Repository files navigation

An implementation of Tacotron speech synthesis in TensorFlow.

• Audio Samples from models trained using this repo.
  • The first set was trained for 441K steps on theLJ Speech Dataset
    • Speech started to become intelligible around 20K steps.
  • The second set was trained by@MXGray for 140K steps on theNancy Corpus.

1. @npuichigofixed a bug where dropout was not being applied in the prenet.

2. @begeekmyfriend created afork that adds location-sensitive attention and the stop token from theTacotron 2 paper. This can greatly reduce the amount of data required to train a model.

In April 2017, Google published a paper,Tacotron: Towards End-to-End Speech Synthesis,where they present a neural text-to-speech model that learns to synthesize speech directly from(text, audio) pairs. However, they didn't release their source code or training data. This is anindependent attempt to provide an open-source implementation of the model described in their paper.

The quality isn't as good as Google's demo yet, but hopefully it will get there someday :-).Pull requests are welcome!

1. Install Python 3.

2. Install the latest version ofTensorFlow for your platform. For betterperformance, install with GPU support if it's available. This code works with TensorFlow 1.3 and later.

3. Install requirements:

   pip install -r requirements.txt

1. Download and unpack a model:

   curl https://data.keithito.com/data/speech/tacotron-20180906.tar.gz | tar xzC /tmp

2. Run the demo server:

   python3 demo_server.py --checkpoint /tmp/tacotron-20180906/model.ckpt

3. Point your browser at localhost:9000

   • Type what you want to synthesize

Note: you need at least 40GB of free disk space to train a model.

1. Download a speech dataset.

   The following are supported out of the box:

   • LJ Speech (Public Domain)
   • Blizzard 2012 (Creative Commons Attribution Share-Alike)

   You can use other datasets if you convert them to the right format. SeeTRAINING_DATA.md for more info.

2. Unpack the dataset into~/tacotron

   After unpacking, your tree should look like this for LJ Speech:

   tacotron  |- LJSpeech-1.1      |- metadata.csv      |- wavs

   or like this for Blizzard 2012:

   tacotron  |- Blizzard2012      |- ATrampAbroad      |   |- sentence_index.txt      |   |- lab      |   |- wav      |- TheManThatCorruptedHadleyburg          |- sentence_index.txt          |- lab          |- wav

3. Preprocess the data

   python3 preprocess.py --dataset ljspeech

   • Use--dataset blizzard for Blizzard data

4. Train a model

   python3 train.py

   Tunable hyperparameters are found inhparams.py. You can adjust these at the commandline using the--hparams flag, for example--hparams="batch_size=16,outputs_per_step=2".Hyperparameters should generally be set to the same values at both training and eval time.The default hyperparameters are recommended for LJ Speech and other English-language data.SeeTRAINING_DATA.md for other languages.

5. Monitor with Tensorboard (optional)

   tensorboard --logdir ~/tacotron/logs-tacotron

   The trainer dumps audio and alignments every 1000 steps. You can find these in~/tacotron/logs-tacotron.

6. Synthesize from a checkpoint

   python3 demo_server.py --checkpoint ~/tacotron/logs-tacotron/model.ckpt-185000

   Replace "185000" with the checkpoint number that you want to use, then open a browsertolocalhost:9000 and type what you want to speak. Alternately, you canruneval.py at the command line:

   python3 eval.py --checkpoint ~/tacotron/logs-tacotron/model.ckpt-185000

   If you set the--hparams flag when training, set the same value here.

• TCMalloc seems to improvetraining speed and avoids occasional slowdowns seen with the default allocator. Youcan enable it by installing it and settingLD_PRELOAD=/usr/lib/libtcmalloc.so. With TCMalloc,you can get around 1.1 sec/step on a GTX 1080Ti.

• You can train withCMUDict by downloading thedictionary to ~/tacotron/training and then passing the flag--hparams="use_cmudict=True" totrain.py. This will allow you to pass ARPAbet phonemes enclosed in curly braces at evaltime to force a particular pronunciation, e.g.Turn left on {HH AW1 S S T AH0 N} Street.

• If you pass a Slack incoming webhook URL as the--slack_url flag to train.py, it will sendyou progress updates every 1000 steps.

• Occasionally, you may see a spike in loss and the model will forget how to attend (thealignments will no longer make sense). Although it will recover eventually, it maysave time to restart at a checkpoint prior to the spike by passing the--restore_step=150000 flag to train.py (replacing 150000 with a step number prior to thespike).Update: a recentfix to gradientclipping by @candlewill may have fixed this.

• During eval and training, audio length is limited tomax_iters * outputs_per_step * frame_shift_msmilliseconds. With the defaults (max_iters=200, outputs_per_step=5, frame_shift_ms=12.5), this is12.5 seconds.

  If your training examples are longer, you will see an error like this:Incompatible shapes: [32,1340,80] vs. [32,1000,80]

  To fix this, you can set a larger value ofmax_iters by passing--hparams="max_iters=300" totrain.py (replace "300" with a value based on how long your audio is and the formula above).

• Here is the expected loss curve when training on LJ Speech with the default hyperparameters:

• By Alex Barron:https://github.com/barronalex/Tacotron
• By Kyubyong Park:https://github.com/Kyubyong/tacotron