https://github.com/israelg99/deepvoice
Deep Voice: Real-time Neural Text-to-Speech
https://github.com/israelg99/deepvoice
deep-learning keras machine-learning phonemes voice
Last synced: about 2 months ago
JSON representation
Deep Voice: Real-time Neural Text-to-Speech
- Host: GitHub
- URL: https://github.com/israelg99/deepvoice
- Owner: israelg99
- License: apache-2.0
- Created: 2017-03-02T12:58:46.000Z (about 8 years ago)
- Default Branch: master
- Last Pushed: 2017-03-21T18:46:59.000Z (about 8 years ago)
- Last Synced: 2024-08-02T16:02:53.165Z (10 months ago)
- Topics: deep-learning, keras, machine-learning, phonemes, voice
- Language: Python
- Size: 42 KB
- Stars: 354
- Watchers: 43
- Forks: 94
- Open Issues: 6
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Deep Voice
[](https://gitter.im/deep-voice/Lobby?utm_source=badge&utm_medium=badge&utm_campaign=pr-badge&utm_content=badge)
*Based on the [Deep Voice paper](https://arxiv.org/pdf/1702.07825.pdf).*
This repository depends [on my Keras fork](https://github.com/israelg99/keras/tree/master) until it is merged with the official [Keras](https://github.com/fchollet/keras/tree/master) repository.
To install: `pip3 install git+https://github.com/israelg99/keras.git`
**This will override your previously installed Keras version.**
Deep Voice is a text-to-speech system based entirely on deep neural networks.
Deep Voice comprises five models:
- Grapheme-to-phoneme converter.
- Phoneme Segmentation.
- Phoneme duration predictor.
- Frequency predictor.
- Audio synthesis.
## Grapheme-to-phoneme
##### Abstract
The grapheme-to-phoneme converter converts from written text (e.g English characters) to phonemes (encoded using a phonemic alphabet such as ARPABET).
#### Architecture
Based on [this architecture](https://arxiv.org/pdf/1506.00196.pdf) but with some changes.
The Grapheme-to-phoneme converter is an encoder-decoder:
- **Encoder**: multi-layer, bidirectional encoder, with a gated recurrent unit (GRU) nonlinearity.
- **Decoder**: identical to the encoder but unidirectional.
It takes written text as input.
#### Setup
- **Initialization**: every decoder layer is initialized to the final hidden state of the corresponding encoder forward layer.
- **Training**: the architecture is trained with teacher forcing.
- **Decoding**: performed using beam search.
#### Hyperparameters
- **Encoder**: 3 bidirectional layers with 1024 units each.
- **Decoder**: 3 unidirectional layers of the same size as the encoder.
- **Beam Search**: width of 5 candidates.
- **Dropout**: 0.95 rate after each recurrent layer.
## Phoneme Segmentation
##### Abstract
- The phoneme segmentation model locates phoneme boundaries in the voice dataset.
- Given an audio file and a phoneme-by-phoneme transcription of the audio, the segmentation model identifies where in the audio each phoneme begins and ends.
- The phoneme segmentation model is trained to output the alignment between a given utterance and a sequence of target phonemes. This task is similar to the problem of aligning speech to written output in speech recognition.
### Architecture
The segmentation model uses the convolutional recurrent neural network based on [Deep Speech 2](https://arxiv.org/pdf/1512.02595.pdf).
***The architecture graph***
1. Audio vector.
2. 20 MFCCs with 10ms stride.
2. Double 2D convolutions (frequency bins * time).
3. Triple bidirectional recurrent GRUs.
4. Softmax.
5. Output sequence of pairs.
### Hyperparameters
***Convolutions***
- **Stride**: (9, 5).
- **Dropout**: 0.95 rate after last convolution.
***Recurrent layers***
- **Dimensionality**: 512 GRU cells for each direction.
- **Dropout**: 0.95 rate after the last recurrent layer.
### Training
The segmentation model uses the [connectionist temporal classification (CTC) loss](ftp://ftp.idsia.ch/pub/juergen/icml2006.pdf).
## Phoneme Duration + Frequency Predictor
#### Abstract
A single architecture is used to jointly predict phoneme duration and time-dependent fundamental frequency.
#### *Phoneme Duration* Abstract
The phoneme duration predictor predicts the temporal duration of every phoneme in a phoneme sequence (an utterance).
#### *Frequency Predictor* Abstract
The frequency predictor predicts whether a phoneme is voiced. If it is, the model predicts the fundamental frequency (F0) throughout the phoneme’s duration.
### Architecture
1. A sequence of phonemes with stresses, encoded in one-hot vector.
2. Double fully-connected layers.
3. Double unidirectional recurrent layers.
4. Fully-connected layer.
### Hyperparameters
**Double fully-connected layers**
- **Dimensionality**: 256.
- **Dropout**: 0.8 rate after last layer.
**Double unidirectional recurrent layers**
- **Dimensionality**: 128 GRUs.
- **Dropout**: 0.8 rate after last layer.
## Audio Synthesis
#### Abstract
* Combines the outputs of the grapheme-to-phoneme, phoneme duration, and frequency predictor models.
* Synthesizes audio at a high sampling rate, corresponding to the desired text.
* Uses a WaveNet variant which requires less parameters and is faster to train.
### Architecture
The architecture is based on [WaveNet](https://arxiv.org/pdf/1609.03499.pdf) but with some changes.
***Will be updated soon.***