mcamara/10-random-configurations-pink-trombone

Articulatory Speech Synthesis Dataset with Pink Trombone

This dataset contains audio files generated using the Pink Trombone model, along with articulatory control parameters. It is divided into train and test sets, with corresponding parameter files stored in compressed formats for efficiency.

📂 Dataset Structure

/my_dataset/
│── train/                  # Training audio samples
│   ├── audio_1.wav
│   ├── audio_2.wav
│── test/                   # Testing audio samples
│   ├── audio_1.wav
│   ├── audio_2.wav
│── train.pkl               # Compressed parameters for train set (Pickle)
│── test.pkl                # Compressed parameters for test set (Pickle)

🎯 Description

• Each audio sample lasts 1 second and corresponds to a specific set of articulatory parameters.
• train.pkl and test.pkl store the original control parameters for each audio file.
• The dataset provides a direct mapping between articulatory parameters and generated sounds.

📊 Metadata

• Audio format: WAV, 48 kHz
• Control parameters (in order):
  1. f0 - Fundamental frequency
  2. voiceness - Voicing parameter
  3. tongue_index - Tongue position index
  4. tongue_diam - Tongue diameter
  5. lip_diam - Lip diameter
  6. constriction_index - Constriction position index
  7. constriction_diam - Constriction diameter
  8. throat_diam - Throat diameter
• Recommended tasks: Speech synthesis, articulatory analysis

🔗 Cite This Dataset

If you use this dataset, please cite it as follows:

@phdthesis{upm85089,
            year = {2024},
           title = {Generative and parametric models for interactive neural synthesis in speech and audio},
             doi = {10.20868/UPM.thesis.85089},
           month = {December},
          school = {Telecomunicacion},
          author = {C{\'a}mara, Mateo},
             url = {https://oa.upm.es/85089/},
         }