Ecosyste.ms: Timeline
Browse the timeline of events for every public repo on GitHub. Data updated hourly from GH Archive.
JiajunDou created a comment on an issue on SWivid/F5-TTS
@SWivid 感谢回复，看来遇到类似问题只能自己微调了， 关于落下我又测试了几遍，这个版本的确是读le4 xia4
Narcooo opened an issue on SWivid/F5-TTS
推理的音频偶现叠音或吞字
我使用了main.flac这个作为ref_audio，同时拿掉了输入的ref_text使用了whisper转录，在我自己测试的样例上他会出现类似于’如果果‘，‘存存在于’的音频，出现的位置相对随机（不在chunk的首尾），亦或者是'存于'，这个问题怎么缓解呢。
SWivid created a comment on an issue on SWivid/F5-TTS
@JiajunDou 我试了下这个case确实有问题，而且暂时没有很好的解决方案 应该是训练集里`落下`基本只有`luo4 xia4`读法的语音（尽管文本在转拼音时候会既有`luo4 xia4`也有`la4 xia4`），所以模型在这两个词一起出现时候，都会对应成`luo4 xia4`的读法 而像是`蜡烛`->`la4 zhu2`就可以正常读对`la4` 这种需要在包含正确读法...
JiajunDou opened an issue on SWivid/F5-TTS
convert_char_to_pinyin(text_list) 生成的拼音和最终生成语音结果不符
> 我们的技术完全美没有被全世界落下。 > [['wo3', ' ', 'men', ' ', 'de', ' ', 'ji4', ' ', 'shu4', ' ', 'wan2', ' ', 'quan2', ' ', 'mei2', ' ', 'you3', ' ', 'bei4', ' ', 'quan2', ' ', 'shi4', ' ', 'jie4', ' ', 'la4',...
SWivid created a comment on an issue on SWivid/F5-TTS
@justzmq 现有的code已经支持这种形式的inference（即batch，可以看`cfm.py`的`sample()`函数 只是如上面所说，因为一个batch里长短不一样所以要加mask有额外开销 目前用fp16推理速度已经还不错的，这些更多是工程上的优化。欢迎PR~~
PhamDangNguyen closed an issue on SWivid/F5-TTS
When I fine-tune model, Do i need to modify "convert_char_to_pinyin" by other function ?
Chinese is a unique language. When I want to train and process with other languages, such as Vietnamese, Spanish, or Lao, do I need to change the data processing method? Specifically, the function ...
PhamDangNguyen created a comment on an issue on SWivid/F5-TTS
@vu-the-dung, oh no problem. But i think the data needs additional periods and commas because the model will learn to pause accordingly. Also, capital letters will help the model learn to emphasize...
justzmq created a comment on an issue on SWivid/F5-TTS
> 可以，但是需要加口罩，不一定会快 感谢大佬回复！那么之后有测试或者添加这个功能的计划么，有的话我会持续关注的😊
PhamDangNguyen created a comment on an issue on SWivid/F5-TTS
@vu-the-dung I think the data needs additional periods and commas because the model will learn to pause accordingly. Also, capital letters will help the model learn to emphasize tone in the audio. ...
PhamDangNguyen created a comment on an issue on SWivid/F5-TTS
@vu-the-dung please give me some example, i'm newbie in TTS
