Cadence’s Tensilica HiFi iQ DSP Features Voice AI and Audio Processing
The new DSP IP core improves AI and audio performance with FP8/BF16 support, enhanced SIMD processing, and lower power consumption for voice-enabled embedded systems.
HiFi iQ DSP
Cadence has introduced the Tensilica HiFi iQ DSP, a new digital signal processor (DSP) IP core designed for advanced voice AI and immersive audio applications. The processor itself belongs to the recent generation of Tensilica HiFi DSPs offered by the company and is used in smartphones, smart speakers, and automotive infotainment systems, as well as other edge AI devices, where high audio processing power and power efficiency are needed.
Architecture and Performance
The HiFi iQ DSP is built on an updated architecture optimized for both audio signal processing and AI workloads. According to Cadence, the processor delivers up to 2x higher compute performance compared to the previous HiFi 5s DSP. It also improves AI processing capability by up to 8x, enabling faster execution of modern machine learning models used in voice recognition and audio enhancement.
Block Diagram of Tensilica HiFi iQ DSP
The other essential improvement is the efficiency of power. The new DSP is able to save over 25% of energy on most tasks, and this makes it ideal in battery-operated devices that need voice processing all the time.
AI and Data Format Support
Given the needs of modern AI workloads, the HiFi iQ DSP has long-data-type support of formats (e.g. FP8 and BF16) frequently used by machine learning models. These extensions enable developers to execute neural networks and voice AI algorithms on the DSP with better efficiency.
The processor also has a larger SIMD vector processing and better auto-vectorization, which assists in faster processing of complex signals and AI processes. This simplifies software optimization by the developers who do not require much manual tuning.
Tensilica processor development flow
Audio Processing Capabilities
The DSP is developed to have capabilities to perform a wide variety of audio processing, such as keyword location, beamforming, noise cancelling and automatic speech recognition. It also implements high-end multi-channel audio rendering, which is applied in modern entertainment systems.
Moreover, the processor has a number of advanced audio codecs such as Dolby MS12, Opus HD, Audio Vivid and Eclipsa Audio, which can provide immersive audio experiences in consumer and automotive devices.
Voice Processing Pipeline On-Device
Ecosystems Software and Development
Cadence offers a full development environment for the HiFi iQ DSP. The processor is compatible with the Cadence NeuroWeave SDK and features a range of frameworks, including TensorFlow Lite for Microcontrollers, enabling developers to run AI models effectively on embedded systems.
According to the press release, the DSP can also work alongside dedicated neural processing units (NPUs) in a system-on-chip to further improve AI performance when required.
Tensilica HiFi DSP audio and voice applications
Video
Applications
The Tensilica HiFi iQ DSP is created to be used in a very broad applicability, such as:
- Handheld phones and devices.
- Voice assistants and smart speakers.
- Car information systems.
- Home entertainment products and smart televisions.
- Edge AI audio signal processing.
The Cadence Tensilica HiFi iQ DSP is a more powerful DSP that could meet the increasing requirement of voice-enabled and AI-driven audio applications in next-generation embedded systems based on its superior compute capability, better AI acceleration capability, and reduced power consumption.
Images used courtesy of Cadence
