Multispeaker & Emotional TTS based on Tacotron 2 and Waveglow

Overview

Multispeaker & Emotional TTS based on Tacotron 2 and Waveglow

Table of Contents

General description

This Repository contains a sample code for Tacotron 2, WaveGlow with multi-speaker, emotion embeddings together with a script for data preprocessing.
Checkpoints and code originate from following sources:

Done:

  • took all the best code parts from all of the 5 sources above
  • clean the code and fixed some of the mistakes
  • change code structure
  • add multi-speaker and emotion embendings
  • add preprocessing
  • move all the configs from command line args into experiment config file under configs/experiments folder
  • add restoring / checkpointing mechanism
  • add tensorboard
  • make decoder work with n > 1 frames per step
  • make training work at FP16

TODO:

  • make it work with pytorch-1.4.0
  • add multi-spot instance training for AWS

Getting Started

The following section lists the requirements in order to start training the Tacotron 2 and WaveGlow models.

Clone the repository:

git clone https://github.com/ide8/tacotron2  
cd tacotron2
PROJDIR=$(pwd)
export PYTHONPATH=$PROJDIR:$PYTHONPATH

Requirements

This repository contains Dockerfile which extends the PyTorch NGC container and encapsulates some dependencies. Aside from these dependencies, ensure you have the following components:

Setup

Build an image from Docker file:

docker build --tag taco .

Run docker container:

docker run --shm-size=8G --runtime=nvidia -v /absolute/path/to/your/code:/app -v /absolute/path/to/your/training_data:/mnt/train -v /absolute/path/to/your/logs:/mnt/logs -v /absolute/path/to/your/raw-data:/mnt/raw-data -v /absolute/path/to/your/pretrained-checkpoint:/mnt/pretrained -detach taco sleep inf

Check container id:

docker ps

Select container id of image with tag taco and log into container with:

docker exec -it container_id bash

Code structure description

Folders tacotron2 and waveglow have scripts for Tacotron 2, WaveGlow models and consist of:

  • /model.py - model architecture
  • /data_function.py - data loading functions
  • /loss_function.py - loss function

Folder common contains common layers for both models (common/layers.py), utils (common/utils.py) and audio processing (common/audio_processing.py and common/stft.py).

Folder router is used by training script to select an appropriate model

In the root directory:

  • train.py - script for model training
  • preprocess.py - performs audio processing and creates training and validation datasets
  • inference.ipynb - notebook for running inference

Folder configs contains __init__.py with all parameters needed for training and data processing. Folder configs/experiments consists of all the experiments. waveglow.py and tacotron2.py are provided as examples for WaveGlow and Tacotron 2. On training or data processing start, parameters are copied from your experiment (in our case - from waveglow.py or from tacotron2.py) to __init__.py, from which they are used by the system.

Data preprocessing

Preparing for data preprocessing

  1. For each speaker you have to have a folder named with speaker name, containing wavs folder and metadata.csv file with the next line format: file_name.wav|text.
  2. All necessary parameters for preprocessing should be set in configs/experiments/waveglow.py or in configs/experiments/tacotron2.py, in the class PreprocessingConfig.
  3. If you're running preprocessing first time, set start_from_preprocessed flag to False. preprocess.py performs trimming of audio files up to PreprocessingConfig.top_db (cuts the silence in the beginning and the end), applies ffmpeg command in order to mono, make same sampling rate and bit rate for all the wavs in dataset.
  4. It saves a folder wavs with processed audio files and data.csv file in PreprocessingConfig.output_directory with the following format: path|text|speaker_name|speaker_id|emotion|text_len|duration.
  5. Trimming and ffmpeg command are applied only to speakers, for which flag process_audio is True. Speakers with flag emotion_present is False, are treated as with emotion neutral-normal.
  6. You won't need start_from_preprocessed = False once you finish running preprocessing script. Only exception in case of new raw data comes in.
  7. Once start_from_preprocessed is set to True, script loads file data.csv (created by the start_from_preprocessed = False run), and forms train.txt and val.txt out from data.csv.
  8. Main PreprocessingConfig parameters:
    1. cpus - defines number of cores for batch generator
    2. sr - defines sample ratio for reading and writing audio
    3. emo_id_map - dictionary for emotion name to emotion_id mapping
    4. data[{'path'}] - is path to folder named with speaker name and containing wavs folder and metadata.csv with the following line format: file_name.wav|text|emotion (optional)
  9. Preprocessing script forms training and validation datasets in the following way:
    1. selects rows with audio duration and text length less or equal those for speaker PreprocessingConfig.limit_by (this step is needed for proper batch size)
    2. if such speaker is not present, than it selects rows within PreprocessingConfig.text_limit and PreprocessingConfig.dur_limit. Lower limit for audio is defined by PreprocessingConfig.minimum_viable_dur
    3. in order to be able to use the same batch size as NVIDIA guys, set PreprocessingConfig.text_limit to linda_jonson
    4. splits dataset randomly by ratio train : val = 0.95 : 0.05
    5. if speaker train set is bigger than PreprocessingConfig.n - samples n rows
    6. saves train.txt and val.txt to PreprocessingConfig.output_directory
    7. saves emotion_coefficients.json and speaker_coefficients.json with coefficients for loss balancing (used by train.py).

