Re: [Espeak-general] [ANNOUNCE] Orator - Text-To-Speech GUI frontend

[Reece D. <ms...@go...>]

2009/5/19 Jonathan Duddington <jo...@js...>:
> On 19 May, Reece Dunn <ms...@go...> wrote:
>
>> The eSpeak engine is using the speak_lib.h API. At the moment, there
>> are a few issues with the eSpeak engine:
>>   *  the speak_lib.h API does not report the sentence length, so
>> cannot correctly highlight the sentences being read.
>
> The events list which eSpeak passes to your SynthCallback() callback
> function includes events which indicate the start of each word and
> sentence (espeakEVENT_WORD and espeakEVENT_SENTENCE).  So you can
> highlight each word or each sentence as it is being read.

If I handle espeakEVENT_WORD, the espeak_EVENT.length contains the
length of the word being spoken, but espeakEVENT_SENTENCE returns 0. I
am using espeak_EVENT.text_position to get the offset (which works for
both events).

How do I determine how long the sentence is? Even from the API docs:
>    int text_position;    // the number of characters from the start of the text
>    int length;           // word length, in characters (for espeakEVENT_WORD)

>>   *  the speak_lib.h API does not support phrase events, so the UI
>> treats them the same as sentences.
>
> Please explain what you mean by a "phrase event".  How do you define a
> "phrase"?  You should be able to use the espeakEVENT_WORD events to
> match the current speaking position with words on your text, and
> therefore determine which phrase is being spoken.

When using the phrase event in Cepstral, it returns the range between
the current speaking position and any commas, semi-colons, etc. E.g.
"Hello, how are you?" would be split into "Hello," and "how are you?"
phrases.

>>   *  there is no pause functionality available in speak_lib.h, so this
>> option is disabled.
>
> You could implement pause by either of two methods:
>
> 1.  Stop speaking, and note the position of the last word that was
> spoken (using the espeakEVENT_WORD event).  To resume, start speaking
> from the next word in the text.
>
> or
>
> 2.  Don't let your SynthCallback() function return until you want to
> resume speaking text.

The first method would probably be easiest to implement.

Thanks for the feedback,
- Reece