The French AI TTS is excellent, and it nearly as good as an audiobook. Only the inevitable bumpiness between sentences makes it sound slightly less than natural.
For Japanese, however, the pronunciation errors are so numerous and the pacing so slow and choppy, that it’s useless. Words written in hiragana or katakana are pronounced properly, but anytime there are kanji, the pronunciation error rate is at least 20%. Even relatively common words are regularly mispronounced.
What’s odd is that the words are totally mispronounced when part of a sentence, but pronounced perfectly when a single word is selected. I understand that Japanese is a challenging language to work with, but if this is supplied by ElevenLabs and included in the highest-tier subscription, shouldn’t the quality be substantially better? It seems rather slapdash and shoddy tbh, arguably even worse than a robotic voice from 15 years ago.
An example (private lesson):
- 受容 is pronounced properly as じゅよう when the word is selected, but incorrectly as しゅよう in the sentence.
- 潜む is pronounced properly as ひそむ when the word is selected, but incorrectly as すむ (or something similar - it’s mumbled) in the sentence.
- 実践的 is pronounced properly as じっせんてき when the word is selected, but incorrectly as じっけんてき in the sentence.
- 極めて is pronounced properly as きわめて when the word is selected, but incorrectly as あつめて (or something similar) in the sentence.
This is at least 4 pronunciation errors in a single sentence. Note that this is the lesson generated audio. I used Shizuka but the other AI voices like Hinata encounter the same problems.