Run preprocessing

Since both scripts waveglow.py and tacotron2.py contain the class PreprocessingConfig, training and validation dataset can be produced by running any of them:

python preprocess.py --exp tacotron2

or

python preprocess.py --exp waveglow

Training

Preparing for training

Tacotron 2

In configs/experiment/tacotron2.py, in the class Config set:

  1. training_files and validation_files - paths to train.txt, val.txt;
  2. tacotron_checkpoint - path to pretrained Tacotron 2 if it exist (we were able to restore Waveglow from Nvidia, but Tacotron 2 code was edited to add speakers and emotions, so Tacotron 2 needs to be trained from scratch);
  3. speaker_coefficients - path to speaker_coefficients.json;
  4. emotion_coefficients - path to emotion_coefficients.json;
  5. output_directory - path for writing logs and checkpoints;
  6. use_emotions - flag indicating emotions usage;
  7. use_loss_coefficients - flag indicating loss scaling due to possible data disbalance in terms of both speakers and emotions; for balancing loss, set paths to jsons with coefficients in emotion_coefficients and speaker_coefficients;
  8. model_name - "Tacotron2".
  • Launch training
    • Single gpu:
      python train.py --exp tacotron2
      
    • Multigpu training:
      python -m multiproc train.py --exp tacotron2
      

WaveGlow:

In configs/experiment/waveglow.py, in the class Config set:

  1. training_files and validation_files - paths to train.txt, val.txt;
  2. waveglow_checkpoint - path to pretrained Waveglow, restored from Nvidia. Download checkopoint.
  3. output_directory - path for writing logs and checkpoints;
  4. use_emotions - False;
  5. use_loss_coefficients - False;
  6. model_name - "WaveGlow".
  • Launch training
    • Single gpu:
      python train.py --exp waveglow
      
    • Multigpu training:
      python -m multiproc train.py --exp waveglow
      

Running Tensorboard

Once you made your model start training, you might want to see some progress of training:

docker ps

Select container id of image with tag taco and run:

docker exec -it container_id bash

Start Tensorboard:

 tensorboard --logdir=path_to_folder_with_logs --host=0.0.0.0

Loss is being written into tensorboard:

Tensorboard Scalars

Audio samples together with attention alignments are saved into tensorbaord each Config.epochs_per_checkpoint. Transcripts for audios are listed in Config.phrases

Tensorboard Audio

Inference

Running inference with the inference.ipynb notebook.

