VoxForge
VoxForge
Hi
for a project in an exhibiton i have to develop an application wich is able to recognize one single word (two sylab, a name). This word will be spoken be visitors: many different voices, different languages, different pronunciation.
I have 20 recordings from different peaple.
I want to do it with sphinx (and maybe additionally blather).
What are your recommendation for this task:
- used dictonary
- training
- how to avaoid false positives (is negativ trainig possible?)
thanks in advance and greetings
Christian Schweingruber
Hello
For the competent help on CMUSphinx it's better to use CMUSphinx forums.
> for a project in an exhibiton i have to develop an application wich is able to recognize one single word (two sylab, a name). This word will be spoken be visitors: many different voices, different languages, different pronunciation.
> I have 20 recordings from different peaple.
> I want to do it with sphinx (and maybe additionally blather).
> What are your recommendation for this task: used dictonary
Our most accurate model Generic en-us should be good.
> training
Not needed
> how to avaoid false positives (is negativ trainig possible?)
You need to tune keyword spotting threshold on your specific keyphrase (-kws_threshold value)