I'm working on a project that would like to modify sphinx to use floats rather than doubles (we're simulating running sphinx on hardware where simulating double precision arithmetic is really slow). does anyone know if sphinx truly requires double-precision arithmetic?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I'm working on a project that would like to modify sphinx to use floats rather than doubles (we're simulating running sphinx on hardware where simulating double precision arithmetic is really slow). does anyone know if sphinx truly requires double-precision arithmetic?
1, What kind of hardware?
2, float64 is necessary on parameter such as beam width and scores.
-Arthur