Audio and Prompts Discussions

Nested
Dynamic prompt creation
User: scadiforever
Date: 7/24/2010 6:47 am
Views: 5195
Rating: 12

Hi everybody.

I'm here simply to ask you to point me at some resources regarding the generation of prompts as mixtures of phonemes. I've many questions about what kind of mixes are best to be recorded in order to create a complete and balanced acoustic model, such as:

  • are all pronounceable phoneme-combinations to be recorded, even if some are not used in the target language?
  • are phoneme-combinations to be recorded proportionally to their frequency in the language?
  • what if text-voice coupling is done with not existing words?
  • ......

This is because I'd like to start developing, in future, a python desktop application to assist volounteers on serially recording prompts with short collections of phonemes, then zipping and sending them automatically when online. I'm not sure this would be useful, so I started asking.
At the moment I'm isolating valid "atomic" phoneme mixtures for Italian, so that dynamic creation, prompting and recording would be quite easy.

I know these ideas might seem crazy, feel free to be sincere when replying.


Thanks

--- (Edited on 7/24/2010 6:47 am [GMT-0500] by scadiforever) ---

Re: Dynamic prompt creation
User: nsh
Date: 7/25/2010 1:04 pm
Views: 2354
Rating: 13

> I'm here simply to ask you to point me at some resources regarding the generation of prompts as mixtures of phonemes.

People are rarely doing that, that's why it's not really covered anywhere.

> I've many questions about what kind of mixes are best to be recorded in order to create a complete and balanced acoustic model

To get good acoustic model you need to record real speech, not any artifical phoneme combinations.

 

--- (Edited on 7/25/2010 22:04 [GMT+0400] by nsh) ---

PreviousNext