Record new words options

[Pgordon]

Hi all.

I\'ve never previously hit any (serious) problems with the default wordlist in my Comfort panels. Therefore I\'ve never looked at any of the wordlist management options, so I\'m completely in the dark regarding this area of Comfort configuration.

I\'ve read the posts in the wordlist forum, and I\'ve downloaded & installed the voiceproc & voiceplay tools... not done anything serious with them yet, - still trying to get my head round the instruction document..

For my own understanding, can I just check/confirm a couple of basic (mis)understandings?

- All of the words on the wordlist detailed in the documentation have been recorded by the same voice actor, but only a subset are downloaded to Comfort at any one time?

- Using the voiceproc tools, individual words selected for download to the panel can be altered? - So if a word I want is on the wordlist I can have it installed in the panel in place of any other word currently there that I don\'t use?

- However, if I want a word that has never been anticipated and thus recorded by the voice actor, I can still get that word in there, but it would have to be recorded first as a new sound file?

It\'s the last part that I may wish to investigate in greater detail... I\'m not overly keen on having a disjointed voice experience with just the odd word in the vocabulary spoken in a distinctly different voice... so I think I\'ve gathered that there are a couple of options...

- Pay Cytech to call in the same voice actor to record the specific extra words I need

- re-record the entire wordlist myself

- *IF* I can tolerate just the very occasional voice difference, I could record just my custom words myself?...

The last option would obviously be the cheapest & least effort, and it might be the case that among my friends & family I may be able to find someone who can do a reasonable impression of the current Comfort voice... but then I got to thinking... TTS via the PC can be pretty darn good these days.... I have used in the past a number of alternate TTS engines such as AT&T Natural Voices which I\'ve been reasonably impressed with, and could certainly tolerate having my entire Comfort vocabulary spoken in some of the TTS voices I have previously heard....

And if I were to use a decent, natural sounding TTS engine, I could quite easily have it record the entire wordlist, as I could just feed in the published wordlist as a text file, & get a bunch of WAVs out the other end with little more than a few clicks...

Has anyone done this? I\'ve just seen there is a section of the instruction document that deals with this; still have to read up a bit more.... What\'s considered currently the \'best\' (i.e. most natural sounding) TTS engine for the Windows platform?

TIA

Paul G.

[slychiu]

- All of the words on the wordlist detailed in the documentation have been recorded by the same voice actor, but only a subset are downloaded to Comfort at any one time?

Yes there are many  recorded words which have not been used in the vocabulary

- Using the voiceproc tools, individual words selected for download to the panel can be altered? - So if a word I want is on the wordlist I can have it installed in the panel in place of any other word currently there that I don\'t use?

Yes

- However, if I want a word that has never been anticipated and thus recorded by the voice actor, I can still get that word in there, but it would have to be recorded first as a new sound file?

Yes

It\'s the last part that I may wish to investigate in greater detail... I\'m not overly keen on having a disjointed voice experience with just the odd word in the vocabulary spoken in a distinctly different voice... so I think I\'ve gathered that there are a couple of options... - Pay Cytech to call in the same voice actor to record the specific extra words I need

It may be dfficult to find the same voice talent (Myma) as she did the recording many years ago, and recording a few words might be quite expensive due to studio time etc

- re-record the entire wordlist myself

that is possible

- *IF* I can tolerate just the very occasional voice difference, I could record just my custom words myself?...

yes

What are the words you need? There is a chance that it has been recorded and you may not have to do any recording

[Pgordon]

Thanks Chiu.

In this particular instance, there are just a couple of words I\'ve come across that I wanted to use... these included: good, bad, morning, afternoon, evening, and perhaps one or two others that I\'ve since forgotten.
It occurred to me in the short term, perhaps I could record the exact phrase I want as a reminder message, and play that in my response in lieu of actually having Comfort able to speak them...
If I find a TTS voice that is similar enough to the current voice, I could record the TTS output rather than speaking it myself... Could I use a UCM-Audio to capture the output from my PC in order to get the best possible recording? - Stereo out from my soundcard directly to Line-in jack socket on audio submodule?

