Hacker News new | ask | show | jobs
by ubercow13 922 days ago
Pretty sure his channel is successful due in no small part to his voice. It couldn't not be, considering the whole thing is just voiceovers. You also often see people commenting as such on his videos.
1 comments

He takes great care of that; I remember him mentioning some early video, where he laboriously edited out the plosives (hissing sounds) out of his recorded voiceover
He spoke about that in a recent podcast as well. He really hates mouth noises in general so he takes the time to fix all of that in post. Quite alot of work.
I think I know what you mean, but the idea of "hating mouth noises" meant literally is funny to me because that's all that speech is
The sound of voice is largely a result of things happening behind the mouth, in the throat. I think most of what people refer to when they say mouth noises is lips and cheeks.
I thought that only vowels were considered to be "throat" sounds, and that most consonants came from the mouth? I guess a lot of them from tongue and teeth (like "s", and "t"), but I'm not sure how I could make an "m" or a "p" sound without using my lips
When you make a noise from the voicebox, the noise is what's heard. It's modulated by the lips and tongue but the lips and tongue don't add noises, they just envelope them.

You have a point about "s" and "th" which is a whistling between tongue and palate.

Munching sounds, those can... Be distracting. Some super sensitive microphones seem to pick up those a lot more than a human ear
Your username is hilariously appropriate for someone commenting on what the most significant part of human speech is!
This also made his voice unusually easy to clone, so there's an early AI parody of him called NileGreen.
I wonder how many people are doing that much extra work behind the scenes. Ross Scott was recently talking about manually editing out some very subtle sound that wasn't being filtered out from certain words.
You can remove alot of it with mic choice/technique, in the same ways that you can enhance it.

There are also plugins that take care of at least most of it now, but it's not going to get rid of every instance of it. In current year, if you want 100% of it, it makes more sense to just train a model on your speech and fake all of the audio instead of recording it, then re-record only the parts you don't find acceptable.

Do you know of any free and open source voice cloning models? I once read a suggestion to use a recording of one's own voice as an alarm clock, as we respond strongly more our own voices. Perhaps there is some nascent narcissism of mine to be developed for fun and profit!