Run Jupyter Notebook:

jupyter notebook --ip 0.0.0.0 --port 6006 --no-browser --allow-root

output:

[email protected]:/app# jupyter notebook --ip 0.0.0.0 --port 6006 --no-browser --allow-root
[I 09:31:25.393 NotebookApp] JupyterLab extension loaded from /opt/conda/lib/python3.6/site-packages/jupyterlab
[I 09:31:25.393 NotebookApp] JupyterLab application directory is /opt/conda/share/jupyter/lab
[I 09:31:25.395 NotebookApp] Serving notebooks from local directory: /app
[I 09:31:25.395 NotebookApp] The Jupyter Notebook is running at:
[I 09:31:25.395 NotebookApp] http://(04096a19c266 or 127.0.0.1):6006/?token=bbd413aef225c1394be3b9de144242075e651bea937eecce
[I 09:31:25.395 NotebookApp] Use Control-C to stop this server and shut down all kernels (twice to skip confirmation).
[C 09:31:25.398 NotebookApp] 
    
    To access the notebook, open this file in a browser:
        file:///root/.local/share/jupyter/runtime/nbserver-15398-open.html
    Or copy and paste one of these URLs:
        http://(04096a19c266 or 127.0.0.1):6006/?token=bbd413aef225c1394be3b9de144242075e651bea937eecce

Select adress with 127.0.0.1 and put it in the browser. In this case: http://127.0.0.1:6006/?token=bbd413aef225c1394be3b9de144242075e651bea937eecce

This script takes text as input and runs Tacotron 2 and then WaveGlow inference to produce an audio file. It requires pre-trained checkpoints from Tacotron 2 and WaveGlow models, input text, speaker_id and emotion_id.

Change paths to checkpoints of pretrained Tacotron 2 and WaveGlow in the cell [2] of the inference.ipynb.
Write a text to be displayed in the cell [7] of the inference.ipynb.

Parameters

In this section, we list the most important hyperparameters, together with their default values that are used to train Tacotron 2 and WaveGlow models.

Shared parameters

  • epochs - number of epochs (Tacotron 2: 1501, WaveGlow: 1001)
  • learning-rate - learning rate (Tacotron 2: 1e-3, WaveGlow: 1e-4)
  • batch-size - batch size (Tacotron 2: 64, WaveGlow: 11)
  • grad_clip_thresh - gradient clipping treshold (0.1)

Shared audio/STFT parameters

  • sampling-rate - sampling rate in Hz of input and output audio (22050)
  • filter-length - (1024)
  • hop-length - hop length for FFT, i.e., sample stride between consecutive FFTs (256)
  • win-length - window size for FFT (1024)
  • mel-fmin - lowest frequency in Hz (0.0)
  • mel-fmax - highest frequency in Hz (8.000)

Tacotron parameters

  • anneal-steps - epochs at which to anneal the learning rate (500/ 1000/ 1500)
  • anneal-factor - factor by which to anneal the learning rate (0.1) These two parameters are used to change learning rate at the points defined in anneal-steps according to:
    learning_rate = learning_rate * ( anneal_factor ** p),
    where p = 0 at the first step and increments by 1 each step.

WaveGlow parameters

  • segment-length - segment length of input audio processed by the neural network (8000). Before passing to input, audio is padded or croped to segment-length.
  • wn_config - dictionary with parameters of affine coupling layers. Contains n_layers, n_chanels, kernel_size.

Contributing

If you've ever wanted to contribute to open source, and a great cause, now is your chance!

See the contributing docs for more information

Owner
Ivan Didur
CTO at data root labs
Ivan Didur
AI Assistant for Building Reliable, High-performing and Fair Multilingual NLP Systems

AI Assistant for Building Reliable, High-performing and Fair Multilingual NLP Systems

Microsoft 37 Nov 29, 2022
An open source library for deep learning end-to-end dialog systems and chatbots.

DeepPavlov is an open-source conversational AI library built on TensorFlow, Keras and PyTorch. DeepPavlov is designed for development of production re

Neural Networks and Deep Learning lab, MIPT 6k Dec 30, 2022
Ceaser-Cipher - The Caesar Cipher technique is one of the earliest and simplest method of encryption technique

Ceaser-Cipher The Caesar Cipher technique is one of the earliest and simplest me

Lateefah Ajadi 2 May 12, 2022
A simple visual front end to the Maya UE4 RBF plugin delivered with MetaHumans

poseWrangler Overview PoseWrangler is a simple UI to create and edit pose-driven relationships in Maya using the MayaUE4RBF plugin. This plugin is dis