Is there perhaps some prospect here of an area for future development for Comfort? - move over from a real human voice to a TTS voice, and roll the functionality into either Comfigurator or Voiceproc? - I could forsee a day when all questions & queries regarding matching new words with the existing voice go away... Customers could give their comfort panel *any* vocabulary they want in any TTS voice they care to use... you could make known what original TTS voice you use to create the wordlist shipped with Comfort so customers could choose to purchase the same voice if they wished, alternatively, you just have to code to allow any SAPI voice to be controlled... In terms of the user experience, it would be *fabulous* to have Comfigurator, in addition to offering the drop-down list of words in the \"Announce Words\" action, there is also a text-entry item where absolutely any word could be entered... Comfigurator recognises the word is not in the current wordlist, and invokes the TTS engine to speak it directly into a file which is then processed just as it is in wordproc today... the text entered in the box becomes the text required for the keypad screens... so new vocabulary download files are built completely automatically.... Comfigurator knows that new words have been generated, and so automatically invokes the vocab upgrade procedure when the user next chooses to write the config to the panel...

I think everything necessary to achieve this is already available... we have a highly manual process for doing this today, so no hardware changes would be required, just the engineering effort in the software...

Just a thought... :-)

Paul G.

[tman]

[user=139]Pgordon[/user] wrote:

If I find a TTS voice that is similar enough to the current voice, I could record the TTS output rather than speaking it myself... Could I use a UCM-Audio to capture the output from my PC in order to get the best possible recording? - Stereo out from my soundcard directly to Line-in jack socket on audio submodule?

You want the word as a trimmed WAV file on your PC for the VoiceProc software to convert into the vocab file which you upload into the Comfort board. Playing it into a UCM with the AUD01 plugin won\'t work for what you\'re trying to do. It isn\'t necessary to play it out of any soundcard and doing so will actually degrade the quality since you\'ll be doing an unecessary digital to analogue to digital series of steps.

[slychiu]

If you download Myma_IVS.zip from https://www.cytech.biz/myma.html you should be able to see all the words which have been recorded in WAV files

You should be able to find the words you need from one of the TTS services which sound similar to Myma whichcan be downloaded in wav format. We have done this before and the results are acceptable. Bear in mind that the voice is compressed so you are not hearing very high quality speech anyway

Having Comfigurator select any random text and getting the file would not be difficult but the problem is the voice must be stored in the Flash and there may be memory limitations in the current hardware

[juwi_uk]

Maybe incorporate support for plugging a MicroSD flash module into the system somewhere so for example 64GB could be say available for storage and use by the system and easier to upgrade in future?

J

[juwi_uk]

BTW is it correct that you can only store 240 words in total?

[slychiu]

yes the number of words is limited to 240

[juwi_uk]

Ah.  Next on the list for 16bit addressing then?

[Pgordon]

 

[user=139]Pgordon[/user] wrote:

If I find a TTS voice that is similar enough to the current voice, I could record the TTS output rather than speaking it myself... Could I use a UCM-Audio to capture the output from my PC in order to get the best possible recording? - Stereo out from my soundcard directly to Line-in jack socket on audio submodule?

You want the word as a trimmed WAV file on your PC for the VoiceProc software to convert into the vocab file which you upload into the Comfort board. Playing it into a UCM with the AUD01 plugin won\'t work for what you\'re trying to do. It isn\'t necessary to play it out of any soundcard and doing so will actually degrade the quality since you\'ll be doing an unecessary digital to analogue to digital series of steps.

I think you misunderstood my point re the audio module... at this stage I\'m referring to the temporary workaround of recording a reminder message and playing that back rather than using the real Comfort voice... to record a reminder message, we\'d normally speak across open air into a microphone via either a keypad or a house phone... I\'m thinking that I should be able to record directly from the TTS engine on my PC into the line-in on the audio module rather than speaking into a keypad...