search

myshell-ai / MeloTTS

High-quality multi-lingual text-to-speech library by MyShell.ai. Support English, Spanish, French, Chinese, Japanese and Korean.
MIT License
4.49k stars 561 forks source link

读中文的年份的时候有错误 #85

Open jonahzheng opened 6 months ago

jonahzheng commented 6 months ago

比如‘2018年’ 生成的语音会念成er qian ling yi shi ba nian，应该是‘liang qian ling yi ba nian’ 或 ‘er ling yi ba nian’ 这个要如何修正，重新用语料训练吗？

MissingTwins commented 5 months ago

Yes, there are many issues with how years are pronounced, and the percent symbol (%) is not pronounced at all for both Chinese and English. Also, many capitalized abbreviations are not correctly pronounced, such as HBO, CCTV, XBOX, and HIV.

··· from melo.api import TTS

device = 'auto' speed = 1.0 text = '说中文, ni3 hao3吗, BBC广播, CNN新闻, 美国HBO, NBC转播, CCTV一套, HIV病毒, 100%不合格, 列宁格勒，希特勒，P2P软件, Sony的PSP, 微软的XBOX游戏机, 2018年初，1931年, 轰炸地堡，数了数几只羊'

model_tts = TTS(language=‘ZH’, device=device) speaker_ids = model_tts.hps.data.spk2id model_tts.tts_to_file(translated_output_text, speaker_ids['ZH'], 1.wav, speed=speed) ···

![Demo Audio]

liuxiong21 commented 5 months ago

你的中文预训练模型在哪里下载的？

Simon-chai commented 3 months ago

model_tts.tts_to_file(translated_output_text, speaker_ids['ZH'], 1.wav, speed=speed) speaker_ids['ZH'] raise error AttributeError: 'HParams' object has no attribute 'ZH' what have I missed?

gurudk commented 2 months ago

中文开源生态真的道阻且长，按理说，这种开放语料就应该通过众包的方式解决，英文的开放语料库多多了