To use this software, read the wiki.
This project is a prototype funded by the Knight Foundation.
The goal of the project is to determine whether it is possible to provide full-text search of audio files, using open source software to automate the creation of text transcriptions of APM digital audio. We are not aiming for 100% accurate, publish-able transcriptions. We are aiming for good-enough keyword recognition to improve the find-ability of our audio.
APM partnered with Cantab Research Limited to create the ASR (automatic speech recognition) models required, and then built a web service application to process and manage automatic transcriptions.
On CentOS 5 systems, the following RPMs are required to compile Kaldi:
atlas atlas-devel blas blas-devel lapack lapack-devel
In addition, the SoX program is required. The TQ software has been tested with SoX v14.4.1.