[espeak-ng:master] New Comment on Pull Request #810 Add some patterns to ja_rules
By rhdunn:

1. Add some patterns to ja_rules -- This commit looks good.
2. Add more ja_rules patterns -- This commit looks good.

Remove the pattern that causes the conversion to fail / Remove more unnecessary parts

Why are you removing the 'かあ' etc. patterns in these commits? These are supported "lengthen" combinations -- see e.g. https://en.wiktionary.org/wiki/%E3%81%8B%E3%81%82 (かあ). From https://en.wikipedia.org/wiki/Hiragana: > Hiragana usually spells long vowels with the addition of a second vowel kana; for example, おかあさん (o-ka-a-sa-n, "mother").

Thus, when a vowel Hiragana matches the vowel part of the previous Hiragana, the vowel is lengthened.

NOTE: The removed parts are not the small versions of the characters (ぁ) but are the full-sized versions (あ) so the two rules are different. This is causing the tests to fail.

[espeak-ng:master] New Comment on Pull Request #810 Add some patterns to ja_rules
By rhdunn:

Using separate vowels would mean that ああ would be pronounced differently to あぁ and あー, as /aa/ would have a slight pause compared to /a:/.

A possible solution would be to define a group for each long vowel set (a, e, i, o, u), e.g.:

.L22    ぁ ー // long a

then have a new rule such as:

    あ   a       // a
    あ (あL22 a       // aā
    ああ  a:      // ā
    あぁ  a:      // ā
    あー  a:      // ā

where the aā rule is a new rule to fix the case you describe. This would need to be done for each of the doubling rules.

This means that ああぁ and ああー will be pronounced as /aa:/. These rules cannot currently be simplified due to limitations in the espeak/espeak-ng rule processing logic. I'll raise a separate issue for that.

[espeak-ng:master] New Comment on Pull Request #810 Add some patterns to ja_rules
By tset-tset-tset:

I've modified this PR to only "add the missing pattern". The かあ problem is still there, but will be resolved after #812 is resolved.