Custom TIMIT Dataset

This dataset is a restructured, Hugging Face-compatible version of the TIMIT Acoustic-Phonetic Continuous Speech Corpus. It contains high-quality 16kHz audio recordings of American English speakers, paired with rich metadata including precise time-aligned word and phoneme transcriptions.

Dataset Structure

The dataset is formatted using the audiofolder structure and uses JSON Lines (metadata.jsonl) to map audio files to their corresponding metadata and time-aligned labels.

Features

Each instance in the dataset contains the following features:

• audio_id (string): The unique identifier for the audio file (e.g., "0.wav").
• audio (Audio): The audio signal loaded at a 16kHz sampling rate.
• dialect_region (string): The geographic dialect region of the speaker in the United States.
• speaker_id (string): A unique alphanumeric identifier for the speaker.
• words (Sequence): Time-aligned word transcriptions consisting of:
  • start (int32): The start time (in samples).
  • end (int32): The end time (in samples).
  • text (string): The transcribed word.
• phonemes (Sequence): Time-aligned phonetic transcriptions consisting of:
  • start (int32): The start time (in samples).
  • end (int32): The end time (in samples).
  • text (string): The transcribed phoneme.

Example Usage

You can load the dataset using the datasets library:

from datasets import load_dataset

# Replace 'your-hf-username/custom-timit' with your actual repository ID
dataset = load_dataset("your-hf-username/custom-timit")

# Access the first training example
sample = dataset["train"][0]
print(sample["words"]["text"])
print(sample["phonemes"]["text"])

Source Data and Attribution

This dataset is derived from the original TIMIT dataset developed jointly by the Massachusetts Institute of Technology (MIT), SRI International (SRI), and Texas Instruments, Inc. (TI).

Original Dataset Link: Linguistic Data Consortium (LDC93S1)

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

@misc{garofolo1993timit,
  title={TIMIT Acoustic-Phonetic Continuous Speech Corpus LDC93S1},
  author={Garofolo, John S. and Lamel, Lori F. and Fisher, William M. and Fiscus, Jonathan G. and Pallett, David S. and Dahlgren, Nancy L. and Zue, Victor W.},
  year={1993},
  publisher={Linguistic Data Consortium, Philadelphia}
}