ml-talking-face / docs /description.md
deepkyu's picture
Add buymeacoffee link
5d56d4a
|
raw
history blame
2.78 kB

This system generates a talking face video based on the input text. You can provide the input text in one of the four languages: Chinese (Mandarin), English, Japanese, and Korean. You may also select the target language, the language of the output speech. If the input text language and the target language are different, the input text will be translated to the target language using Google Translate API.

Updates

(2023.10.20.) It has been a year since the demonstration has suddenly shut down by MINDsLab (MAUM.AI for now). And today, I'm happy to share that ⭐I have restored the demonstration⭐ in my own lambdalabs instance! Over the past year, there have been numerous advancements in Gen AI, including multilingual TTS and talking face generation. This demo may become "old-fashioned" at this time 😅... but I hope that it would help other researchers taking a journey in the same field.

⚠️By the way, I'm using A10G instance from lambdalabs with my own expense... I'm sorry, but I don't know when it will shut down again. 😵‍💫 I'll keep you posted on the status.

Buy Me A Coffee

(2022.06.17.) Thank you for visiting our demo!😊 This demo attracted a lot more attention than we anticipated. This, unfortunately, means that the computational burden is heavier than this demo was designed for. So, to maximize everyone's experience, we capped the length of the translated texts at:

  • 200 characters for English
  • 100 characters for Chinese, Japaense, and Korean.

(2022.06.17.) We were originally planning to support any input text. However, when checking the logs recently, we found that there were a lot of inappropriate input texts. So, we decided to filter the inputs based on toxicity using Perspective API @Google. Now, if you enter a possibily toxic text, the video generation will fail. We hope you understand.

(2022.06.05.) Due to the latency from HuggingFace Spaces and video rendering, it takes 15 ~ 30 seconds to get a video result.

Outdated updates

(2022.09.29.) The core part of the demonstration has been working on the AWS instance of MINDsLab, and I found that it can't connect to the instance now. I want to fix this issue, but I'm sorry to say that I left the company last week. I've contacted the company, but it takes some time to restore the session. If you're in a hurry, please send the e-mail directly to MINDsLab ([email protected]).
Whatever the reason, I'm sorry again. Hope you understand.