Re: [Espeak-general] Problems building espeak and voice data from source

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

On 28 June 2010 23:39, Jonathan Duddington <jo...@ta...> wrote:
> On 28 Jun, Reece Dunn <ms...@go...> wrote:
>
>> I now have this building, but running espeak causes a segmentation
>> fault
>
>> with $PLATFORM=big_endian
>
> I may be a endian compatibility problem with the data.  Unfortunately I
> am not able to test on a big-endian processor here.

Actually, that step is not needed as the platform is actually little
endian (64-bit Intel i7):

platforms/big_endian/espeak-phoneme-data espeak-data-local
espeak-data-local espeak-data-local/phondata-manifest
Host seems to be little-endian ..

I am defaulting to that so that the makefile will work -- I would get
errors if PLATFORM is not defined.

> The little-endian data in espeak-data in the eSpeak download
> (specifically espeak-data/phondata,phonindex,phontab) is wrong for a
> big-endian processor.  Either you must re-compile the data using
> espeakedit (which I think you have done) or use a program
> 'platforms/big_endian/espeak-phoneme-data' in the eSpeak-source
> download to convert it.

So you don't need to run espeak-phoneme-data if you run espeakedit?

Q: Do you need to run espeakedit to construct the espeak-data, or is
it sufficient to use what is in the svn repository?

>> 0x00007ffff7867286 in CalcLengths (tr=0x614560) at setlengths.cpp:751
>> 751                           pitch_start = p->pitch1
>> ((p->pitch2-p->pitch1)*pitch_env[0])/256;
>
> What's the immediate cause of the segfault?  What's the value of
> variable  'env2' before the segfault (you could use a printf() to get
> its value).

I get:

----- 8< -----
env2 = 1
env2 = 105
Segmentation fault
----- >8 -----

> How did you run eSpeak?  Did you use a command such as:
>
> espeak -v en "Hello world"
> ?
> Which voice (eg. 'en') did you use?

$ espeak -v en hello
env2 = 1
env2 = 105
Segmentation fault

$ espeak -v en-us hello
Bad voice option: option 12 1
Unknown phoneme table: 'en_us'
env2 = 1
env2 = 105
Segmentation fault

$ espeak -v es hello
env2 = 1
env2 = 1

$ espeak -v de hello
env2 = 101
Segmentation fault

So es (and possibly others) are ok, but en, en-us and de are not.

>> This is with:
>
>> espeak-data-local: espeak-phoneme-data espeak-data
>>       rm -rf espeak-data-local
>>       cp -a $(HOME)/espeak-data espeak-data-local
>>       platforms/$(PLATFORM)/espeak-phoneme-data espeak-data-local
>> espeak-data-local espeak-data-local/phondata-manifest
>>       mkdir -p espeak-data-local/mbrola
>>       mkdir -p espeak-data-local/mbrola_ph
>>       mkdir -p espeak-data-local/soundicons
>>       mkdir -p espeak-data-local/voices/test
>>       rm -rf espeak-data-local/dictsource
>>       rm -rf espeak-data-local/phsource
>>       rm -rf espeak-data-local/phondata-manifest
>> ----- 8< -----
>
>> to generate the espeak-data directory (from the content of
>> epeak-data-local, with $PLATFORM=big_endian).
>
> I'm sorry, but I don't understand what this means.

I apologise for not making this clear.

I am copying voices, phsource and dictsource to ~/espeak-data. Then I
am building the voices using espeakedit.

I am then copying the built voices to ./espeak-data-local and running
the platform's espeak-phoneme-data on it. I am then making directories
to match what directories are present in the Ubuntu espeak-data
package (probably not necessary) and removing the sources (not
strictly necessary, but keeps the installed espeak-data directory
cleaner).

> Does eSpeak pick up your data from $HOME/espeak-data which has been
> compiled or converted for big-endian?

>From an installed version that has been through `espeakedit --compile`
and `espeak-phoneme-data` (installed to /usr/share/espeak-data).

- Reece