TTS Engine Comparison

I thought I might investigate a clearer sounding text-to-speech engine for Carl, but found out that clarity comes with a long wait:

( using phrase: "<engine-name> Hi. My name is Carl. How do you like this voice?" )
$ ./
real 0m0.565s
user 0m0.122s
sys 0m0.050s

real 0m0.283s
user 0m0.064s
sys 0m0.044s

real 0m7.479s
user 0m3.727s
sys 0m0.981s

plib espeak-ng option rate 150 volume 125
real 0m0.386s
user 0m0.101s
sys 0m0.058s

$ ls -1 samples/
flite_ssml.wav <- sample with flite reading a file with SSML (speech synthesis markup language)
plib_espeak_ng.wav <- voice I am currently using for Carl -s150 -a125 option


The very clear Cepstral Charlie voice comes at a cost of a 5 to 7 second delay before speaking, and a US$31.50 price!


Even if you ignore the relatively steep price tag for something like a clearer voice, (which is a design decision / judgement call), the five-plus second delay is entirely unacceptable from a UI perspective.

Congratulations for understanding that an engineering accomplishment should not only work, but be reasonable from a non-engineer user’s point of view.  Many engineering types are totally incapable of making that mental jump in point-of-view.

Total props and a good call.

1 Like

And Carl’s POV…

1 Like