SpinVox Shares Details of its Speech Technology Breakthroughs
SpinVox took the opportunity to reveal some of the details about the technological breakthroughs it has achieved in the development of its Voice Message Conversion System (VMCS).
The core technology around which the VMCS is built is unique to the speech market, so there is little to which it can be compared. It is based on world-leading breakthoughs in automatic speech recognition (ASR) combined with artificial intelligence, semantics and natural linguistics which have been developed by SpinVox Cambridge-based Advanced Speech Group (ASG).
VMCS already knows more than 99 per cent of anything a user is likely to say. It contains over two billion words and phrases derived from the equivalent of 72 years of audio training – making it the world's largest corpus of spoken language. This knowledge system is growing at an ever-increasing rate and is further accelerating SpinVox ability to automate.
Dr Tony Robinson, a peer of Prof. Philip Woodland of the Cambridge University Machine Intelligence Laboratory, leads the SpinVox ASG of more than 20 speech specialists and PhDs who are working continually to develop and refine the system.
Professor Woodland explains: "Much of this technology derives from the world-leading research undertaken by my group at Cambridge University Machine Intelligence Laboratory. I am a consultant to SpinVox and its Automatic Speech Group and their unique approach allows the automatic system to be exposed to huge amounts of spoken data, from which highly accurate acoustic and language models can be built."
This combined group of technologies and processes is the main engine for converting voice messages to text, rather than human intervention.
Having experimented with purely automatic speech conversion, SpinVox decided early on in its development that because its voice to text service converts real-life, dynamic and fast-evolving language and messages that we use and exchange every day (known in the industry as 'free form speech'), it was essential that the system had the capability to evolve at the same rate, converting the latest words, phrases, brand names and colloquialisms to ensure a high level of accuracy. This is why it describes the system as 'live-learning'.
SpinVox realised that only by combining its rapidly evolving state-of-the art technology with human quality control and training, could it create a system which could complete elements of messages which could not be automatically converted. As a result, the system constantly learns new words and phrases, making it increasingly efficient and reliable.
Explains SpinVox CIO Rob Wheatley: "Quality Control agents are an important part of the SpinVox service because their constant minute-by-minute input actually improves the quality of text conversions in a process we call 'live learning'. The technology is a bit like a human brain, in that, the more it is exposed to input, the more it learns.
"This process has helped us improve our accuracy massively. Since its inception in 2007, the technology has improved to the extent that the system requires only two per cent of the input it required just two years ago and can even now predict more than 99 per cent of what most people speaking in English or Spanish will say next. Or to put it another way, in just two years, we have reduced the requirement for human intervention to just a few hundred agents per market compared to the thousands per market when we started. Our world-class speech scientists in the Advanced Speech Group have helped make this system unchallenged in terms of accuracy, speed and reliability."
As discussed above, SpinVox VMCS learns from human intervention. This is the reason SpinVox works with five, world-class, call centres which have been chosen after SpinVox put around 50 call centres through its stringent quality control and security procedures.
Every message is dealt with initially by the automated system. Only in cases where speech is too indistinct to be dealt with by the system, or contains unfamiliar or new words or phrases, is the completely anonymised and encrypted message sent to a QC agent for help. The agents will only ever see the messages that need input and do not know how many other messages have been converted, processed and sent automatically by VMCS.
SpinVox is fully compliant with industry standards relating to the processing of information, including the Data Protection Act 1998. To this end, any part of a message seen by the Quality Control centres is anonymous, encrypted and randomised, meaning that it is impossible to determine where the messages are from or where they are going to. SpinVox has achieved two prestigious ISO qualifications: ISO 27001 (the international Information Security Standard) and ISO 9001:2008 quality certification.
Jaime Tronqued is President of ScopeWorks Asia, Inc., a Quality Control house which handles private and secure customer services for companies in telecommunications, travel, banking, healthcare and insurance.
Scopeworks has been working with SpinVox for the past two years and Tronqued says: "I can categorically assure people that SpinVox messages are both private and secure. There are many layers of security and privacy that are used to ensure this and SpinVox was extremely thorough in its audit of our operations, our security and our privacy procedures as they ran a training pilot with our Quality Control agents on test conversion messages, prior to contracting with us to deliver a live customer service using encrypted and anonymised messages. "
Adds Ragindra Persaud, CEO of NPIC, a South American-based Quality Control house, which has also been working with SpinVox for the past two years. "SpinVox went through a thorough review of our processes and procedures, and it was not until they were fully satisfied that we comply with their stringent requirements, that they contracted with us officially. In the live customer system there is no way for any Quality Control agent to know where a message has come from or to whom it is being sent, nor copy or abuse this."
write your comments about the article :: © 2009 Computing News :: home page