A few days ago the Google developers released via a blog post that they have taken the decision to make Lyra open source. Lyra is based on machine learning to enable high-quality voice calls in low-bandwidth situations.
With which this benefits and allows other developers to feed their applications communication and develop Lyra in new directions.
A staple of media applications for decades, codecs have enabled bandwidth-intensive applications to transmit data efficiently.
As such, codec development, for both video and audio, presents an ongoing challenge- Provide ever-higher quality, use less data, and minimize latency for real-time communication.
Although video may appear to consume much more bandwidth than audio, modern video codecs can achieve lower bit rates than some of the high-quality speech codecs in use today.
Combining Low bit rate voice and video codecs can bring high quality video calling experience even on low bandwidth networks. However, historically, the lower the bit rate of an audio codec, the less intelligible the voice signal is and the more robotic it is.
Also, although some people have access to a consistent high-quality broadband network, this level of connectivity is not universal, and even people who live in well-connected areas sometimes face poor network connections, poor network connections, and connectivity. deficient.
To solve this problem, Google created Lyra, a high-quality, ultra-low-bit-rate speech codec which makes voice communication available even on the slowest networks.
To do this, Google applied traditional coding techniques while taking advantage of advances in machine learning with models trained over thousands of hours of data to create a new method of compression and transmission of voice signals.
Lyra's code is written in C ++ for speed, efficiency and interoperability, plus it uses the Bazel framework with Abseil and the GoogleTest framework for full unit testing.
The Basic API provides an interface for encoding and decoding at the packet and file level. The complete signal processing toolchain is also provided and includes various filters and transforms.
“Our sample application integrates with the Android NDK to show how to integrate Lyra's native code into a Java-based Android application. We also provide the vector weights and quantifiers needed to run Lyra, ”Google said. This release provides the necessary tools for developers to encode and decode audio with Lyra, optimized for the 64-bit Android ARM platform, with a version for Linux.
Features are decoded into wave form using a generative model. Generative models are a special type of machine learning model well suited for recreating a complete audio waveform from a limited number of functions.
Lyra's architecture is very similar to traditional audio codecs, which have been the backbone of Internet communication for decades. While these traditional codecs are based on digital signal processing techniques, Lyra resides in the generative model's ability to reconstruct a high-quality speech signal.
Google implemented Lyra in its free video calling app Duo and said it was making the code open source because it thinks it might be suitable for other apps.
Google thinks there are a number of apps that Lyra might be suitable for, whether it's for archiving large amounts of voice, saving battery life, or alleviating network congestion in busy situations.
"We look forward to seeing the creativity that characterizes the open source community applied to Lyra to deliver powerful and unique applications," said Google.
Source: https://opensource.googleblog.com