Dolphin Design unveils IP for sound classification cutting down energy by 99%

1 min read

Dolphin Design, a specialist in semiconductor IP solutions for power management, audio, and processing with ASIC design services, has launched WhisperExtractor, a mixed-signal IP for voice and audio-enabled SoCs.

WhisperExtractor IP enables voice and sound classification at a µW level power budget, enabling always-on voice user interface and sound detection. This IP is intended for battery-powered applications aiming at Keyword Spotting (KWS), Automatic Speech Recognition (ASR), Natural Language Processing (NLP), or power-efficient sound classification found in intelligent security camera.

WhisperExtractor relies on a disruptive technology that addresses one of the major challenges of battery-powered devices that want to enable voice-user interaction or sound classification, which is power efficiency. It has become critical to have efficient and low-power SoC design to enhance the end-user experience.

Conventional signal processing methods to process Mel Frequency Spectrum Coefficients (MFCC) require a significant amount of energy to convert analogue signals to digital and perform signal processing calculations and advanced software capabilities. This makes it difficult to implement always-on voice listening or sound classification without draining too much current from the battery.

WhisperExtractor is based on mixed-signal architecture to efficiently extract the MFCC needed for audio classification without requiring any software programming. It connects directly to an analogue microphone and outputs the data needed by an AI accelerator block to handle inference/classification.

The advanced architecture of this IP helps reduce power consumption by up to 99% at system level, enabling an always-on listening function at a power budget of just µW, with a simple 32 kHz clock.

It frees massive headroom on DSP or CPUs and wakes up the system only when necessary.

Key Features and benefits of WhisperExtractor:

Power Efficiency: µW level power consumption, extending application live time of battery-operated devices.

Always on domain friendly: 32 kHz clock input, RC or crystal oscillator.

Seamless software integration: MFCCs without any software programming.

Versatile: Built-in support of various use cases, voice classification, sound classification…

Commenting Hakim Jaafar VP Marketing at Dolphin Design, said, “WishperExtractor furthers our investment in our comprehensive set of IPs that enable the cost-effective design of power efficient SoCs. Be it our audio converters, power management, or processing IPs.“

The Dolphin Design IP catalogue encompasses more than 300 references of audio converters, power management, and processing IPs that are commodity features in SoCs.