:yum: TensorFlowTTS
Real-Time State-of-the-art Speech Synthesis for Tensorflow 2 :zany_face: TensorFlowTTS provides real-time state-of-the-art speech synthesis architectures such as Tacotron-2, Melgan, Multiband-Melgan, FastSpeech, FastSpeech2 based-on TensorFlow 2. With Tensorflow 2, we can speed-up training/inference progress, optimizer further by using [fake-quantize aware](https://www.tensorflow.org/model_optimization/guide/quantization/training_comprehensive_guide) and [pruning](https://www.tensorflow.org/model_optimization/guide/pruning/pruning_with_keras), make TTS models can be run faster than real-time and be able to deploy on mobile devices or embedded systems. ## What's new - 2020/12/02 **(NEW!)** Support German TTS with [Thorsten dataset](https://github.com/thorstenMueller/deep-learning-german-tts). See the [Colab](https://colab.research.google.com/drive/1W0nSFpsz32M0OcIkY9uMOiGrLTPKVhTy?usp=sharing). Thanks [thorstenMueller](https://github.com/thorstenMueller) and [monatis](https://github.com/monatis). - 2020/11/24 **(NEW!)** Add HiFi-GAN vocoder. See [here](https://github.com/TensorSpeech/TensorFlowTTS/tree/master/examples/hifigan) - 2020/11/19 **(NEW!)** Add Multi-GPU gradient accumulator. See [here](https://github.com/TensorSpeech/TensorFlowTTS/pull/377) - 2020/08/23 Add Parallel WaveGAN tensorflow implementation. See [here](https://github.com/TensorSpeech/TensorFlowTTS/tree/master/examples/parallel_wavegan) - 2020/08/23 Add MBMelGAN G + ParallelWaveGAN G example. See [here](https://github.com/TensorSpeech/TensorFlowTTS/tree/master/examples/multiband_pwgan) - 2020/08/20 Add C++ inference code. Thank [@ZDisket](https://github.com/ZDisket). See [here](https://github.com/TensorSpeech/TensorFlowTTS/tree/master/examples/cppwin) - 2020/08/18 Update [new base processor](https://github.com/TensorSpeech/TensorFlowTTS/blob/master/tensorflow_tts/processor/base_processor.py). Add [AutoProcessor](https://github.com/TensorSpeech/TensorFlowTTS/blob/master/tensorflow_tts/inference/auto_processor.py) and [pretrained processor](https://github.com/TensorSpeech/TensorFlowTTS/blob/master/tensorflow_tts/processor/pretrained/) json file - 2020/08/14 Support Chinese TTS. Pls see the [colab](https://colab.research.google.com/drive/1YpSHRBRPBI7cnTkQn1UcVTWEQVbsUm1S?usp=sharing). Thank [@azraelkuan](https://github.com/azraelkuan) - 2020/08/05 Support Korean TTS. Pls see the [colab](https://colab.research.google.com/drive/1ybWwOS5tipgPFttNulp77P6DAB5MtiuN?usp=sharing). Thank [@crux153](https://github.com/crux153) - 2020/07/17 Support MultiGPU for all Trainer - 2020/07/05 Support Convert Tacotron-2, FastSpeech to Tflite. Pls see the [colab](https://colab.research.google.com/drive/1HudLLpT9CQdh2k04c06bHUwLubhGTWxA?usp=sharing). Thank @jaeyoo from the TFlite team for his support - 2020/06/20 [FastSpeech2](https://arxiv.org/abs/2006.04558) implementation with Tensorflow is supported. - 2020/06/07 [Multi-band MelGAN (MB MelGAN)](https://github.com/tensorspeech/TensorFlowTTS/blob/master/examples/multiband_melgan/) implementation with Tensorflow is supported ## Features - High performance on Speech Synthesis. - Be able to fine-tune on other languages. - Fast, Scalable, and Reliable. - Suitable for deployment. - Easy to implement a new model, based-on abstract class. - Mixed precision to speed-up training if possible. - Support Single/Multi GPU gradient Accumulate. - Support both Single/Multi GPU in base trainer class. - TFlite conversion for all supported models. - Android example. - Support many languages (currently, we support Chinese, Korean, English.) - Support C++ inference. - Support Convert weight for some models from PyTorch to TensorFlow to accelerate speed. ## Requirements This repository is tested on Ubuntu 18.04 with: - Python 3.7+ - Cuda 10.1 - CuDNN 7.6.5 - Tensorflow 2.2/2.3 - [Tensorflow Addons](https://github.com/tensorflow/addons) >= 0.10.0 Different Tensorflow version should be working but not tested yet. This repo will try to work with the latest stable TensorFlow version. **We recommend you install TensorFlow 2.3.0 to training in case you want to use MultiGPU.