DQ AutoVoice

VoxDev provides the world’s most advanced voice automated database query and telephone directory enquiry technology. DQ AutoVoice enables speech recognition to achieve its own potential.

Our technology enables natural language use of voice automated interfaces and resolves issues related to the accuracy and scalability of voice recognition solutions resulting in a ‘break-through’ in highly scalable carrier grade ‘Voice Recognition’ software for fixed line and mobile telephones.

DQ AutoVoice is a language independent software product that can dramatically increase speech recognition accuracy when used with a large database. Specifically developed for automating DQ 118 and name/address capture services, it achieves accuracy of over 90% on residential directory enquiries using a real database of 20 million people (over 48 million using the electoral roll). Working in conjunction with leading speech recognition systems (from Nuance for example) DQ AutoVoice interfaces with standard databases (SQL Server/Oracle) to find the best matching records using sophisticated cross-matching database search techniques.

In development for over the 10 years, DQ AutoVoice was recently independently tested by a large speech software vendor using real call data and databases where it achieved 92% accuracy for UK English, 96% for Danish and 94% for Swedish, compared to around 60% using the speech recogniser alone. DQ AutoVoice is the best software to choose for (fully or semi-) automating directory enquiries or name/address capture. DQ AutoVoice is unique in being able to handle very large databases of records.

The DQ AutoVoice search engine technology is radically different from traditional approaches to speech recognition and is designed specifically for voice searching extremely large directories / databases. It is the best speech technology available for automating directory enquiries 118XXX applications. Suitable for applications such as speech-driven auto attendants, name and address capture, bet taking, payment processing, call routing, password reset, toll fraud protection and directory information centers.

Unlike our main competitors our technology is able to handle high volumes of database query requests initiated even from ‘low quality’ GSM phones with extreme accuracy enabling large reductions in staff overheads. Existing speech systems will ask the user more questions and often need the names or street names to be spelt out as confirmation this is unnecessary with DQ AutoVoice.

PRINCIPLE OF THE SYSTEM

For many years, speech recognition developers have been obsessed by achieving a 0.01% increase in recognition accuracy. This often, though, misses the point of the problem that needs solving. ‘Management of Uncertainty’ is a technique developed over the past years by VoxDev. Management of Uncertainty improves task accuracy in scenarios where spoken input can be broken down into related parts For example, a typical residential Directory Enquiry has town, surname, forename and street, the combination of these pieces of information uniquely identifies the record that the caller is seeking. Although any one of the individual utterances may be poorly recognised, the other recognised utterances may be combined to find a smaller candidate record set that is much smaller (thousands of records) when compared to the original data set (millions of records). Sophisticated cross matching of the database yields the most likely records in a short period of time.

During a dialogue, the main technique is to defer decision making as late as possible to avoid unnecessary asking of additional questions – confirming each answer during a call can be frustrating for the caller and may introduce additional ambiguity if the line is noisy.

UNIQUENESS OF THE SYSTEM

DQ AutoVoice is unique in being able to handle very large databases of records. Whilst other vendors may claim automation on smaller databases in other countries, the techniques that they employ simply do not scale to very large real databases. VoxDev is unique in offering a solution that can be applied to the UK market place. The 48 million record UK electoral roll, for example, contains over 375,000 distinct first names, 700,000 distinct surnames and 320,000 distinct street names.

DQ AutoVoice works in conjunction with the world’s leading speech recognisers, it does not contain its own proprietary engine – we prefer to take advantage of any technical improvements offered by much larger speech vendors. We don’t use brute force speech recognition to find the most likely answer. When developing DQ AutoVoice we took a step back to analyse the problem of identifying a single record in very large databases. Our solution is highly accurate and very efficient. It is language independent and works for any application where the goal is to identify a record in a database using speech recognition. The larger the database and the more utterances (corresponding to fields in the database e.g. surname, street name) that are available, then the more that DQ AutoVoice stands out over any opposition.

CASE STUDIES

DQ AutoVoice is currently being used live across several sites in the UK, the largest being the personal directory service “Speech Directory” provided by Speech Solutions Ltd. The ease of use of DQ AutoVoice has been demonstrated by the fast and simple integration with a variety of other well known IVR products. The Speech Directory service has been implemented on the InVade and Envox IVR platforms. Speech Solutions investigated a number of solutions and finally settled on a solution developed by VoxDev based on DQ AutoVoice. Speech Solutions first undertook their own testing of VoxDev’s sample 118 and Name/Address capture applications and found them to be far superior to anything else in the market place. Their target market for the Speech Directory was large corporates as well as SMEs. The directory sizes ranged from a few thousand names to a less than ten. The solution provided by VoxDev performed highly accurate name recognition in noise (line noise and verbal noise) against a SQL server database containing names of employees.

VALUE TO THE 118 SECTOR

DQ AutoVoice can be used to fully or partially automate a 118 directory enquiries service. A fully automated solution may of course not be acceptable to all callers, but a semi-automated service could be used in conjunction with a human call centre agent. The automated part could be used to obtain the information from the caller and make an initial search in the database whilst leaving the human agent (who could also listen to the callers utterances just before speaking to them) to handle the final identification of the desired number. A fully automated solution could of course be used to handle any overflow calls where insufficient call centre agents are available.

Using standard speech recognition techniques to 118 automation, 60% accuracy for the service would not acceptable to callers, but, by making use of DQ AutoVoice, one approaching 90% accuracy would much more acceptable. Given problems of staff retention in call centres and long queuing times often experienced by callers, automation could be the key to solving some of the recent problems in the de-regulation of 118 services. Providing even a semi-automated solution backed up by human intervention at the last stage could dramatically reduce waiting times and improve customer satisfaction.

DQ AutoVoice clearly provides an opportunity for some 118 service providers to provide a highly accurate, reliable, efficient yet low-cost 24×7 automated service where appropriate. This is only made available in the UK by using VoxDev software.

BENEFITS OF USING DQ AUTOVOICE

DQ AutoVoice can used to provide an entirely automated 118 service or used to complement a human agent service. From our many visits to call centres, it is clear that a fully automated service would not be appropriate for many callers who are uncertain of many of the pieces of information required to identify a number (name, address etc), but for a certain caller, perhaps a regular business user who has all of the information to hand and wants a low-cost, fast and efficient service, with minimal waiting time, a fully automated 118 solution based on DQ AutoVoice would be suitable.

A partially automated solution, where DQ AutoVoice is used in conjunction with a human agent, would certainly help reduce call waiting times. The automated part could be used to obtain information from the caller, undertake speech recognition and database search, and present the most likely database records to the human agent. The advantage to the caller is that there would be no waiting and the emotion level of the prompts would be consistently high (automated systems don’t get tired of asking the same questions 24×7). DQ AutoVoice is able to feedback confidences in its own success to the human agent, and in cases where this is low, the human agent could perform their own database search based on the caller’s utterances.

To summarise, DQ AutoVoice offers the following benefits:

• High accuracy at 118-like tasks
• Works with very large databases of information
• Full or partial automation of 118 services, including overflow handling
• Consist and reliable performance over long periods of time
• Integrates with industry leading speech recognition systems and database engines