laitimes

Google's first AI codec to process speech, SoundStream's latest technology detailed, efficient compression and noise reduction

author:Smart stuff
Google's first AI codec to process speech, SoundStream's latest technology detailed, efficient compression and noise reduction

Zhi DongXi (public number: zhidxcom)

Compile the | Li Huinan

Edit | Jiang Xinbai

According to foreign media VentureBeat, on August 12, US time, Google conducted a technical detailed explanation of its audio codec SoundStream, which can not only process different types of sounds, but also provide high-quality audio. SoundStream is also the first AI codec to handle speech and music, and the codec can also run on smartphones.

SoundStream is an end-to-end "neural" audio codec that can process audio including speech, music, and ambient sounds. At the same time, SoundStream simultaneously compresses and enhances audio to eliminate noise in the background.

According to Google, the performance of the 3kbps SoundStream is close to that of the 9.6kbps US EVS processor, and the performance exceeds the 12kbps Opus codec. In addition, at the same bit rate, SoundStream performs better than the current version of Lyra.

Users can use SoundStream to compress audio, alleviating the need for high storage and bandwidth. At the same time, the decoded audio is not significantly different in perception from the original audio.

In traditional audio processing pipelines, compressed audio and enhanced audio are often performed by different plates. But SoundStream is both compressed and enhanced.

In May, Google released a neural audio codec called Lyra that can be used to compress low bitrate audio. Lyra is a system built by SoundStream using encoders, decoders, and quantizers.

But Google said that SoundStream is still in the experimental stage, and the follow-up plan is to update the version of Lyra, which will have higher audio quality and less complexity.

"When people are transmitting audio, effective compression is necessary. SoundStream is an important step in improving machine-driven audio codecs, which have outperformed the most advanced Opus and EVS codecs and can enhance audio as needed. Google researcher Neil Zeghidour said.

Marco Tagliasacchi, another Google researcher, also wrote in a blog that by integrating SoundStream with Lyra, developers can use existing tools to develop, which is a good use of resources and provides better sound quality.

<h2>Conclusion: The follow-up version of SoundStream is worth consumers looking forward to</h2>

It is reported that SoundStream currently has the functions and features of the best audio codec on the market. SoundStream's highly efficient audio processor not only saves consumer time, but also provides better sound quality. Therefore, SoundStream may be more popular with consumers in the market for audio codecs.

Although SoundStream is still in the experimental stage, as Google advances the technology development, SoundStream may be widely used in the future.

Source: VentureBeat

Google's first AI codec to process speech, SoundStream's latest technology detailed, efficient compression and noise reduction

Read on