Tacotron with Location Relative Attention

A PyTorch implementation of Location-Relative Attention Mechanisms For Robust Long-Form Speech Synthesis. Audio samples can be found here. Colab demo can be found here.

^{Fig 1:Tacotron (with Dynamic Convolution Attention).}

^{Fig 2:Example Mel-spectrogram and attention plot.}

Quick Start

Ensure you have Python 3.6 and PyTorch 1.7 or greater installed. Then install this package (along with the univoc vocoder):

pip install tacotron univoc

Example Usage

import torch
import soundfile as sf
from univoc import Vocoder
from tacotron import load_cmudict, text_to_id, Tacotron

# download pretrained weights for the vocoder (and optionally move to GPU)
vocoder = Vocoder.from_pretrained(
    "https://github.com/bshall/UniversalVocoding/releases/download/v0.2/univoc-ljspeech-7mtpaq.pt"
).cuda()

# download pretrained weights for tacotron (and optionally move to GPU)
tacotron = Tacotron.from_pretrained(
    "https://github.com/bshall/Tacotron/releases/download/v0.1/tacotron-ljspeech-yspjx3.pt"
).cuda()

# load cmudict and add pronunciation of PyTorch
cmudict = load_cmudict()
cmudict["PYTORCH"] = "P AY1 T AO2 R CH"

text = "A PyTorch implementation of Location-Relative Attention Mechanisms For Robust Long-Form Speech Synthesis."

# convert text to phone ids
x = torch.LongTensor(text_to_id(text, cmudict)).unsqueeze(0).cuda()

# synthesize audio
with torch.no_grad():
    mel, _ = tacotron.generate(x)
    wav, sr = vocoder.generate(mel.transpose(1, 2))

# save output
sf.write("location_relative_attention.wav", wav, sr)

Train from Scatch

1. Clone the repo:

git clone https://github.com/bshall/Tacotron
cd ./Tacotron

2. Install requirements:

pipenv install

3. Download and extract the LJ-Speech dataset:

wget https://data.keithito.com/data/speech/LJSpeech-1.1.tar.bz2
tar -xvjf LJSpeech-1.1.tar.bz2

4. Download the train split here and extract it in the root directory of the repo.
5. Extract Mel spectrograms and preprocess audio:

pipenv run python preprocess.py path/to/LJSpeech-1.1 datasets/LJSpeech-1.1

usage: preprocess.py [-h] in_dir out_dir

Preprocess an audio dataset.

positional arguments:
  in_dir      Path to the dataset directory
  out_dir     Path to the output directory

optional arguments:
  -h, --help  show this help message and exit

6. Train the model:

pipenv run python train.py ljspeech datasets/LJSpeech-1.1 path/to/LJSpeech-1.1/metadata.csv

usage: train.py [-h] [--resume RESUME] checkpoint_dir text_path dataset_dir

Train Tacotron with dynamic convolution attention.

positional arguments:
  checkpoint_dir   Path to the directory where model checkpoints will be saved
  text_path        Path to the dataset transcripts
  dataset_dir      Path to the preprocessed data directory

optional arguments:
  -h, --help       show this help message and exit
  --resume RESUME  Path to the checkpoint to resume from

Pretrained Models

Pretrained weights for the LJSpeech model are available here.

Notable Differences from the Paper

1. Trained using a batch size of 64 on a single GPU (using automatic mixed precision).
2. Used a gradient clipping threshold of 0.05 as it seems to stabilize the alignment with the smaller batch size.
3. Used a different learning rate schedule (again to deal with smaller batch size).
4. Used 80-bin (instead of 128 bin) log-Mel spectrograms.

Acknowlegements

• https://github.com/keithito/tacotron
• https://github.com/PetrochukM/PyTorch-NLP
• https://github.com/fatchord/WaveRNN