Context

#

Learned in this study

#

Things to explore

#

Overview

#

• Uses 5 building blocks/components in their pipeline
• Training take more than a week on 8 TitanX Maxwell GPUs
• Inference is more efficient if done on CPU than on GPU
• Efficient CPU inference appears to require a lot of complex engineering involving multithreading, synchronization, minimizing nonlinearity FLOPs, avoiding cache trashing and thread contention via thread pinning, as well as using custom hardware-optimized routines for matrix multiplication and convolution

Notes

#

Abstract

#

• Five major building blocks:
  • A segmentation model for locating phoneme boundaries
  • A grapheme-to-phoneme conversion model
  • A phoneme duration prediction model
  • A fundamental frequency prediction model
  • An audio synthesis model

1. Introduction

#

• Our only features are phonemes with stress annotations, phoneme durations, and fundamental frequency (F0)
• We train our entire pipeline on datasets that contains solely audio and unaligned textual transcriptions and generate relatively high quality speech

2. Related Work

#

• Deep Voice is completely standalone; training a new Deep Voice system does not require a pre-existing TTS system, and can be done from scratch using a dataset of short audio clips and corresponding textual transcripts
• Deep Voice minimizes the use of hand-engineered features; it uses one-hot encoded characters for grapheme to phoneme conversion, one-hot encoded phonemes and stresses, phoneme durations in milliseconds, and normalized log fundamental frequency that can be computed from waveforms using any F0 estimation algorithm
• Deep Voice can synthesize audio in fractions of a second, and offers a tunable trade-off between synthesis speed and audio quality

3. TTS System Components

#

3.1. Grapheme-to-Phoneme Model

#

• Model based on the encoder-decoder architecture by Yao & Zweig
• However, we use a multi-layer bidirectional encoder with a gated recurrent unit (GRU) nonlinearity and an equally deep unidirectional GRU decoder
• The initial state of every decoder layer is initialized to the final hidden state of the corresponding encoder forward layer
• The architecture is trained with teacher forcing and decoding is performed using beam search

3.2. Segmentation Model

#

• The connectionist temporal classification (CTC) loss function has been shown to focus on character alignments to learn a mapping between sound and text
• We adapt the convolutional recurrent neural network architecture from a state-of-the-art speech recognition system for phoneme boundary detection
• A network trained with CTC to generate sequences of phonemes will produce brief peaks for every output phoneme
  • Although this is sufficient to roughly align the phonemes to the audio, it is insufficient to detect precise phoneme boundaries
  • To overcome this, we train to predict sequences of phoneme pairs rather than single phonemes

3.3. Phoneme Duration and Fundamental Frequency Model

#

• We use a single architecture to jointly predict phoneme duration and time-dependent fundamental frequency
• The architecture comprises two fully connected layers followed by two unidirectional recurrent layers with GRU cells and finally a fully connected output layer

3.4. Audio Synthesis Model

#

• Our audio synthesis model is a variant of WaveNet

See also

#

References

#

• Arik, Sercan O., et al. “Deep Voice: Real-time neural text-to-speech.” arXiv preprint arXiv:1702.07825 (2017).