This article is more than 1 year old

Say 'Yes' after the tone

Where are we at with speech recognition?

Speech recognition that actually works, Star-Trek style, is a genuinely tough computing problem. So, when the HomeTalk consortium said it was to start European trials of its vision of voice-activated domestic bliss, we wanted to find out some more.

Hometalk is the latest in a series of high-profile speech recognition announcements: at the SpeechTEK conference in San Francisco last month Opera launched a voice-enabled browser and IBM and Microsoft launched new voice products for big business. IBM updated its Websphere Voice product range and Microsoft announced Speech Server 2004.

The important question, of course, is how long until we can have Picard-esque conversations with our networks and PCs? This might take a while, according to Dr. Phillip Hanna, a natural language processing specialist at Queen's University in Belfast.

"Although speaker-dependent packages (like IBM's Via Voice, which, incidentally uses the same basic technology as the Websphere Voice products- Ed) - where the user trains the software to recognise their voice - work reasonably well in accoustically clean environments, background noise, accents and unclear (i.e. normal) speech can play havoc with the recognition percentages," he said.

Can you hear me at the back?

But this shouldn't matter for most uses. The technology is good enough if you don't want to catch every single word, or if you have a defined set of acceptable responses. So for call centres, for example, speech can replace touch tones as a way for customers to interact with the answering system. For simple voice instructions, Hanna says, recognition systems can get enough understanding to provide an appropriate response by stripping out non-essential words in the input.

According to IBM's Mike Howell, European voice sales manager, demand for speech recognition is driven by an increasingly self-service world: people don't want to queue in banks to check on balances or move money around, and banks would rather we phoned them, too, he says.

There is a pull on the consumer side too: he notes that voice recognition is also making its way onto phones and PDAs, which will prompt demand for more sophisticated applications, and better recognition technology.

HomeTalk, then, has launched into a strange environment of almost-good-enough technology, and some consumer demand.

The HomeTalk project is an open source platform for voice-enabled home automation. Christos Georgopoulos, CEO of inAccess Networks in Greece, argues that the project has to be open source to get developers on board. Closed platform projects, all running in different languages needing specialist developer "[do] not help the progress of the service development market and could not build a critical mass of service developers", he says.

It takes previously unconnected appliances, like a telephone and an oven, and puts them on one platform connected through a Residential Gateway, provided by inAccess Networks. The gateway holds the hardware interfaces and software protocol stacks to get all the various technologies talking nicely to one another.

Doing what comes naturally

The system's user interface is speech enabled, using IBM's ViaVoice embedded platform. It provides voice recognition and text-to-speech functions that the project organisers say will provide a more natural way of interacting with the network. Users can either use a voice-activated PDA or a normal phone to access the system remotely.

Speech is touted as the most natural way for us to interact with technology, as this is how we primarily interact with each other. But does that mean it is the way we want to interact with our toasters? (See Red Dwarf's intelligent toaster for more debate of this topic.)

The push into the enterprise space makes sense. There are fast returns on investment in speech technology for call-centres: speeding up call handling undoubtedly saves money, even if it irritates the customers.

The HomeTalk project uses relatively simple voice recognition technology: word matching and trained voice systems, but we suspect Howell is right when he suggests people will start asking for more sophisticated recognition as the technology spreads.

The Hometalk platform is now finalised and all the technical information is available online for developers on the project's website. ®

Related stories

Boffins test voice-activated secure credit card
Hardware will be (almost) free Chairman Bill
Opera browser to recognise speech
NASA pulls off mindreading act

More about

TIP US OFF

Send us news


Other stories you might like