Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
A TensorFlow Implementation of Tacotron: A Fully End-to-End Text-To-Speech Synthesis Model
| Date | Stars |
|---|---|
| 2026-07-24 | 1833 |
| 2026-07-25 | 1833 |
| 2026-07-28 | 1833 |
| 2026-07-30 | 1833 |
| 2026-07-31 | 1833 |
| 2026-08-06 | 1833 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# A (Heavily Documented) TensorFlow Implementation of Tacotron: A Fully End-to-End Text-To-Speech Synthesis Model ## Requirements * NumPy >= 1.11.1 * TensorFlow >= 1.3 * librosa * tqdm * matplotlib * scipy ## Data <img src="https://upload.wikimedia.org/wikipedia/commons/7/72/World_English_Bible_Cover.jpg" height="200" align="right"> <img src="https://upload.wikimedia.org/wikipedia/commons/thumb/f/f6/Nick_Offerman_at_UMBC_%28cropped%29.jpg/440px-Nick_Offerman_at_UMBC_%28cropped%29.jpg" height="200" align="right"> <img src="https://image.shutterstock.com/z/stock-vector-lj-letters-four-colors-in-abstract-background-logo-design-identity-in-circle-alphabet-letter-418687846.jpg" height="200" align="right"> We train the model on three different speech datasets. 1. [LJ Speech Dataset](https://keithito.com/LJ-Speech-Dataset/) 2. [Nick Offerman's Audiobooks](https://www.audible.com.au/search?searchNarrator=Nick+Offerman) 3. [The World English Bible](https://www.kaggle.com/bryanpark/the-world-english-bible-speech-dataset) LJ Speech Dataset is recently widely used as a benchmark dataset in the TTS task because it is publicly available. It has 24 hours of reasonable quality samples. Nick's audiobooks are additionally used to see if the model can learn even with less data, variable speech samples. They are 18 hours long. [The World English Bible](https://en.wikipedia.org/wiki/World_English_Bible) is a public domain update of the American Standard Version of 1901 into modern English. Its original audios are freely available [here](http://www.audiotreasure.com/webindex.htm). Kyubyong split each chapter by verse manually and aligned the segmented audio clips to the text. They are 72 hours in total. You can download them at [Kaggle Datasets](https://www.kaggle.com/bryanpark/the-world-english-bible-speech-dataset). ## Training * STEP 0. Download [LJ Speech Dataset](https://keithito.com/LJ-Speech-Dataset/) or prepare your own data. * STEP 1. Adjust hyper parameters in `hyperparams.py`. (If you want to do preprocessing, set `prepro` True`. * STEP 2. Run `python train.py`. (If you set `prepro` True, run `python prepro.py` first) * STEP 3. Run `python eval.py` regularly during training. ## Sample Synthesis We generate speech samples based on [Harvard Sentences](http://www.cs.columbia.edu/~hgs/audio/harvard.html) as the original paper does. It is already included in the repo. * Run `python synthesize.py` and check the files in `samples`. ## Training Curve <img src="fig/training_curve.png"> ## Attention Plot <img src="fig/attention.gif"> ## Generated Samples * [LJ at 200k steps](https://soundcloud.com/kyubyong-park/sets/tacotron_lj_200k) * [Nick at 215k steps](https://soundcloud.com/kyubyong-park/sets/tacotron_nick_215k) * [WEB at 183k steps](https://soundcloud.com/kyubyong-park/sets/tacotron_web_183k) ## Pretrained Files * Keep in mind 200k steps may not be enough for the best performance. * [LJ 200k](https://www.dropbox.com/s/8kxa3xh2vfna3s9/LJ_logdir.zip?dl=0) * [WEB 200k](https://www.dropbox.com/s/g7m6xhd350ozkz7/WEB_logdir.zip?dl=0) ## Notes * It's important to monitor the attention plots during training. If the attention plots look good (alignment looks linear), and then they look bad (the plots will look similar to what they looked like in the begining of training), then training has gone awry and most likely will need to be restarted from a checkpoint where the attention looked good, because we've learned that it's unlikely that the loss will ever recover. This deterioration of attention will correspond with a spike in the loss. * In the original paper, the authors said, "An important trick we discovered was predicting multiple, non-overlapping output frames at each decoder step" where the number of of multiple frame is the reduction factor, `r`. We originally interpretted this as predicting non-sequential frames during each decoding step `t`. Thus were using the following
Excerpt of 6,147 characters
Read on GitHub
Kyubyong Park · TUNiB · South Korea
107
Yunchao He · China
6
Qingying Chen · Peking University · China
3
3
Sungkyun Chang · Centre for Digital Music, QMUL
1
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:183b36d3f586024d, topic:tts, desc:text-to-speech, readme:text-to-speech
matched fp:183b36d3f586024d, topic:tensorflow