Google has released an update for the Lyra audio codec 0.0.2, optimized for achieving maximum speech transmission quality over very slow communication channels. The codec was unveiled in early April but was initially tied to a proprietary mathematical library. In version 0.0.2, this limitation has been resolved, and an open replacement has been created for the specified library — sparse_matmul, which, like the codec itself, is distributed under the Apache 2.0 license. Other improvements include the ability to use the Bazel build system with the GCC compiler, which is now the default in Linux instead of Bazel+Clang.
It is worth noting that in terms of voice data transmission quality at low speeds, Lyra significantly outperforms traditional codecs that use digital signal processing methods. To achieve high-quality voice transmission in environments with limited data volume, in addition to standard audio compression and signal transformation methods, Lyra utilizes a speech model based on a machine learning system, which reconstructs missing information based on typical speech characteristics. The model used for sound generation has been trained with thousands of hours of recordings of voices in more than 70 languages. The performance of the proposed implementation is sufficient for real-time speech encoding and decoding on mid-range smartphones, with a signal transmission delay of around 90 milliseconds.
Source: opennet.ru
