Nuance batchrec analysis tool

The nuance-batchrec script is a tool is for Nuance developers who wish to analyze the results of Nuance's 'batchrec' program. The nuance-batchrec script reads in the results of a Nuance batch recognition run, calculates Word Error Rate (WER) using sclite, and sticks the results in a database for subsequent analysis. Once the results of several batch runs have been entered in the database, you can write SQL queries to compare them. For example, you might want to find out which Nuance grammar or parameter settings resulted in the lowest overall WER, or produced the most accurate NLU slots, or ran the fastest.

The nuance-batchrec documentation explains how to create the batchrec analysis database, but not how to perform the kind of analysis you need. We assume you are comfortable with SQL and can design the appropriate queries yourself.

Requirements

The following software must be installed on your system:

A Unix-like enviroment (e.g., Linux, Solaris, cygwin, etc.)
Ruby
Ruby/DBI, the database-independent interface for accessing databases. Be sure your Ruby/DBI installation includes the support libraries for your particular database (MySQL, PostgreSQL, etc.). For example, for MySQL in Debian Linux, you would need the packages libdbi-ruby and libdbd-mysql-ruby.
The SCLITE scoring package from NIST. In particular, the command 'sclite' must be in your execution path.
A running database. Only MySQL has been tested, but any DBI-compatible database that supports the :insert_id function should work.

Download nuance-batchrec