** ## Installation ### With pip ```bash $ pip install TensorFlowTTS ``` ### From source Examples are included in the repository but are not shipped with the framework. Therefore, to run the latest version of examples, you need to install the source below. ```bash $ git clone https://github.com/TensorSpeech/TensorFlowTTS.git $ cd TensorFlowTTS $ pip install . ``` If you want to upgrade the repository and its dependencies: ```bash $ git pull $ pip install --upgrade . ``` # Supported Model architectures TensorFlowTTS currently provides the following architectures: 1. **MelGAN** released with the paper [MelGAN: Generative Adversarial Networks for Conditional Waveform Synthesis](https://arxiv.org/abs/1910.06711) by Kundan Kumar, Rithesh Kumar, Thibault de Boissiere, Lucas Gestin, Wei Zhen Teoh, Jose Sotelo, Alexandre de Brebisson, Yoshua Bengio, Aaron Courville. 2. **Tacotron-2** released with the paper [Natural TTS Synthesis by Conditioning WaveNet on Mel Spectrogram Predictions](https://arxiv.org/abs/1712.05884) by Jonathan Shen, Ruoming Pang, Ron J. Weiss, Mike Schuster, Navdeep Jaitly, Zongheng Yang, Zhifeng Chen, Yu Zhang, Yuxuan Wang, RJ Skerry-Ryan, Rif A. Saurous, Yannis Agiomyrgiannakis, Yonghui Wu. 3. **FastSpeech** released with the paper [FastSpeech: Fast, Robust, and Controllable Text to Speech](https://arxiv.org/abs/1905.09263) by Yi Ren, Yangjun Ruan, Xu Tan, Tao Qin, Sheng Zhao, Zhou Zhao, Tie-Yan Liu. 4. **Multi-band MelGAN** released with the paper [Multi-band MelGAN: Faster Waveform Generation for High-Quality Text-to-Speech](https://arxiv.org/abs/2005.05106) by Geng Yang, Shan Yang, Kai Liu, Peng Fang, Wei Chen, Lei Xie. 5. **FastSpeech2** released with the paper [FastSpeech 2: Fast and High-Quality End-to-End Text to Speech](https://arxiv.org/abs/2006.04558) by Yi Ren, Chenxu Hu, Xu Tan, Tao Qin, Sheng Zhao, Zhou Zhao, Tie-Yan Liu. 6. **Parallel WaveGAN** released with the paper [Parallel WaveGAN: A fast waveform generation model based on generative adversarial networks with multi-resolution spectrogram](https://arxiv.org/abs/1910.11480) by Ryuichi Yamamoto, Eunwoo Song, Jae-Min Kim. 7. **HiFi-GAN** released with the paper [HiFi-GAN: Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesis](https://arxiv.org/abs/2010.05646) by Jungil Kong, Jaehyeon Kim, Jaekyoung Bae. We are also implementing some techniques to improve quality and convergence speed from the following papers: 2. **Guided Attention Loss** released with the paper [Efficiently Trainable Text-to-Speech System Based on Deep Convolutional Networks with Guided Attention ](https://arxiv.org/abs/1710.08969) by Hideyuki Tachibana, Katsuya Uenoyama, Shunsuke Aihara. # Audio Samples Here in an audio samples on valid set. [tacotron-2](https://drive.google.com/open?id=1kaPXRdLg9gZrll9KtvH3-feOBMM8sn3_), [fastspeech](https://drive.google.com/open?id=1f69ujszFeGnIy7PMwc8AkUckhIaT2OD0), [melgan](https://drive.google.com/open?id=1mBwGVchwtNkgFsURl7g4nMiqx4gquAC2), [melgan.stft](https://drive.google.com/open?id=1xUkDjbciupEkM3N4obiJAYySTo6J9z6b), [fastspeech2](https://drive.google.com/drive/u/1/folders/1NG7oOfNuXSh7WyAoM1hI8P5BxDALY_mU), [multiband_melgan](https://drive.google.com/drive/folders/1DCV3sa6VTyoJzZmKATYvYVDUAFXlQ_Zp) # Tutorial End-to-End ## Prepare Dataset Prepare a dataset in the following format: ``` |- [NAME_DATASET]/ | |- metadata.csv | |- wavs/ | |- file1.wav | |- ... ``` Where `metadata.csv` has the following format: `id|transcription`. This is a ljspeech-like format; you can ignore preprocessing steps if you have other format datasets. Note that `NAME_DATASET` should be `[ljspeech/kss/baker/libritts]` for example. ## Preprocessing The preprocessing has two steps: 1. Preprocess audio features - Convert