TTS

This project is a part of Mozilla Common Voice. TTS aims a Text2Speech engine low in cost and high in quality. To begin with, you can hear a sample here.

The model here is highly inspired from Tacotron: A Fully End-to-End Text-To-Speech Synthesis Model however, it has many important updates over the baseline model that make training faster and computationally very efficient. Feel free to experiment new ideas and propose changes.

You can find here a brief note pointing possible TTS architectures and their comparisons.

Requirements and Installation

Highly recommended to use miniconda for easier installation.

python>=3.6
pytorch>=0.4.1
librosa
tensorboard
tensorboardX
matplotlib
unidecode

Install TTS using setup.py. It will install all of the requirements automatically and make TTS available to all python environment as an ordinary python module. This makes things easier to run your model outside of the project folder.

python setup.py develop

Or you can use requirements.txt to install the requirements only.

pip install -r requirements.txt

Docker

A barebone Dockerfile exists at the root of the project, which should let you quickly setup the environment. By default, it will start the server and let you query it. Make sure to use nvidia-docker to use your GPUs. Make sure you follow the instructions in the server README before you build your image so that the server can find the model within the image.

docker build -t mozilla-tts .
nvidia-docker run -it --rm -p 5002:5002 mozilla-tts

Checkpoints and Audio Samples

Checkout here to compare the samples (except the first) below.

Models	Commit	Audio Sample	Details
iter-62410	99d56f7	link	First model with plain Tacotron implementation.
iter-170K	e00bc66	link	More stable and longer trained model.
iter-270K	256ed63	link	Stop-Token prediction is added, to detect end of speech.
Best: iter-120K	bf7590	link	Better for longer sentences

Example Model Outputs

Below you see model state after 16K iterations with batch-size 32.

"Recent research at Harvard has shown meditating for as little as 8 weeks can actually increase the grey matter in the parts of the brain responsible for emotional regulation and learning."

Audio output: https://soundcloud.com/user-565970875/iter16k-f48c3b

Runtime

The most time-consuming part is the vocoder algorithm (Griffin-Lim) which runs on CPU. By setting its number of iterations, you might have faster execution with a small loss of quality. Some of the experimental values are below.

Sentence: "It took me quite a long time to develop a voice, and now that I have it I'm not going to be silent."

Audio length is approximately 6 secs.

Time (secs)	System	# GL iters
2.00	GTX1080Ti	30
3.01	GTX1080Ti	60

Data

TTS provides a generic dataloder easy to use for new datasets. You need to write an adaptor to formatyour dataset.Check datasets/preprocess.py to see example adaptors. After your adaptor, you need to set dataset field in config.json accordingly. Some example datasets, we successfuly applied TTS, are linked below.

Training and Fine-tuning LJ-Speech

Click Here for hands on Notebook example, training LJSpeech.

Split metadata.csv into train and validation subsets respectively metadata_train.csv and metadata_val.csv. Note that having a validation split does not work well as oppose to other ML problems since at the validation time model generates spectrogram slices without "Teacher-Forcing" and that leads misalignment between the ground-truth and the prediction. Therefore, validation loss does not really show the model performance. Rather, you might use the all data for training and check the model performance by relying on human inspection.

shuf metadata.csv > metadata_shuf.csv
head -n 12000 metadata_shuf.csv > metadata_train.csv
tail -n 11000 metadata_shuf.csv > metadata_val.csv

To train a new model, you need to define your own config.json file (check the example) and call with the command below.

train.py --config_path config.json

To fine-tune a model, use --restore_path.

train.py --config_path config.json --restore_path /path/to/your/model.pth.tar

If you like to use specific set of GPUs, you need set an environment variable. The code uses automatically all the available GPUs for data parallel training. If you don't specify the GPUs, it uses the all.

CUDA_VISIBLE_DEVICES="0,1,4" train.py --config_path config.json

Each run creates a new output folder and config.json is copied under this folder.

In case of any error or intercepted execution, if there is no checkpoint yet under the output folder, the whole folder is going to be removed.

You can also enjoy Tensorboard, if you point the Tensorboard argument--logdir to the experiment folder.

Testing

Best way to test your pre-trained network is to use Notebooks under notebooks folder.

What is new with TTS

If you train TTS with LJSpeech dataset, you start to hear reasonable results after 12.5K iterations with batch size 32. This is the fastest training with character based methods up to our knowledge. Out implementation is also quite robust against long sentences.

Location sensitive attention (ref). Attention is the vital part of text2speech models. Therefore, it is important to use an attention mechanism that suits the diagonal nature of the problem where the output strictly aligns with the text monotonically. Location sensitive attention performs better by looking into the previous alignment vectors and learns diagonal attention more easily. Yet, I believe there is a good space for research at this front to find a better solution.
Attention smoothing with sigmoid (ref). Attention weights are computed by normalized sigmoid values instead of softmax for sharper values. That enables the model to pick multiple highly scored inputs for alignments while reducing the noise.
Weight decay (ref). After a certain point of the training, you might observe the model over-fitting. That is, model is able to pronounce words probably better but quality of the speech quality gets lower and sometimes attention alignment gets disoriented.
Stop token prediction with an additional module. The original Tacotron model does not propose a stop token to stop the decoding process. Therefore, you need to use heuristic measures to stop the decoder. Here, we prefer to use additional layers at the end to decide when to stop.
Applying sigmoid to the model outputs. Since the output values are expected to be in the range [0, 1], we apply sigmoid to make things easier to approximate the expected output distribution.

One common question is to ask why we don't use Tacotron2 architecture. According to our ablation experiments, nothing, except Location Sensitive Attention, improves the performance, given the big increase in the model size.

Please feel free to offer new changes and pull things off. We are happy to discuss and make things better.

Problems waiting to be solved.

Punctuations at the end of a sentence sometimes affect the pronounciation of the last word. Because punctuation sign is attended by the attention module , that forces network to create a voice signal or at least modify the voice signal being generated for neighboring frames.
~~Simpler stop-token prediction. Right now we use RNN to keep the history of the previous frames. However, we never tested, if something simpler would work as well.~~ Yet RNN based model gives more stable predictions.
Train for better mel-specs. Mel-spectrograms are not good enough to be fed Neural Vocoder. Easy solution to this problem is to train the model with r=1. However,in this case model struggles to align the attention.
irregular words: "minute", "focus", "aren't" etc. Even though, ~~it might be solved~~ (Nancy dataset give much better results compared to LJSpeech) it is solved by a larger or better dataset, some of irregular words cause network to mis-pronounce. Irregular means in this context is that written form and pronounciation of a word have a unique disparity.

Major TODOs

Implement the model.
Generate human-like speech on LJSpeech dataset.
Generate human-like speech on a different dataset (Nancy).
Train TTS with r=1 successfully.
Enable process based distributed training. Similar [to] (https://github.com/fastai/imagenet-fast/).
Adapting Neural Vocoder. The most active work is [here] (https://github.com/erogol/WaveRNN)
Multi-speaker embedding.

References

Precursor implementations

https://github.com/keithito/tacotron (Dataset and Test processing)
https://github.com/r9y9/tacotron_pytorch (Initial Tacotron architecture)

sh-mahdi / TTS

TTS