Cambridge start-up AudioTelligence has developed one of the first digital solutions for voice-activated devices to comprehend speech from loud background noise.
The company has raised £3.1million from Cambridge Innovation Capital and Cambridge Enterprise to grow its team to meet the considerable demand for its technology, which, in addition to helping humans to hear and understand speakers in the presence of loud background noise, helps to improve the performance of voice recognition systems.
AudioTelligence CEO Ken Roberts (pictured with Andrew Williamson, Investment Director, CIC) explains that one of the major motivations in developing the technology was to overcome the “cocktail party problem”. This is a particular area of interest to Professor Peter Rayner who founded the Signal Processing and Communications Laboratory at the University of Cambridge.
Roberts says: “Peter’s expertise was called upon in 1985 when the National Sound Archive first considered digitising its collection to preserve its decaying recordings. His research led to the establishment of CEDAR Audio Ltd in Cambridge in 1988, to develop the new science of audio restoration and to commercialise it worldwide.”
CEDAR has grown substantially since those early days, providing high value products for markets such as film and TV production and broadcasting, as well as law enforcement and security, where high quality sound and improved intelligibility are urgently required.
Roberts continues: “Today, Peter uses a hearing aid and is therefore acutely aware of the difficulties faced by the hearing impaired when different sources of sound are mixed together. It was his vision to solve this problem by using advanced digital signal processing techniques that led CEDAR to conduct eight years of research in this area, and consequently to the founding of AudioTelligence.”
Now a new application for signal processing technology has emerged. The Cambridge cluster is a world-leading hub for voice command and control technologies, and in recent years Amazon, Google and Apple have all set up R&D centres in the city to integrate such technologies into their products and services.
The CEDAR Audio team, headed by Gordon Reid, Dave Betts, Dr Christopher Hicks and Clive Osborn, together with Professor Rayner and Professor Simon Godsill, saw that their recent work had huge potential in this new market but recognised that a different business model would be required to achieve commercial success. So it was agreed to spin out a new company, AudioTelligence, to license the technology for consumer products.
The team brought in Dr Robert Swann as a non-executive director and Ken Roberts as CEO to help to define a business plan, working alongside the patent holders Dave Betts and Mohammad Dmour. Swann is known as co-founder of AlphaMosaic (which was subsequently sold to Broadcom) and Roberts has a strong track record in business development and commerce, including roles as CEO of Pulsic Ltd and board/advisor positions at Akya, Undo and Ateeda.
To secure contracts with industry leaders, a sizeable funding round was required, so the company looked to Cambridge Innovation Capital (CIC) for support. CIC was established to help companies with significant IP and strong Cambridge connections to grow within the cluster.
Andrew Williamson, Investment Director for CIC, sees a considerable opportunity for AudioTelligence and says that Cambridge is the best place for it to build the strong team required.
He says: “The cluster has specialist expertise in voice recognition and signal processing based on decades of research and engineering. The company is very well connected with the University and already has considerable technical and commercial expertise. It is well placed to capitalise on the opportunity to provide a digital solution for voice driven applications.”
Cambridge University’s commercialisation arm, Cambridge Enterprise, has also been involved with AudioTelligence since the outset because of its long association with CEDAR.
Elaine Loukes, Investment Manager at Cambridge Enterprise commented, “AudioTelligence is a spin-out from CEDAR Audio, which is the oldest company in our portfolio. It’s a real pleasure to be working with a new company that we know to have such an excellent pedigree.”
AudioTelligence is unique because, instead of using high-quality microphones to help distinguish the target voice from background noise, it uses a software approach called blind audio signal separation. This uses Bayesian statistics and machine learning to split the soundscape into separate streams, each representing one source of sound.
Furthermore, the system is adaptive so, if new sounds such as air-conditioning, other fans or equipment noise appear, these are also eliminated from the sound heard by the listener. The resulting sound is high quality with low latency and no artefacts or echo.
When reading a list of 100 items to an Amazon Echo Dot using existing technology, the device will recognise 84 items under good conditions, dropping to 22 when a second sound source that creates interference is introduced. Use of AudioTelligence software to pre-process the sound enables the system to recognise 94 items even with significant background noise.
The AudioTelligence software will be available on a licence basis and offers a low cost, high quality solution for voice activated devices such as in-car navigation and entertainment systems, personal assistants, hearing aid devices and amplifiers.
Chief Scientific Officer Dave Betts explains: “CEDAR is recognised as an industry leader in signal processing and its products are used by major film companies and broadcasters, mastering studios, and live sound companies worldwide. It is also a standard in many parts of the world for cleaning recordings and live audio streams for audio forensic and security purposes.
“AudioTelligence builds upon this expertise to provide a low-cost option for integration into consumer devices, but this is a very different market, which is why we decided to spin out the company.
“It is an exciting time for both companies and for the Cambridge cluster as we are pioneers in this technology.”
Both Mr Betts and Dr Hicks were awarded Academy Awards in February 2005 for developing the noise-reduction systems used in the production of most major movies coming out of Hollywood and elsewhere.