Cadence Tensilica HiFi 5 DSP for audio and voice processing
The new generation of HiFi DSP has double the audio processing power and 4X the neural network (NN) processing power as its predecessor, the HiFi 4 DSP. The extra processing ability makes it ideal for voice-controlled user interfaces for digital assistants.
The popularity of digital home assistants has led more manufacturers incorporating the technology in newer areas, such as automotive infotainment. These manufacturers also want a greater number of features, for areas where the Cloud is hard to access, or for simple tasks that should not need a cloud connection or always-on internet to perform. Having additional neural networking capabilities combined with advanced noise reduction algorithms extends the number of commands a digital assistant can understand, which can then be interpreted and performed without connecting to the cloud. This type of solution is more efficient and enhances privacy.
Key features of the HiFi 5 DSP include:
– Five very long instruction word (VLIW)-slot architecture capable of issuing two
128-bit loads per cycle
– 2X MAC capability versus HiFi 4 DSP for pre- and post-processing, includes:
• Support for eight 32×32-bit or 16 16×16-bit MACs per cycle
• Optional eight single-precision floating-point MACs per cycle
– 4X MAC capability versus HiFi 4 DSP for NN processing, includes:
• 32 16×8 or 16×4 MACs per cycle
• Optional 16 half-precision floating-point MACs per cycle
– The HiFi NN library offers a highly optimised set of library functions commonly used in NN processing. These library functions can easily be integrated into popular machine learning frameworks.
– Software compatibility with the complete HiFi product line totalling over 300 HiFi-optimised audio and voice codecs and audio enhancement software packages.