E.g. for 'port' this goes through following transformations:
1. written 'port' to phonetic [p'o*@-tS] for eSpeak NG
2. espeak [*] was mapped to MBROLA [r] and [@-] was just skipped
resulting [rt] which is not allowed diphone in br1 voice.
Description of fix:
mapping phonetic [*@-] to [r2] (and skipping [@-]) makes [r2-...] which is
in diphone database. That fixed warnings:
mbrola: Warning: r-t unkown, replaced with _-_
mbrola: Warning: r-d unkown, replaced with _-_
mbrola: Warning: r-n unkown, replaced with _-_
mbrola: Warning: r-m unkown, replaced with _-_
mbrola: Warning: r-s unkown, replaced with _-_
It is possible -- especially at higher speeds -- for the n at the
end of a word to be velarised if the next word starts with a velar
plosive. I prefer the velarised sound between word boundaries, but
others do not. As such, limit the velarisation to within the word
only.
[1] https://en.wikipedia.org/wiki/English_phonology
There is no fr-CA language dialect dialect file, nor has there
been in the espeak/espeak-ng history.
The fr-CA phoneme table defined the additional phonemes [U], [y"],
[E:], and [A]. The fr-FR phoneme table mapped these to [u], [y],
[E], and [a] respectively. This meant to describe the phonemes that
have not been merged in fr-CA, but have merged in fr-FR, as described
in https://en.wikipedia.org/wiki/Quebec_French_phonology.
The only use of these phonemes in the French dictionary was with
the word "tramway" using [E:]. This has been replaced with [E] to
match the rest of the dictionary file transcriptions, and those
phonemes for fr-CA have been removed.
If fr-CA is supported in the future, the French dictionary would
need to be modified to reverse the merger of the phonemes that are
not merged in fr-CA, with the merge logic being performed in the
fr-FR phoneme table like in the code being removed in this commit.
These are some improvements to the german voices, both default and mbrola. It fixes some sounds of phonemes and also pronunciation of words. I also modified the mbrola phoneme and voice files to make them sound more natural. The stressLength setting is that it matches the way espeak deals with the lengths.
Valdis Vitolins