TTSVowelViz

TTSVowelViz is a tool for visualizing static and dynamic vowel spaces during the training of text-to-speech ( TTS) models. This helps researchers and developers monitor the progression of vowel quality over training steps.

✨ Features

• 📊 Visualize static and dynamic vowel spaces across training steps.
• 📈 Track vowel space evolution during model training.
• 🔍 Examine the shape of the learned vowel space.
• 🆚 Compare learned vowel spaces at training steps against the ground truth.
• 🛠️ Customize visualizations with various user-defined inputs and configurations.
• 🧠 Analyze, evaluate, and interpret TTS systems.
• 🧩 Easily integrate into TTS training pipelines with minimal effort.

📦 Installation

Install using pip:

pip install ttsvowelviz

Or install the latest version from source:

git clone https://github.com/pasindu-ud/ttsvowelviz.git
cd ttsvowelviz
pip install .

🔧 Usage

Basic Example

from typing import List, Union

from ttsvowelviz import Synthesizer, TTSVowelViz
from ttsvowelviz.forced_aligner import ForcedAligner, WebMAUSBasicAligner
from ttsvowelviz.formant_extractor import FormantExtractor, PraatFormantExtractor


class ExampleSynthesizer(Synthesizer):
    def synthesize(self, step: int, text: str) -> str:
        # Code to generate speech from text at a given step
        return "Path to the synthesized audio file"


static_vowels: List[str] = ["3:", "6", "6:", "I", "O", "U", "e", "i:", "o:", "{", "}:"]
static_time_points: List[Union[int, float]] = [50]
point_vowels: List[str] = ["i:", "o:", "6:"]
dynamic_vowels: List[str] = ["@}", "Ae", "e:", "oI", "{I", "{O"]
dynamic_time_points: List[Union[int, float]] = [20, 50, 80]
intermediate_steps: List[int] = [0, 1000, 3000]
synthesizer: Synthesizer = ExampleSynthesizer()
forced_aligner: ForcedAligner = WebMAUSBasicAligner(language="eng-NZ")
formant_extractor: FormantExtractor = PraatFormantExtractor()
text_list: List[str] = ["Heard foot hud heed head had hard hod thought goose hid heard.",
                        "How'd hear oat hide lloyd hare aid how'd."]
ground_truth_src_dir_path: str = "Path to the ground truth directory"
vowel_space_dst_dir_path: str = "Path to the directory where vowel spaces should be saved"

tool: TTSVowelViz = TTSVowelViz(static_vowels=static_vowels, static_time_points=static_time_points,
                                point_vowels=point_vowels, dynamic_vowels=dynamic_vowels,
                                dynamic_time_points=dynamic_time_points, intermediate_steps=intermediate_steps,
                                synthesizer=synthesizer, forced_aligner=forced_aligner,
                                formant_extractor=formant_extractor, text_list=text_list,
                                ground_truth_src_dir_path=ground_truth_src_dir_path,
                                vowel_space_dst_dir_path=vowel_space_dst_dir_path)
for s in intermediate_steps:
    tool.execute(step=s)

📚 Citation

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

@inproceedings{ttsvowelviz-2026,
  author    = {Pasindu Udawatta and Jesin James and B.T. Balamurali and Catherine I. Watson and Ake Nicholas and Binu Abeysinghe},
  title     = {{TTSVowelViz: A Tool for Visualising Text-to-Speech Model Training via Vowel Spaces}},
  booktitle = {Proceedings of the Language Resources and Evaluation Conference},
  month     = {May},
  year      = {2026},
  address   = {Palma, Spain},
  publisher = {European Language Resources Association},
  url       = {https://pypi.org/project/ttsvowelviz/}}

📄 License

MIT License. See LICENSE file for details.