CosyVoice icon indicating copy to clipboard operation
CosyVoice copied to clipboard

Mispronunciations in Japanese TTS results / 日语TTS出现拼读错误

Open addicted0010 opened this issue 6 months ago • 3 comments

Locally deployed on my MACBookPro (M2 chip). Results: cozyvoice_results.zip

  1. Input text: 自民党は、岸田総理大臣の後任を選ぶ総裁選挙について、9月12日に告示し、27日に投開票を行うことを決めました。総裁選挙の期間は、いまの規程が設けられて以降、最も長くなります。 Cozyvoice result: audio1.wav

Issues: There are 6 mispronunciations occurred in the result. I have marked with BOLD font.

  1. Input text: 晴子、お前が見つけてきた変な男は、湘北に必要な男になったぞ Cozyvoice result: audio2.wav

Issues: '晴子' was wrong and not sure what it said. Only the middle part remains. The last part was missing.

  1. Input text: 自分がバスケの試合をしているも、そうでないも、心が折れそうな自分を奮起させる最高の名言です。この言葉から、何度も力をいました。 Cozyvoice result: audio3.wav

Issues: 4 mispronunciations occurred in the result. I have marked with BOLD font.

In conclusion, this result means cozyvoice is basically unusable for Japanese language. The mispronunciation issue looks quite strange. These are all simple words in Japanese. Is it caused by the polluted original Japanese training material? Is there any improvement plan for Japanese capability?

*BTW, Chinese TTS result is great (wow!)

addicted0010 avatar Aug 21 '24 02:08 addicted0010