characters to IDs - Compute mel spectrograms - Normalize mel spectrograms to [-1, 1] range - Split the dataset into train and validation - Compute the mean and standard deviation of multiple features from the **training** split 2. Standardize mel spectrogram based on computed statistics To reproduce the steps above: ``` tensorflow-tts-preprocess --rootdir ./[ljspeech/kss/baker/libritts/thorsten] --outdir ./dump_[ljspeech/kss/baker/libritts/thorsten] --config preprocess/[ljspeech/kss/baker/thorsten]_preprocess.yaml --dataset [ljspeech/kss/baker/libritts/thorsten] tensorflow-tts-normalize --rootdir ./dump_[ljspeech/kss/baker/libritts/thorsten] --outdir ./dump_[ljspeech/kss/baker/libritts/thorsten] --config preprocess/[ljspeech/kss/baker/libritts/thorsten]_preprocess.yaml --dataset [ljspeech/kss/baker/libritts/thorsten] ``` Right now we only support [`ljspeech`](https://keithito.com/LJ-Speech-Dataset/), [`kss`](https://www.kaggle.com/bryanpark/korean-single-speaker-speech-dataset), [`baker`](https://weixinxcxdb.oss-cn-beijing.aliyuncs.com/gwYinPinKu/BZNSYP.rar), [`libritts`](http://www.openslr.org/60/) and [`thorsten`](https://github.com/thorstenMueller/deep-learning-german-tts) for dataset argument. In the future, we intend to support more datasets. **Note**: To run `libritts` preprocessing, please first read the instruction in [examples/fastspeech2_libritts](https://github.com/TensorSpeech/TensorFlowTTS/tree/master/examples/fastspeech2_libritts). We need to reformat it first before run preprocessing. After preprocessing, the structure of the project folder should be: ``` |- [NAME_DATASET]/ | |- metadata.csv | |- wav/ | |- file1.wav | |- ... |- dump_[ljspeech/kss/baker/libritts/thorsten]/ | |- train/ | |- ids/ | |- LJ001-0001-ids.npy | |- ... | |- raw-feats/ | |- LJ001-0001-raw-feats.npy | |- ... | |- raw-f0/ | |- LJ001-0001-raw-f0.npy | |- ... | |- raw-energies/ | |- LJ001-0001-raw-energy.npy | |- ... | |- norm-feats/ | |- LJ001-0001-norm-feats.npy | |- ... | |- wavs/ | |- LJ001-0001-wave.npy | |- ... | |- valid/ | |- ids/ | |- LJ001-0009-ids.npy | |- ... | |- raw-feats/ | |- LJ001-0009-raw-feats.npy | |- ... | |- raw-f0/ | |- LJ001-0001-raw-f0.npy | |- ... | |- raw-energies/ | |- LJ001-0001-raw-energy.npy | |- ... | |- norm-feats/ | |- LJ001-0009-norm-feats.npy | |- ... | |- wavs/ | |- LJ001-0009-wave.npy | |- ... | |- stats.npy | |- stats_f0.npy | |- stats_energy.npy | |- train_utt_ids.npy | |- valid_utt_ids.npy |- examples/ | |- melgan/ | |- fastspeech/ | |- tacotron2/ | ... ``` - `stats.npy` contains the mean and std from the training split mel spectrograms - `stats_energy.npy` contains the mean and std of energy values from the training split - `stats_f0.npy` contains the mean and std of F0 values in the training split - `train_utt_ids.npy` / `valid_utt_ids.npy` contains training and validation utterances IDs respectively We use suffix (`ids`, `raw-feats`, `raw-energy`, `raw-f0`, `norm-feats`, and `wave`) for each input type. **IMPORTANT NOTES**: - This preprocessing step is based on [ESPnet](https://github.com/espnet/espnet) so you can combine all models here with other models from ESPnet repository. - Regardless of how your dataset is formatted, the final structure of the `dump` folder **SHOULD** follow the above structure to be able to use the training script, or you can modify it by yourself 😄. ## Training models To know how to train model from scratch or fine-tune with other datasets/languages, please see detail at example directory. - For Tacotron-2 tutorial, pls see [examples/tacotron2](https://github.com/tensorspeech/TensorFlowTTS/tree/master/examples/tacotron2) - For FastSpeech tutorial, pls see [examples/fastspeech](https://github.com/tensorspeech/TensorFlowTTS/tree/master/examples/fastspeech) - For FastSpeech2 tutorial, pls see [examples/fastspeech2](https://github.com/tensorspeech/TensorFlowTTS/tree/master/examples/fastspeech2) - For FastSpeech2 + MFA