Multi-Speaker Tacotron in TensorFlow

TensorFlow implementation of:

Samples audios (in Korean) can be found here.

Prerequisites

Python 3.6+
FFmpeg
Tensorflow 1.3

Usage

1. Install prerequisites

After preparing Tensorflow, install prerequisites with:

pip3 install -r requirements.txt
python -c "import nltk; nltk.download('punkt')"

If you want to synthesize a speech in Korean dicrectly, follow 2-3. Download pre-trained models.

2-1. Generate custom datasets

The datasets directory should look like:

datasets
├── son
│   ├── alignment.json
│   └── audio
│       ├── 1.mp3
│       ├── 2.mp3
│       ├── 3.mp3
│       └── ...
└── YOUR_DATASET
    ├── alignment.json
    └── audio
        ├── 1.mp3
        ├── 2.mp3
        ├── 3.mp3
        └── ...

and YOUR_DATASET/alignment.json should look like:

{
    "./datasets/YOUR_DATASET/audio/001.mp3": "My name is Taehoon Kim.",
    "./datasets/YOUR_DATASET/audio/002.mp3": "The buses aren't the problem.",
    "./datasets/YOUR_DATASET/audio/003.mp3": "They have discovered a new particle.",
}

After you prepare as described, you should genearte preprocessed data with:

python3 -m datasets.generate_data ./datasets/YOUR_DATASET/alignment.json

2-2. Generate Korean datasets

Follow below commands. (explain with son dataset)

To automate an alignment between sounds and texts, prepare GOOGLE_APPLICATION_CREDENTIALS to use Google Speech Recognition API. To get credentials, read this.
```
export GOOGLE_APPLICATION_CREDENTIALS="YOUR-GOOGLE.CREDENTIALS.json"
```
Download speech(or video) and text.
```
python3 -m datasets.son.download
```

Segment all audios on silence.

python3 -m audio.silence --audio_pattern "./datasets/son/audio/*.wav" --method=pydub

By using Google Speech Recognition API, we predict sentences for all segmented audios.
```
python3 -m recognition.google --audio_pattern "./datasets/son/audio/*.*.wav"
```
By comparing original text and recognised text, save audio<->text pair information into ./datasets/son/alignment.json.
```
python3 -m recognition.alignment --recognition_path "./datasets/son/recognition.json" --score_threshold=0.5
```

Finally, generated numpy files which will be used in training.

python3 -m datasets.generate_data ./datasets/son/alignment.json

Because the automatic generation is extremely naive, the dataset is noisy. However, if you have enough datasets (20+ hours with random initialization or 5+ hours with pretrained model initialization), you can expect an acceptable quality of audio synthesis.

2-3. Generate English datasets

Download speech dataset at https://keithito.com/LJ-Speech-Dataset/
Convert metadata CSV file to json file. (arguments are available for changing preferences)
```
 python3 -m datasets.LJSpeech_1_0.prepare
```

Finally, generate numpy files which will be used in training.

 python3 -m datasets.generate_data ./datasets/LJSpeech_1_0

3. Train a model

The important hyperparameters for a models are defined in hparams.py.

(Change cleaners in hparams.py from korean_cleaners to english_cleaners to train with English dataset)

To train a single-speaker model:

python3 train.py --data_path=datasets/son
python3 train.py --data_path=datasets/son --initialize_path=PATH_TO_CHECKPOINT

To train a multi-speaker model:

# after change `model_type` in `hparams.py` to `deepvoice` or `simple`
python3 train.py --data_path=datasets/son1,datasets/son2

To restart a training from previous experiments such as logs/son-20171015:

python3 train.py --data_path=datasets/son --load_path logs/son-20171015

If you don't have good and enough (10+ hours) dataset, it would be better to use --initialize_path to use a well-trained model as initial parameters.

4. Synthesize audio

You can train your own models with:

python3 app.py --load_path logs/son-20171015 --num_speakers=1

or generate audio directly with:

python3 synthesizer.py --load_path logs/son-20171015 --text "이거 실화냐?"

4-1. Synthesizing non-korean(english) audio

For generating non-korean audio, you must set the argument --is_korean False.

python3 app.py --load_path logs/LJSpeech_1_0-20180108 --num_speakers=1 --is_korean=False
python3 synthesizer.py --load_path logs/LJSpeech_1_0-20180108 --text="Winter is coming." --is_korean=False

Results

Training attention on single speaker model:

Training attention on multi speaker model:

Disclaimer

This is not an official DEVSISTERS product. This project is not responsible for misuse or for any damage that you may cause. You agree that you use this software at your own risk.

References

Author

Taehoon Kim / @carpedm20

symbiote-research/multi-speaker-tacotron-tensorflow