From 5559fbd3e3ed60324a4cbbc992ea706860c5de16 Mon Sep 17 00:00:00 2001 From: Taylor Jung Date: Tue, 19 May 2020 16:22:30 +0900 Subject: [PATCH] update readme to help users who is migrating data Users who is migrating their data from NVIDIA's tacotron 2 repository will have to change their filelist format in order to train their own data. --- README.md | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/README.md b/README.md index 0ee74b0..1c29363 100644 --- a/README.md +++ b/README.md @@ -34,8 +34,9 @@ Visit our [website] for audio samples. ## Training 1. Update the filelists inside the filelists folder to point to your data -2. `python train.py -c config.json -p train_config.output_directory=outdir` -3. (OPTIONAL) `tensorboard --logdir=outdir/logdir` +2. (OPTIONAL) If your filelist have no speaker ids inside, you can add dummy speaker id. `sed -e 's/$/|0/' -i your_filelist.txt` +3. `python train.py -c config.json -p train_config.output_directory=outdir` +4. (OPTIONAL) `tensorboard --logdir=outdir/logdir` ## Training using a pre-trained model Training using a pre-trained model can lead to faster convergence. @@ -68,4 +69,4 @@ Seetharaman](https://github.com/pseeth/pytorch-stft) as described in our code. [PyTorch]: https://github.com/pytorch/pytorch#installation [website]: https://nv-adlr.github.io/Flowtron [AMP]: https://github.com/NVIDIA/apex/tree/master/apex/amp -[Tacotron]: https://arxiv.org/abs/1712.05884 \ No newline at end of file +[Tacotron]: https://arxiv.org/abs/1712.05884