News & Analysis
Comment
WireMan
Spansion already sells memory, so it's no surprise acoustic data gets saved in a ...
VUI Guy
DrLAL is correct. The bottom line is small footprint, low resource drain and ...
Spansion announces voice recognition chip
Rick Merritt
6/19/2012 3:50 PM EDT
SAN JOSE – NOR flash maker Spansion Inc. officially launched a new business unit today that aims to roll out a family of human interface accelerators starting with a voice recognition chip. The company teamed with Nuance Communications to create its Spansion Acoustic Coprocessor.
The first two members of the new family are targeted at car infotainment systems. The two companies will work together on future chips for industrial and consumer markets, said Alvin Wong, vice president of marketing for Spansion’s new programmable systems group.
“We really believe human-machine interfaces are the future, so we are also looking into gesture and image recognition products,” said Wong, calling the voice recognition accelerator an industry first.
Voice recognition has been on the computing horizon for decades, making slow in-roads into a variety of markets. Most recently, Apple’s Siri interface for the iPhone 4S has captured the public imagination about the potential for the technology.
Spansion promises its accelerators will cut in half both system response time and the CPU workload for voice recognition. The chip essentially stores acoustic databases and performs parallel searches across them.
The company will initially offer two versions of the chip, one supporting acoustic models for up to a dozen languages and the other supporting three languages. They use parallel search logic developed in house by Spansion along with algorithms and acoustic databases supplied by Nuance.
The chips link to applications processors over an SPI bus. They include an internal 1.2 Gbytes/second data bus to support fast memory searches.
The company is not releasing details of how much memory the chips contain or how much they will cost.
The chips consume from 100 milliwatts to 1.5W, depending on their size and fit in a 10mm2 QFN package. They will sample in the third quarter and be in production in the first quarter of next year.
Automotive is a natural target for Spansion. Car makers are under pressure to deliver more information to drivers with greater safety, making voice interfaces a natural direction. Spansion claims it supplies as much as 70 percent of the NOR flash used in cars.
Given the relatively long design cycles at car makers, Spansion is not expecting revenues material to its results in its first year. However, it notes market watchers forecast a potential automotive market of as many as 35 million units growing to as much as 90 million units over the next several years.
The first two members of the new family are targeted at car infotainment systems. The two companies will work together on future chips for industrial and consumer markets, said Alvin Wong, vice president of marketing for Spansion’s new programmable systems group.
“We really believe human-machine interfaces are the future, so we are also looking into gesture and image recognition products,” said Wong, calling the voice recognition accelerator an industry first.
Voice recognition has been on the computing horizon for decades, making slow in-roads into a variety of markets. Most recently, Apple’s Siri interface for the iPhone 4S has captured the public imagination about the potential for the technology.
Spansion promises its accelerators will cut in half both system response time and the CPU workload for voice recognition. The chip essentially stores acoustic databases and performs parallel searches across them.
The company will initially offer two versions of the chip, one supporting acoustic models for up to a dozen languages and the other supporting three languages. They use parallel search logic developed in house by Spansion along with algorithms and acoustic databases supplied by Nuance. The chips link to applications processors over an SPI bus. They include an internal 1.2 Gbytes/second data bus to support fast memory searches.
The company is not releasing details of how much memory the chips contain or how much they will cost.
The chips consume from 100 milliwatts to 1.5W, depending on their size and fit in a 10mm2 QFN package. They will sample in the third quarter and be in production in the first quarter of next year.
Automotive is a natural target for Spansion. Car makers are under pressure to deliver more information to drivers with greater safety, making voice interfaces a natural direction. Spansion claims it supplies as much as 70 percent of the NOR flash used in cars.
Given the relatively long design cycles at car makers, Spansion is not expecting revenues material to its results in its first year. However, it notes market watchers forecast a potential automotive market of as many as 35 million units growing to as much as 90 million units over the next several years.
|
|
|
|
|
Navigate to related information
chanj
6/19/2012 4:21 PM EDT
With a lot of car makers coming to the Valley to do R&D of infotainment, voice recognition seems like a good move. Voice recognition seems to be the MMI in the future. Samsung SmartTV supports it. The technology may likely apply everywhere.
Sign in to Reply
DrLAL
6/19/2012 5:15 PM EDT
"Spansion promises its accelerators will cut in half both system response time and the CPU workload for voice recognition. The chip essentially stores acoustic databases and performs parallel searches across them". Won't it run out of storage space?
Sign in to Reply
daleste
6/19/2012 9:38 PM EDT
Smart move to diversify. The memory business has been shown to not be a stable market.
Sign in to Reply
nicolas.mokhoff
6/20/2012 2:20 PM EDT
I tend to agree with DrLAL. You need a lot of acoustic databases to be accommodated on the chip for the parallel searches to come up with meaningful results. If they stick to the automotive speech recognition domain they can probably get away with a limited vocabulary but they will need pretty fast search capabilities for real-time response in a vehicle setting. Wonder if the Apple folks have not thought to embed their Siri in cars' infotainment system. Seems like a natural.
Sign in to Reply
WireMan
9/25/2012 6:31 PM EDT
Spansion already sells memory, so it's no surprise acoustic data gets saved in a LOT of memory.
Sign in to Reply
VUI Guy
6/21/2012 4:37 AM EDT
DrLAL is correct. The bottom line is small footprint, low resource drain and cost effective for the chip manufacturer. That's why companies prefer a vendor like Rubidium as opposed to one that demands more resources like Nuance.
Sign in to Reply