Home > News content

Damo Academy has released the industry's first special AI-FPGA chip for speech synthesis algorithm design Ouroboros

via:博客园     time:2019/8/21 21:34:24     readed:248

data-croporisrc=https://mmbiz.qpic.cn/mmbiz_jpg/YjhmbbkdV6tbJg67yHL6nvGGPVVkZqCcfgvQB8UrqOW77BYaMFmHmmSK4tLZhadaoeicrAOXNKhMzTPLVzicptMQ/0?wx_fmt=jpeg

This is the industry's first AI FPGA chip structure designed for speech synthesis algorithms, which can increase the computational efficiency of speech generation algorithms by more than 100 times.

Text: Bao Yonggang

Leifeng Network News, Hot chips 31 (2019) is being held in San Francisco, USA. On the second day of the summit, Alibaba brought a speech on "Ouroboros: A WaveNet Inference Engine for TTS Applications on Embedded Devices" and released a new generation of AI voice FPGA chips. Technical Ouroboros.

According to Alibaba, this is the industry's first AI FPGA chip design dedicated to speech synthesis algorithms, which can increase the computational efficiency of speech generation algorithms by more than 100 times.

According to the Adriatic Motors Sweeper, WaveNet generates 1 second of voice using the AI ​​speech synthesis algorithm. The CPU and GPU require 50 seconds of computing time, but Ouroboros is only 0.3 seconds in the FPGA environment. A major breakthrough for Ouroboros is the replacement of cloud servers with custom hardware acceleration technology, which avoids strong dependence on network connections and cloud services.

Performance simulations for ASICs show that Ouroboros is designed to run real-time speech-to-speech (TTS) algorithms such as WaveNet in real time for real-time speech synthesis.

It is also known that Ouroboros technology is also applicable to the new generation of speech synthesis algorithm KAN-TTS released by Dharma in July this year. The algorithm increases the similarity between synthesized speech and original speech in commercial systems to over 97%.

It is also reported that the Ouroboros technology is also applicable to the new generation of speech synthesis algorithm KAN-TTS released by Dharma in July this year. The algorithm increases the similarity between synthesized speech and original speech in commercial systems to over 97%. In addition to speech synthesis, Ouroboros chip technology will also support AI speech recognition. Based on ouroboros research and development of a complete voice AI chip, it is expected to be the first to land on the Tmall Elf.

Lei Feng.com noted that, like other chip products released by Ali recently, the naming of this product is also very distinctive. Ouroboros Chinese is a serpent, a symbol that has been handed down from ancient times. The image is a snake (or dragon) that swallows its own tail, resulting in a ring (sometimes also shown as a twisted pattern, ie “∞” ), the name of the name is "self-devourer". This symbol has always had a lot of different symbolic meanings, and the most acceptable one is “Infinity”, “Circular” and so on.

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments