Real-time digital signal processing on the xCORE-200 architecture
Voice interfaces are increasingly popular and ubiquitous in today’s life environment, ranging from well-known teleconferencing systems to upcoming connected natural speech enabled personal assistants. Such new applications are based on a networked topology where processing is distributed on multiple nodes.
Interface nodes for data acquisition or rendering are increasingly required to provide real-time DSP processing of input/output data signals for subsequent handling by other nodes. In the case of a natural speech interface, multiple microphones provide input signals that need to be pre-processed, for instance by beam-forming or echo-cancellation, with minimal latency. Additional processing such as voice activity detection or key words recognition may also be executed locally at interface level.
This whitepaper analyses the real-time DSP capabilities of the xCORE-200 platform in the context of voice interfaces. A voice capture front end application based on eight Pulse Density Modulation microphones is taken as an example. Comparisons with a standard DSP architecture are also discussed.
Read and download the full whitepaper, below.