🗣️ DMTTS: Multi-Language Text-to-Speech System

DMTTS is a VITS-based multilingual text-to-speech toolkit.

🌍 Supported Languages

DMTTS supports English, Chinese, Korean, Japanese, Vietnamese, and Thai.
Each language is currently represented by a single speaker.

---

⚙️ Environment Setup

Using Conda

conda create -n dmtts python=3.11
conda activate dmtts

Simple Installation

pip install git+https://github.com/dmtPlatformInMarch/DMTTS.git 

if Ubuntu 22.04

pip install git+https://github.com/dmtPlatformInMarch/DMTTS.git@ubuntu.22.04

if Ubuntu 20.04

pip install git+https://github.com/dmtPlatformInMarch/DMTTS.git@ubuntu.20.04

for japanese, additionally

pip install git+https://github.com/JRMeyer/jphones.git
python -m unidic download

---

🚀 Quick Start

from dmtts.app.api import TTS

# Default text for each language
DEFAULT_TEXT = {
    "EN": "Did you ever hear a folk tale about a giant turtle?",
    "ZH": "领域近年来发展迅速。",
    "KR": "최근, 텍스트 음성 합성 분야가 급속도로 발전하고 있습니다.",
    "VI": "xâm hại tình dục trẻ em là vấn đề của toàn cầu",
    "TH": "ในช่วงหลังมานี้ เทคโนโลยีสังเคราะห์เสียงพูดได้พัฒนาอย่างรวดเร็ว",
    "JP": "テキスト読み上げの分野は最近急速な発展を遂げています。",
    "RU": "В последнее время сфера синтеза речи развивается быстрыми темпами."
}

LANG_TO_LOCAL_REPO_ID = {
    'EN': 'path/to/your/local/path',
    'JP': 'path/to/your/local/path',
    'ZH': 'path/to/your/local/path',
    'KR': 'path/to/your/local/path',
    'TH': 'path/to/your/local/path',
    'VI': 'path/to/your/local/path',
}
# List of languages to synthesize
Languages = list(DEFAULT_TEXT.keys())

# Speaking speed
speed = 1.0

for language in Languages:
    print(f"Generating speech for [{language}]...")

    # Load model
    # if you want to load model from hugging face
    model = TTS(language=language, device="auto", use_hf=True)
    # or if you want to load from your own directory
    # model = TTS(language=language, device="auto", use_hf=False, local_repo_path_dict=LANG_TO_LOCAL_REPO_ID)

    # Get speaker information
    speaker_ids = model.hps.data.spk2id
    speaker = list(speaker_ids.keys())[0]
    speaker_id = speaker_ids[speaker]

    # Select text
    text = DEFAULT_TEXT[language]

    # Define output file path
    output_path = f"{language}.wav"

    # Run synthesis
    model.tts_to_file(
        text=text,
        speaker_id=speaker_id,
        output_path=output_path,
        speed=speed,
        quiet=True
    )

    print(f"Saved: {output_path}")

---

💾 Clone & Local Installation

git clone https://github.com/kijoongkwon99/DMTTS.git
cd DMTTS
pip install -e .
python -m unidic download

---

---

💾 Model & Config Installation

You can easily download pre-trained DMTTS models for each supported language from Hugging Face.

📁 Download Path in Source Code

All model download and configuration logic is implemented in: dmtts/src/utils/download_utils.py

🌍 Supported Languages

# When downloading from Hugging Face Hub
LANG_TO_HF_REPO_ID = {
    'EN': 'kijoongkwon99/DMTTS-English',
    'JP': 'kijoongkwon99/DMTTS-Japanese',
    'ZH': 'kijoongkwon99/DMTTS-Chinese',
    'KR': 'kijoongkwon99/DMTTS-Korean',
    'TH': 'kijoongkwon99/DMTTS-Thai',
    'VI': 'kijoongkwon99/DMTTS-Vietnamese',
    'RU': 'kijoongkwon99/DMTTS-Russian',
}

If you want to load models locally instead of from Hugging Face,
make sure that BOTH of the following files exist inside each directory:

• checkpoint.pth → the model weights
• config.json → the model configuration

Example:
/to/your/local/path/
    ├── checkpoint.pth
    └── config.json
Otherwise, local loading will fail and you may need to re-download them.

---

---

📁 Project Structure

DMTTS/
│
├── ckpts/                             # Model checkpoints
│   └── README.md
│
├── data/                              # Dataset folder
│
└── src/dmtts/
    ├── __init__.py
    │
    ├── app/                           # API / Main entrypoints
    │   ├── __init__.py
    │   ├── api.py
    │   ├── app.py
    │   ├── main.py
    │   └── README.md
    │
    ├── eval/                          # Evaluation scripts
    │   ├── __init__.py
    │   ├── ecapa_tdnn.py
    │   ├── eval_infer_batch.py
    │   ├── eval_metric_batch.py
    │   ├── infer_metric_batch.sh
    │   ├── result/
    │   └── README.md
    │
    ├── infer/                         # Inference scripts
    │   ├── __init__.py
    │   ├── infer_cli.py
    │   ├── outputs/
    │   └── README.md
    │
    ├── model/                         # Model architecture
    │   ├── __init__.py
    │   ├── backbones/                 
    │   │   ├── __init__.py
    │   │   ├── discriminators.py
    │   │   ├── duration_predictors.py
    │   │   ├── encoders.py
    │   │   ├── flows.py
    │   │   └── generators.py
    │   │
    │   ├── monotonic_align/           # Alignment functions
    │   │   ├── __init__.py
    │   │   └── core.py
    │   │
    │   ├── text/                      # Text processing modules
    │   │   ├── __init__.py
    │   │   ├── chinese.py
    │   │   ├── cleaner.py
    │   │   ├── english.py
    │   │   ├── japanese.py
    │   │   ├── korean.py
    │   │   ├── kr_normalizer.py
    │   │   ├── symbols.py
    │   │   ├── thai.py
    │   │   ├── tone_sandhi.py
    │   │   ├── vietnamese.py
    │   │   ├── english_utils/         # English text normalization utils
    │   │   │   ├── __init__.py
    │   │   │   ├── abbreviations.py
    │   │   │   ├── number_norm.py
    │   │   │   └── time_norm.py
    │   │   └── README.md
    │   │
    │   ├── attentions.py
    │   ├── commons.py
    │   ├── modules.py
    │   ├── synthesizer.py
    │   └── transforms.py
    │
    ├── train/                         # Training scripts & configs
    │   ├── __init__.py
    │   ├── train.py
    │   ├── losses.py
    │   ├── preprocess_text.py
    │   ├── mel_processing.py
    │   ├── train.sh
    │   └── README.md
    │
    └── utils/                         # Helper modules
        ├── __init__.py
        ├── data_utils.py
        ├── download_utils.py
        ├── hparam_utils.py
        ├── infer_utils.py
        └── eval_utils.py

---