tutorial, pls see [examples/fastspeech2_libritts](https://github.com/tensorspeech/TensorFlowTTS/tree/master/examples/fastspeech2_libritts) - For MelGAN tutorial, pls see [examples/melgan](https://github.com/tensorspeech/TensorFlowTTS/tree/master/examples/melgan) - For MelGAN + STFT Loss tutorial, pls see [examples/melgan.stft](https://github.com/tensorspeech/TensorFlowTTS/tree/master/examples/melgan.stft) - For Multiband-MelGAN tutorial, pls see [examples/multiband_melgan](https://github.com/tensorspeech/TensorFlowTTS/tree/master/examples/multiband_melgan) - For Parallel WaveGAN tutorial, pls see [examples/parallel_wavegan](https://github.com/tensorspeech/TensorFlowTTS/tree/master/examples/parallel_wavegan) - For Multiband-MelGAN Generator + Parallel WaveGAN Discriminator tutorial, pls see [examples/multiband_pwgan](https://github.com/tensorspeech/TensorFlowTTS/tree/master/examples/multiband_pwgan) - For HiFi-GAN tutorial, pls see [examples/hifigan](https://github.com/tensorspeech/TensorFlowTTS/tree/master/examples/hifigan) # Abstract Class Explaination ## Abstract DataLoader Tensorflow-based dataset A detail implementation of abstract dataset class from [tensorflow_tts/dataset/abstract_dataset](https://github.com/tensorspeech/TensorFlowTTS/blob/master/tensorflow_tts/datasets/abstract_dataset.py). There are some functions you need overide and understand: 1. **get_args**: This function return argumentation for **generator** class, normally is utt_ids. 2. **generator**: This function have an inputs from **get_args** function and return a inputs for models. **Note that we return a dictionary for all generator functions with the keys that exactly match with the model's parameters because base_trainer will use model(\*\*batch) to do forward step.** 3. **get_output_dtypes**: This function need return dtypes for each element from **generator** function. 4. **get_len_dataset**: Return len of datasets, normaly is len(utt_ids). **IMPORTANT NOTES**: - A pipeline of creating dataset should be: cache -> shuffle -> map_fn -> get_batch -> prefetch. - If you do shuffle before cache, the dataset won't shuffle when it re-iterate over datasets. - You should apply map_fn to make each element return from **generator** function have the same length before getting batch and feed it into a model. Some examples to use this **abstract_dataset** are [tacotron_dataset.py](https://github.com/tensorspeech/TensorFlowTTS/blob/master/examples/tacotron2/tacotron_dataset.py), [fastspeech_dataset.py](https://github.com/tensorspeech/TensorFlowTTS/blob/master/examples/fastspeech/fastspeech_dataset.py), [melgan_dataset.py](https://github.com/tensorspeech/TensorFlowTTS/blob/master/examples/melgan/audio_mel_dataset.py), [fastspeech2_dataset.py](https://github.com/TensorSpeech/TensorFlowTTS/blob/master/examples/fastspeech2/fastspeech2_dataset.py) ## Abstract Trainer Class A detail implementation of base_trainer from [tensorflow_tts/trainer/base_trainer.py](https://github.com/tensorspeech/TensorFlowTTS/blob/master/tensorflow_tts/trainers/base_trainer.py). It include [Seq2SeqBasedTrainer](https://github.com/tensorspeech/TensorFlowTTS/blob/master/tensorflow_tts/trainers/base_trainer.py#L265) and [GanBasedTrainer](https://github.com/tensorspeech/TensorFlowTTS/blob/master/tensorflow_tts/trainers/base_trainer.py#L149) inherit from [BasedTrainer](https://github.com/tensorspeech/TensorFlowTTS/blob/master/tensorflow_tts/trainers/base_trainer.py#L16). All trainer support both single/multi GPU. There a some functions you **MUST** overide when implement new_trainer: - **compile**: This function aim to define a models, and losses. - **generate_and_save_intermediate_result**: This function will save intermediate result such as: plot alignment, save audio generated, plot mel-spectrogram ... - **compute_per_example_losses**: This function will compute per_example_loss for model, note that all element of the loss **MUST** has shape [batch_size]. All models on this repo are trained based-on **GanBasedTrainer** (see [train_melgan.py](https://github.com/tensorspeech/TensorFlowTTS/blob/master/examples/melgan/train_melgan.py), [train_melgan_stft.py](https://github.com/tensorspeech/TensorFlowTTS/blob/master/examples/melgan.stft/train_melgan_stft.py), [train_multiband_melgan.py](https://github.com/tensorspeech/TensorFlowTTS/blob/master/examples/multiband_melgan/train_multiband_melgan.py)) and **Seq2SeqBasedTrainer** (see [train_tacotron2.py](https://github.com/tensorspeech/TensorFlowTTS/blob/master/examples/tacotron2/train_tacotron2.py), [train_fastspeech.py](https://github.com/tensorspeech/TensorFlowTTS/blob/master/examples/fastspeech/train_fastspeech.py)). # End-to-End Examples You can know how to inference each model at [notebooks](https://github.com/tensorspeech/TensorFlowTTS/tree/master/notebooks) or see a [colab](https://colab.research.google.com/drive/1akxtrLZHKuMiQup00tzO2olCaN-y3KiD?usp=sharing) (for English), [colab](https://colab.research.google.com/drive/1ybWwOS5tipgPFttNulp77P6DAB5MtiuN?usp=sharing) (for Korean). Here is an example code for end2end inference with fastspeech and melgan. ```python import numpy as np import soundfile as sf import yaml import tensorflow as tf from tensorflow_tts.inference import AutoConfig from tensorflow_tts.inference import TFAutoModel from tensorflow_tts.inference import AutoProcessor # initialize fastspeech model. fs_config = AutoConfig.from_pretrained('./examples/fastspeech/conf/fastspeech.v1.yaml') fastspeech = TFAutoModel.from_pretrained( config=fs_config, pretrained_path="./examples/fastspeech/pretrained/model-195000.h5" ) # initialize melgan model melgan_config = AutoConfig.from_pretrained('./examples/melgan/conf/melgan.v1.yaml') melgan = TFAutoModel.from_pretrained( config=melgan_config, pretrained_path="./examples/melgan/checkpoint/generator-1500000.h5" ) # inference processor = AutoProcessor.from_pretrained(pretrained_path="./test/files/ljspeech_mapper.json") ids = processor.text_to_sequence("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.") ids = tf.expand_dims(ids, 0) # fastspeech inference masked_mel_before, masked_mel_after, duration_outputs = fastspeech.inference( ids, speaker_ids=tf.zeros(shape=[tf.shape(ids)[0]], dtype=tf.int32), speed_ratios=tf.constant([1.0], dtype=tf.float32) ) # melgan inference audio_before = melgan.inference(masked_mel_before)[0, :, 0] audio_after = melgan.inference(masked_mel_after)[0, :, 0] # save to file sf.write('./audio_before.wav', audio_before, 22050, "PCM_16") sf.write('./audio_after.wav', audio_after, 22050, "PCM_16") ``` # Contact [Minh Nguyen Quan Anh](https://github.com/tensorspeech): nguyenquananhminh@gmail.com, [erogol](https://github.com/erogol): erengolge@gmail.com, [Kuan Chen](https://github.com/azraelkuan): azraelkuan@gmail.com, [Dawid Kobus](https://github.com/machineko): machineko@protonmail.com, [Takuya Ebata](https://github.com/MokkeMeguru): meguru.mokke@gmail.com, [Trinh Le Quang](https://github.com/l4zyf9x): trinhle.cse@gmail.com, [Yunchao He](https://github.com/candlewill): yunchaohe@gmail.com, [Alejandro Miguel Velasquez](https://github.com/ZDisket): xml506ok@gmail.com # License Overall, Almost models here are licensed under the [Apache 2.0](http://www.apache.org/licenses/LICENSE-2.0) for all countries in the world, except in **Viet Nam** this framework cannot be used for production in any way without permission from TensorFlowTTS's Authors. There is an exception, Tacotron-2 can be used with any purpose. If you are Vietnamese and want to use this framework for production, you **Must** contact us in advance. # Acknowledgement We want to thank [Tomoki Hayashi](https://github.com/kan-bayashi), who discussed with us much about Melgan, Multi-band melgan, Fastspeech, and Tacotron. This framework based-on his great open-source [ParallelWaveGan](https://github.com/kan-bayashi/ParallelWaveGAN) project.