Aratako

/

Irodori-TTS-500M-v2

like 141

Thank you for developing a great Japanese TTS for community. I tested and have some views

Advantages noted:

1. Accurately pronounces the pitch of homonyms: 橋 - 箸.

2. Changes the intonation at the end of sentences according to punctuation marks: "!", "~", "?", ".".

3. Automatically adds appropriate expressive SFX when input includes emoticons (laugh, sigh, etc.).

4. Clear audio quality, free from background noise.

5. Natural, native-like voice.

→ Sound quality is better than other TTS models currently in use.

Errors detected during testing:

1. Incorrect reading of numbers when written in numerals: 「２０歳」 is read as 「にじゅっさい」 instead of 「はたち」.

2. Incorrect reading of dates even when written in kanji: 「二月一日付で」 is misread.

3. Unstable voice during mass generation: reference audio is for a female voice but output is a male voice.

4. Sound changes are not applied when combining words: 「三階」 is read as 「さんかい」 instead of 「さんがい」.

5. Incorrect word reading: 「銀行」 is read as 「きんこう」 instead of 「ぎんこう」.

6. A single kanji character has multiple readings, but TTS does not yet support a mechanism to specify a reading. For example: 「角」 is read as 「つの」 while the desired reading is 「かど」; similarly with 「十分」.

7. English words are automatically converted to Japanese katakana readings, but there may still be cases where the reading is incorrect.

8. When reading single words, odd characters, or short sentences without context, TTS may add inappropriate expressions (gasping, cries, etc.). For example: 「あ」, 「母」, 「やめてください」, 「先生」.

9. Mispronouncing proper names: 「秋山先生」 is pronounced 「おきやませんせい」 instead of 「あきやませんせい」.

10. When reading a long passage, there may be strange sound effects inserted (audio interruptions, speaker breathing sounds).