Sound icons handling

[Hynek Hanke]

On 1.9.2010 10:16, Halim Sahin wrote:
> On Wed, Sep 01, 2010 at 01:09:08AM -0400, Trevor Saunders wrote:
>    
>> that makes no sense, we are asking the module to play a wave file.
>>      

Hello all,

we used to do sound icons directly in Speech Dispatcher, but
decided against it. Let me explain.

We are not asking the synthesizer to play a wave file, we are
asking the TTS to include a sound among the speech it produces.
This sound can be embedded in the speech mid-sentence (see SSML
AUDIO element). There can also be a pre-synthesized text in the
form of audio.

In other cases, such as a the SSIP key command, the TTS might
choose either playing synthesize the key name or play
a sound icon, or many times combine those (play a sound
for 'shitf', synthesize the letter).

So the TTS, if it's good, must handle sound icons already.
Handling the special case where the whole message is composed only
of the sound icon itself in Speech Dispatcher would be a duplication
of functionality.

It is also principially not a task for Speech Dispatcher to
fabricate any sound data. We can of course speak of emulations
for those synthesizers who can't support it.

>> Which synth can play soundicons except espeak?
>>      

Festival. It's also a requirement of both TTS API and SSML.

This is unrelated to the issue of resampling, because you can
(and will) get different sample rates from the TTS even when
switching voices, which can also happen mid-sentence.

Best regards,
Hynek