|
In voice transcription, voice content is converted to text so it can be analyzed, impact decisions or trigger actions, improve productivity and provide information accessibility. Transcription adds value and improves business processes whenever there is a large amount of voice data. With the rapid growth of telecommunication, broadcast and information technologies, improved audio compression technology and reduced cost of large storage systems, a vast amount of voice data is being collected every day. Call centers are a major component of CRM (Customer Relationship Management) operations. Analysis of customer interactions through speech analytics can help to increase agent productivity, monitor customer satisfaction, improve quality assurance, and aid in trends detection. .
The transcription server prototype developed in Haifa is Websphere-based scalable middleware. The prototype uses state-of-the-art speech recognition technology from the Watson Research center to automatically transcribe speech content such as phone calls and lectures. Haifa researchers have also developed a novel speech document retrieval technology that can be used to search the recorded audio. Because automatic transcription does not generate 100% accurate transcripts, searching the transcript using a common search engine may produce inadequate results. This is especially true in cases where the search term was not recognized correctly because there are several acoustically close words. The novel algorithm developed in Haifa takes into account not only the most likely word selected for the output transcript, but several candidates with their respective confidence scores, and thus overcomes this problem. The transcription server and audio search developed can work under the framework of the Unstructured Information Management Architecture (UIMA).
| 
