ajbrouw@neabbs.UUCP (ALBERT-JAN BROUWER) (11/01/89)
If people sum up the advantages of owning an Amiga, sooner or later the built-in speech capability will be mentioned. But be honest, you don't use your Amiga to recite documents, eventhough this would free your hands and eyes for other jobs, because the the thing is unintelligible too often. So what can be improved? Firstly, textfiles are designed to be displayed instead of being copied to SPEAK:. So they often contain things like: ************* * Chapter 1 * ************* The current narrator device insists on saying "asterisk asterisk ..." A better way would be to say "eleven asterisks" or to ignore more than 4 sequential occurances of the same character. In general, it would be nice if the narrator device is modified in order to deal with the most frequent ways of graphical text usage. Secondly, about 1 in every 10 words is mispronounced. This isn't a bad score, given the very weak link between how an english word is spelled, and how it is pronounced, but it doesn't suffice for document recital. What follows is a proposal on developing better software for translating words into phonetics. I sure aint gonna program all this, but perhaps someone at Commodore-Amiga would like to take up the challenge. The first thing that's needed is a program that creates a word-frequency table from a representative amount of text files, and allows one to enter the phonetics for, say, the 10.000 most frequently used words. This is a rather boring job, and it can be left to anybody qualified enough to use a lexicon. Next, a program is needed that compiles the given table into a relatively small (10-20K) data table, used by the narrator device for translation. This data table could contain: - Rules for pronouncing words containing certain sub-strings. - Rules for recognizing prefixes/postfixes like un-, non-, -ble etc. (Try echo >speak: "unable" for example) - Rules stating an exception to other rules given certain conditions. - Any other imaginable way of generalizing common relations between spelling and phonetics into rules driven by spelling characteristics. - Full phonetic spelling for the most common words defying analysis. This is only a very superficial description of a rather complicated program. Still, the quality of the output hinges on how well this compilation program performs. Furthermore, the narrator device has to be modified in order to interpret the thus created table when translating. All hard-coded translation control logic won't be needed anymore. Next to better translation performance, this set of programs offers the possibility of creating any number of different translation tables. Imagine the following scenario: The narrator device could be made to contain its own small default translation table, thus supplying backwards compatability. The DEVS: directory might then contain a subdirectory named, say, lingos, (like keymaps) filled with translation tables for a selection of foreign languages, a super large English table for power users, improving the translation hit rate, and perhaps even a techie translation table built by processing news.amigatech messages! I'm rather fond of this idea, it seems in-line with the Amiga way of doing things. It will take some effort to implement it though, so it sure won't make it into the 1.4 O.S. Reactions? Comments? Pro Votes? -- -Albert (hp4nl!neabbs!ajbrouw) Optimize your options, both in quantity and quality.
jonabbey@walt.cc.utexas.edu (Jonathan Abbey) (11/03/89)
In article ajbrouw@neabbs.UUCP (ALBERT-JAN BROUWER) writes: +------ |If people sum up the advantages of owning an Amiga, sooner or later the |built-in speech capability will be mentioned. But be honest, you don't |use your Amiga to recite documents, eventhough this would free your hands |and eyes for other jobs, because the the thing is unintelligible too often. | | ... | |The DEVS: directory might then contain a subdirectory named, say, lingos, |like keymaps) filled with translation tables for a selection of foreign |languages, a super large English table for power users, improving the |translation hit rate, and perhaps even a techie translation table built by |processing news.amigatech messages! | | I'm rather fond of this idea, it seems in-line with the Amiga way of doing |things. It will take some effort to implement it though, so it sure won't |make it into the 1.4 O.S. |Reactions? Comments? Pro Votes? | |-Albert (hp4nl!neabbs!ajbrouw) >+----- Both this and your refinements to the non text handling sound very well thought out. I do remember that one of the improvements in 1.4 mentioned in the Amazing Computing (INFO?) DevCon/1.4 Preview was greatly improved speech handlers. Count this as a positive vote. // /\ /\/\ | Jonathan Abbey - jonabbey@doc.cc.utexas.edu - (512) 926-5934 \X/ / \ / \ | Wanted: Programmers interested in 3d graphics/modem games.
galvey@sts.sts.COM (11/03/89)
I like the idea! I have been using the Amiga for a few months to help my son on him memory verses. I usually spend about 50% of the time translating mis- pronounced words into phonemes (sp?). I also agree with the idea of a table... this could be used in different ways of saying the same word based on where you live (toe-mA-toe vs. toe-mAH-toe 8-). ============================================================================= |Greg Alvey | Internet: galvey@sts.sts.COM | |Semiconductor Test Solutions | Phone : (408) 727-2885 (work) | |4101 Burton Dr. | | |Santa Clara, CA 95054 | When all else fails, read the instructions! | =============================================================================