Making automatic directory assistance more reliable
The challenges of automating the response to business-related requests include recognising names while dealing with large variations in question formulation. The SMADA system tackles some of these challenges by detecting user queries that were not foreseen by automatic directory assistance system designers. These variant question formulations can then be added to the system's vocabulary along with variant pronunciations of proper names to improve its accuracy and reliability. The project analysed directory assistance queries that had been routed to the telephone operator due to the automated system's failure to recognise the requested information. By grouping phonetically similar requests, and based on a hypothesis that such groups would correspond to particular user formulations, the SMADA approach can update the vocabulary of the automated directory assistance system. The project also addressed the challenges of background noise for automatic query recognition. A number of research studies compared the effectiveness of 'de-noising' techniques such as 'soft thresholding' and 'spectral subtraction'. By investigating both signal filtering and noise subtraction approaches the project improved the automatic recognition of directory inquiries in noisy environments.
