Voiceprint Extraction GPU processing

Since version 1.2.0, the Phonexia Voiceprint Extraction microservice is also available in a variant with GPU acceleration support. This can be enabled by setting the device option to cuda (default is cpu). This also requires a special GPU-enabled Docker image. See the microservice readme for more information.

While primarily using GPU acceleration, certain processing tasks will still rely on CPU resources. It is possible to share a GPU for multiple microservice instances for better utilization. GPU acceleration can significantly improve extraction performance. In our measurements, we observed an almost 5x speedup. However, the overall performance also depends on other parts of the server (e.g. ethernet connection).