Hi,

I'm using Julius (In Server Module Mode) for a university project where the spoken command gets recognized by julius and is then sent to a piece of software, which interprets the input and then does something depending on the command given.

To evaluate the system I wanted to measure some times, mostly in the piece of software i developed. But also the following two:

1. Length of each spoken word, which I get from the julius output. (no probs there)

2. The time it took julius to process (or recognize) each word or preferably the entire command.

Does anyone know if and how, if possible, I can get the second time? Haven't found anything in the docs unfortunately.

Thanks in advance,

Lars

Presumably in the dialog manager (the software that communicates with Julius) the socket connection is set up and then there is a function which listens for output from Julius on the socket. Have you tried wrapping this listening function in a timer? You can ask Julius to record the speech input, get the timestamp and length of that input and compare it to the time that the function "hears" something on the socket.

That's actually a nice idea. Unfortunately julius only seems to set off timestamps in seconds (unix timestamp), which is not precise enough.

Example:

<INPUT STATUS="ENDREC" TIME="1452608640"/>

For example:

~/vox/athena> ls -la --time-style=full-iso failwavs/2016.0112.095904.wav
-rw-r--r-- 1 colin users 48044 2016-01-12 09:59:05.658992764 -0500 failwavs/2016.0112.095904.wav

The file creation time is stored in the system to quite fine granularity. The DM ought to have similar granularity to get the heard time. The above is on a Linux openSUSE system. What are you using for OS and DM?

Ubuntu 14.04 and the DM is in C++, written in Qt