GithubHelp home page GithubHelp logo

hcnoh / multi-speaker-tacotron-tensorflow Goto Github PK

View Code? Open in Web Editor NEW
0.0 0.0 0.0 516 KB

A TensorFlow implementation of Multi-Speaker Tacotron which was introduced on Deep Voice 2 paper by Baidu.

Python 100.00%
tensorflow-models python3 text-to-speech tacotron tensorflow deep-learning

multi-speaker-tacotron-tensorflow's Introduction

Multi-Speaker Tacotron Implementation in TensorFlow

Multi-Speaker 기능을 가진 Tacotron 모델의 TensorFlow 구현입니다. 이 저장소는 Baidu의 Deep Voice 2 논문을 기반으로 구현하였습니다. Multi-Speaker 기능은 선택적으로 사용, 또는 사용하지 않을 수 있으며 만약 Multi-Speaker 기능을 사용하지 않는다면 기존의 Tacotron 모델과 동일합니다.

Installing Dependencies

  1. 파이썬 3.5 버전을 설치합니다.

  2. TensorFlow 1.10.0 버전 설치합니다. GPU가 사용가능하다면 GPU 버전을, 그렇지 않다면 CPU 버전을 설치합니다. 참고로 Generating을 수행하는 경우에는 CPU 버전으로도 충분합니다.

  3. 파이썬 패키지(Requirements)를 설치합니다. 설치는 아래와 같은 명령을 통해 수행 가능합니다.

    $ pip install -r requirements.txt

    아래와 같이 virtualenv를 활용하여 더 쉽게 패키지 관리를 할 수도 있습니다. virtualenv가 이미 설치되어 있다는 가정 하에 아래의 명령을 수행하면 됩니다.

    $ virtualenv --system-site-packages -p python3 tacotron-venv
    $ source ./tacotron-venv/bin/activate
    $ pip install -r requirements.txt

Training

Note: 데이터 세트를 다운로드받기 위한 여유분의 저장 공간이 필요합니다. LJ Speech 데이터 세트를 기준으로 약 80GB 이상의 저장 공간이 필요합니다.

  1. 음성 데이터 세트를 다운로드합니다.

    이 저장소는 다음의 데이터 세트에 대한 활용을 제공합니다.

  2. 데이터 세트를 원하는 위치에 Unpack합니다.

    Unpacking 이후에는 디렉토리 구조가 다음과 같아야 합니다.

    dataset
    ├── LJSpeech
        ├── wavs
            ├── LJ001-0001.wav (name doesn't matter)
            ├── LJ001-0002.wav
            └── ...
        ├── metadata.csv
        └── README
    ├── kss
        ├── 1
            ├── 1_0000.wav (name doesn't matter)
            ├── 1_0001.wav
            └── ...
        ├── 2
        ├── 3
        ├── 4
        ├── transcript.txt
        └── transcript.v.1.1.txt
    ├── tfrecord
    ├── VCTK
        ├── VCTK-Corpus
            ├── txt
                ├── p225
                    ├── p225_001.txt (name doesn't matter)
                    ├── p226_002.txt
                    └── ...
                ├── p226
                └── ...
            ├── wav
                ├── p225
                    ├── p225_001.wav (name doesn't matter)
                    ├── p226_002.wav
                    └── ...
                ├── p226
                └── ...
            ├── COPYING
            ├── NOTE
            ├── README
            └── speaker-info.txt
        ├── license_text
        └── README.txt
    └── ...
    
  3. hyparams.py의 스크립트에서 Training을 위한 데이터 세트 이름 및 각각의 Path들을 수정하여 줍니다.

    dataset_paths같은 경우는 LJ Speech와 kss의 경우 LJSpeechkss 디렉토리까지 명시해줘야 하며, VCTK 같은 경우는 VCTK-Corpus 디렉토리까지 명시해줘야 합니다.

  4. 데이터 세트 전처리를 수행합니다.

    $ python preprocess.py
  5. 모델 Training을 수행합니다.

    $ python train.py

Generating

학습된 모델을 이용하여 음성을 생성하기 위해서는 generate.py를 실행하면 됩니다. hyparams.pyfor generating 부분을 원하는 경로로 바꿔주고 바로 실행하면 됩니다. Multi-Speaker 기능 사용 여부에 따라서 실행 방식이 달라질 것이지만 사용하는데 어려움이 없을 것입니다.

$ python generate.py

Results

106000 Steps:

382000 Steps:

Recent Updates

  1. 현재 Multi-Speaker 기능 실험을 위한 여러 기능들이 구현되어 있지만 지금 당장은 사용하기 어려운 상태입니다.
  2. Results 작성을 완료하였습니다.

References

multi-speaker-tacotron-tensorflow's People

Contributors

hcnoh avatar

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.