size: 40149090
  dataset_size: 51232776
---

# Trump Voice Dataset

This dataset contains audio clips of Donald Trump's speech from the World Economic Forum (WEF) 2018, paired with their corresponding transcriptions. The dataset is designed for text-to-speech (TTS) and speech recognition tasks.

## Dataset Description

### Dataset Summary

The Trump Voice Dataset consists of 20 audio samples (10 train, 10 test) extracted from Donald Trump's speech at the World Economic Forum 2018. Each audio clip is approximately 10 seconds long and is paired with its corresponding transcription. The dataset includes additional metadata such as speaker information, pitch mean, and pitch standard deviation.

### Supported Tasks and Leaderboards

This dataset can be used for:
- **Text-to-Speech (TTS)**: Fine-tuning TTS models to generate speech in Donald Trump's voice style
- **Automatic Speech Recognition (ASR)**: Training or testing speech recognition models
- **Voice Analysis**: Analyzing speech patterns, pitch, and other vocal characteristics

### Languages

The dataset contains English speech (en-US) from a single speaker (Donald Trump).

## Dataset Structure

### Data Instances

Each instance in the dataset contains:
- `audio`: The speech audio file (10 seconds, 16kHz sampling rate)
- `text`: The transcription of the speech
- `speaker_name`: The speaker identifier ("Trump")
- `utterance_pitch_mean`: Mean pitch value for the audio clip
- `utterance_pitch_std`: Standard deviation of pitch for the audio clip

Example:
```
{
  'audio': {'array': [...], 'sampling_rate': 16000},
  'text': 'Thank you very much. It's a privilege to be here at this forum where leaders in business science art diplomacy and world affairs have gathered for',
  'speaker_name': 'Trump',
  'utterance_pitch_mean': 451.27,
  'utterance_pitch_std': 314.17
}
```

### Data Splits

The dataset is divided into two splits:
- **Train**: 10 samples
- **Test**: 10 samples

## Dataset Creation

### Source Data

The source data comes from Donald Trump's speech at the World Economic Forum (WEF) 2018.

#### Initial Data Collection and Normalization

The original speech was segmented into 10-second clips. The audio was processed to ensure consistent quality and sampling rate (16kHz).

#### Who are the source language producers?

The source language producer is Donald Trump, the 45th President of the United States, speaking at an official event.

### Annotations

#### Annotation process

The transcriptions were created by aligning the audio segments with the corresponding text from the speech transcript. The alignment was based on timestamps in the original transcript.

## Considerations for Using the Data

### Social Impact of Dataset

This dataset can be used for educational and research purposes in speech technology. It provides real-world speech data from a public figure speaking at a public event.

### Discussion of Biases

The dataset contains speech from only one individual and may not represent diverse speech patterns. Models trained exclusively on this data may exhibit biases toward the specific speaking style, accent, and vocal characteristics of the speaker.

### Other Known Limitations

- Limited to one speaker
- Limited context (formal speech at an international forum)
- Limited sample size (20 samples total)

## Additional Information

### Dataset Curators

This dataset was curated and processed for research and educational purposes.

### Licensing Information

This dataset is released under the MIT License.

### Citation Information

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

```
@dataset{trump_voice_dataset,
  author = {Sakchham19},
  title = {Trump Voice Dataset},
  year = {2023},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/Sakchham19/Trump_Voice_Dataset}
}
```

### Contributions

Contributions to improve the dataset are welcome. Please feel free to open an issue or pull request on the dataset repository.