-
-
Notifications
You must be signed in to change notification settings - Fork 217
Description
The current implementation of the Phonemizer library does not handle currency expressions correctly. For instance, the sentence:
“It now wants to raise up to $260 million, nearly half of the original $504 million it had hoped to reap.”
is phonemized as:
ɪt naʊ wɔnts tə ɹeɪz ʌp tə dɑːlɚ tuːhʌndɹɪd sɪksti mɪliən nɪɹli hæf ʌvðɪ ɚɹɪdʒɪnəl dɑːlɚ faɪvhʌndɹɪd fɔːɹ mɪliən ɪt hæd hoʊpt tə ɹiːp
This output incorrectly phonemizes “$260 million” as “dɑːlɚ tuːhʌndɹɪd sɪksti mɪliən” and “$504 million” as “dɑːlɚ faɪvhʌndɹɪd fɔːɹ mɪliən”.
Suggested Solution:
Implement a preprocessing step to detect and correctly format monetary expressions before passing them to the Phonemizer.
Alternatively, consider using a different phonemizer, such as misaki (sililar to original kokoro tts project and it is the bestest), which has better handling of such expressions.
Additional Notes:
This issue may also affect other special cases like numbers, time, ordinal numbers, measurement units, ratios, and percentages. However, further testing is needed to confirm the extent of the problem.