Google TTS accepts X-SAMPA, an ASCII encoding of the IPA, in lieu of Unicode IPA.
(This is not a request or report but a placeholder for the topic, because I couldn't find one.)
Description
Description
Related Objects
Related Objects
- Mentioned In
- T328705: Phonos - update help
T324738: Add voice="" attribute, for selecting engine voice
T322787: IPA can be optional only if a wikidata item or audio file is provided
T320523: Display error when an engine does not support the requested language - Mentioned Here
- T323912: IPA validation and conversion is flawed/unnecessary
P898 cxserver-restbase
Event Timeline
Comment Actions
Am I right in thinking that the key change here would be to change the alphabet attribute in the SSML? Google's documentation has this example:
<phoneme alphabet="ipa" ph="ˌmænɪˈtoʊbə">manitoba</phoneme>
to
<phoneme alphabet="x-sampa" ph='m@"hA:g@%ni:'>mahogany</phoneme>
Would we add a new x-sampa="" attribute to the Phonos parser tag (that would not be able to be used in conjunction with ipa="")? Or because there's a deterministic mapping from IPA to X-SAMPA (is there?), do we instead translate the IPA? It looks like Wikidata doesn't have a X-SAMPA equivalent of IPA transcription (P898), I assume because it is possible to calculate it from IPA.
Is there any validation required or possible for the X-SAMPA value? I've done a very quick search and haven't found a PHP library for working with it. If we can do the mapping from IPA to X-SAMPA, should we use the mappings from Wikidata (e.g. ʊ, near-close near-back rounded vowel (Q1137807) maps to U)?
Comment Actions
@Samwilson Like I said, this is not a request. More like an inquiry on whether you're going to support X-SAMPA.
I was thinking more like a type="x-sampa" or format="x-sampa" that specifies the type of the input (currently in ipa="", which would probably require renaming), i.e. something that directly translates to alphabet in SSML. If you're going to support Mandarin/Cantonese and therefore Pinyin/Jyutping as input, you're going to need something like that (since creating an attribute for every scheme that becomes available doesn't seem like a good long-term solution).
Is there any validation required or possible for the X-SAMPA value? I've done a very quick search and haven't found a PHP library for working with it. If we can do the mapping from IPA to X-SAMPA, should we use the mappings from Wikidata (e.g. ʊ, near-close near-back rounded vowel (Q1137807) maps to U)?
Wiktionary has Template:x2i that converts X-SAMPA to IPA (data: Module:IPA/data/X-SAMPA). As summarized here, X-SAMPA actually doesn't encode the entire IPA. And as we've established (I hope) at T323912, what is "valid IPA" is a shaky concept. For example, Google lists [p͈] for Korean, of which p* is listed as the X-SAMPA equivalent. But you won't find this in any X-SAMPA specs as [p͈] is not IPA (it's extIPA). Just like with IPA, trying to see if a string is "valid X-SAMPA" strikes me as a lost cause (I guess aside from the fact it must be all ASCII by its very nature).
So I don't actually recommend doing pretty much anything to "support", i.e. accommodate, X-SAMPA in Phonos other than to simply allow the user access to the alphabet option, so the flexibility offered by Google is offered to Phonos users as well.