Technology

Nexiwave is a software developer dedicated to provide solution to the information retrieval needs in media archives.

Our product is an integrated search engine for audio content. We took a different approach from traditional speech-to-text engines. We recognize the real problem to solve that speech technology needs to solve is: given a large audio archive, how an end user can effectively find the information that he/she is looking for. nexiwave Speech Indexing system is designed exactly for this task: our job is not done until the end user found the information. Our system is built around the whole life cycle of information retrieval process: capturing, processing, searching and accessing.

The four steps form a positive causal effect loop within our system.

Capturing

This is a key component in the Nexiwave speech indexing system. nexiwave is capable of extracting audio from nearly any kind of audio/video formats. During this step, audio is extracted, cleansed, converted to the nearest sample rate that we support. Also during this step, speakers are identified from audio based on speaker characteristics and past data from this particular audio source.

Processing

This steps is a fuse of two key concepts: speech processing and search targeted indexing. Nearly all data captured during the capturing steps is fed into our speech processing engine and the core indexing engine. A very tight integration between speech engine and indexing engine means data points at nearly every aspect of speech processing were indexed by the indexing engine. At the end of this step, a highly advanced speech indexing database were created.

Nexiwave is the leader in revolutionizing the speech indexing process. Traditionally, speech processing requires 10x or even 20x times real time. Using advanced highly parallel hardware, nexiwave is capable of processing at 0.5x, 0.2x or even 0.1x speed.

Searching

Upon receiving user's search request (when did we talk about "unified communication"?), data points from the speech indexing database is re-evaluated and searched based on this user request. User's search request, as well as past searches, is key driver in this search process. At the end of this step, the exact audio snippets were identified and ranked.

Accessing

At this step, the end user is presented with the ordered search result. End user's further action within this step is stored and fed back to the processing and searching steps. This feedback provides strong positive causal effects in our system.