Christopher Evans 105 Dec 18, 2022
LeBenchmark: a reproducible framework for assessing SSL from speech

LeBenchmark: a reproducible framework for assessing SSL from speech

11 Nov 30, 2022
Ελληνικά νέα (Python script) / Greek News Feed (Python script)

Ελληνικά νέα (Python script) / Greek News Feed (Python script) Ελληνικά English Το 2017 είχα υλοποιήσει ένα Python script για να εμφανίζει τα τωρινά ν

Loren Kociko 1 Jun 14, 2022
Knowledge Graph,Question Answering System,基于知识图谱和向量检索的医疗诊断问答系统

Knowledge Graph,Question Answering System,基于知识图谱和向量检索的医疗诊断问答系统

wangle 823 Dec 28, 2022
A highly sophisticated sequence-to-sequence model for code generation

CoderX A proof-of-concept AI system by Graham Neubig (June 30, 2021). About CoderX CoderX is a retrieval-based code generation AI system reminiscent o

Graham Neubig 39 Aug 03, 2021
Which Apple Keeps Which Doctor Away? Colorful Word Representations with Visual Oracles

Which Apple Keeps Which Doctor Away? Colorful Word Representations with Visual Oracles (TASLP 2022)

Zhuosheng Zhang 3 Apr 14, 2022
A music comments dataset, containing 39,051 comments for 27,384 songs.

Music Comments Dataset A music comments dataset, containing 39,051 comments for 27,384 songs. For academic research use only. Introduction This datase

Zhang Yixiao 2 Jan 10, 2022
Multilingual Emotion classification using BERT (fine-tuning). Published at the WASSA workshop (ACL2022).

XLM-EMO: Multilingual Emotion Prediction in Social Media Text Abstract Detecting emotion in text allows social and computational scientists to study h

MilaNLP 35 Sep 17, 2022
Snips Python library to extract meaning from text

Snips NLU Snips NLU (Natural Language Understanding) is a Python library that allows to extract structured information from sentences written in natur

Snips 3.7k Dec 30, 2022
A sentence aligner for comparable corpora

About Yalign is a tool for extracting parallel sentences from comparable corpora. Statistical Machine Translation relies on parallel corpora (eg.. eur

Machinalis 128 Aug 24, 2022
NewsMTSC: (Multi-)Target-dependent Sentiment Classification in News Articles

NewsMTSC: (Multi-)Target-dependent Sentiment Classification in News Articles NewsMTSC is a dataset for target-dependent sentiment classification (TSC)

Felix Hamborg 79 Dec 30, 2022
📔️ Generate a text-based journal from a template file.

JGen 📔️ Generate a text-based journal from a template file. Contents Getting Started Example Overview Usage Details Reserved Keywords Gotchas Getting

Harrison Broadbent 21 Sep 25, 2022
Connectionist Temporal Classification (CTC) decoding algorithms: best path, beam search, lexicon search, prefix search, and token passing. Implemented in Python.

CTC Decoding Algorithms Update 2021: installable Python package Python implementation of some common Connectionist Temporal Classification (CTC) decod

Harald Scheidl 736 Jan 03, 2023
Code for EMNLP'21 paper "Types of Out-of-Distribution Texts and How to Detect Them"

Code for EMNLP'21 paper "Types of Out-of-Distribution Texts and How to Detect Them"

Udit Arora 19 Oct 28, 2022
Official code for Spoken ObjectNet: A Bias-Controlled Spoken Caption Dataset

Official code for our Interspeech 2021 - Spoken ObjectNet: A Bias-Controlled Spoken Caption Dataset [1]*. Visually-grounded spoken language datasets c

Ian Palmer 3 Jan 26, 2022
English loanwords in the world's languages

Wiktionary as CLDF Content cldf1 and cldf2 contain cldf-conform data sets with a total of 2 377 756 entries about the vocabulary of all 1403 languages

Viktor Martinović 3 Jan 14, 2022
chaii - hindi & tamil question answering

chaii - hindi & tamil question answering This is the solution for rank 5th in Kaggle competition: chaii - Hindi and Tamil Question Answering. The comp

abhishek thakur 33 Dec 18, 2022