search

kaiidams / Kokoro-Speech-Dataset

A public domain single speaker Japanese speech dataset
38 stars 3 forks source link

Cannot download pretrained model #13

Closed NidhiBharani closed 2 years ago

NidhiBharani commented 2 years ago

Hi, I got to know about your work through this discussion thread in coquiTTS : https://github.com/coqui-ai/TTS/discussions/930 and found your dataset included in TTS/recipes as well. I am interested in Japanese TTS as well so thanks for sharing your work!

Description

I tried to download the pretrained model but the link doesn't seem to be working. I am attaching a screenshot. Screenshot from 2022-05-07 21-41-16

Steps to reproduce error

  1. Tried wget: wget https://github.com/kaiidams/Kokoro-Speech-Dataset/releases/download/1.0/tacotron-kokoro-20210324.zip
  2. Also searched through all releases and tags.
kaiidams commented 2 years ago

@NidhiBharani Added the pretrained model to https://github.com/kaiidams/Kokoro-Speech-Dataset/releases/download/1.0/tacotron-kokoro-20210324.zip.

The model is based on Tacotron 1 + TensorFlow 1.x and the model is not trained long enough. If you are looking for newer model, you can look at this recipe https://github.com/coqui-ai/TTS/tree/main/recipes/kokoro/tacotron2-DDC. Coqui-TTS has a Kokoro model with Tacotron 2 + HiFi GAN.

kaiidams commented 2 years ago

Download link is correct now.