Reddit用户比较六款TTS后看好Omni Voice
一名Reddit用户使用自选声音样本和口音测试,将Omni Voice与Moss-TTS、Breeze-TTS、Chatterbox-TTS、VibeVoice和Fish Audio S2 Pro进行比较。作者称Omni Voice通过了全部受测口音,生成音频的速度显著领先,并且在多段落单样本声音克隆方面优于其他模型。作者同时指出,Omni Voice偶尔会出现伪音,情绪和表现力控制选项少于部分竞品,而Breeze-TTS和Fish Audio S2 Pro的音频听起来更清晰。这些结论来自一次个人主观比较,并非标准化基准测试结果。
该报告为Omni Voice的口音复刻、生成速度和多段落单样本声音克隆表现提供了初步实践信号,但样本由作者自选、评测方法带有主观性,结论的适用范围因此受到限制。
There are tons of use cases out there - storybooks, agents, etc. But my main goal was to find which TTS has the best voice clone options, comparable to Elevenlabs.
I have a shortlist of voices that I run every tts through to see how it handles accents specifically, which to me is the best demonstration of its versatility. I'd recommend you try these out as well, and you'll pretty quickly find the weakness in a model. Some actresses with good accents to test are: Rebecca Ferguson, Charlotte Rampling, Mariangela Melatto (this is the model-killer), and Ornella Muti.
I also like to test generic/fake accents of cartoon characters. My particular go-to is Morgan Lofting's Baroness. Her accent isn't based on any particular culture, so models will struggle with placing it and emulating it.
On the male side, you can snatch Javier Bardem's Stillgar from Dune or any one of Jeff Goldblum's performances (the coffee shop scene of The Fly has a great monologue) to grade a model's capacity.
I ran these voices and others through some of the top TTS models out there, like Moss-TTS, Breeze-TTS, Chatterbox-TTS, VibeVoice, Fish Audio S2 Pro, and Omni Voice.
#Omni Voice#Moss-TTS#Breeze-TTS#Chatterbox-TTS#VibeVoice#Fish Audio S2 Pro#ElevenLabs