NPC voices, recommendations?

Hi All,

Was wondering what you all are using for NPC voices, especially with all the AI options out there now. I had a voice from a site (I type in text and it exported an mp3 file in a women’s voice, it was Nordic reminded me of the girl warrior on the show Vikings), I really liked it but then the website disappeared and I now can’t get any new dialog for that NPC.

I would really like to buy a program that I can be sure will exist until my project is complete, any programs/apps that you would recommend? or less desirable, a solid website.

I don’t have the funds to hire a voice artist/actor… :frowning:

Thanks,

V

I looked into voice AIs not long ago, and only the biggest providers had anything of real quality to offer, like ElevenLabs and Google AI Voices. I also tried some opensource solutions, but most of them sounded robotic., the only tools that produced something acceptable were voice-cloning models, though they struggled with diction.

Then just use text.

so here is a sample of voice I got from a site in the past (the site has since gone away):

Is this considered really poor quality?

I mean its not great, but personally I like it but maybe my hearing is crap… I would take this over text, I really hate text in a game…

I did this text to voice conversion about 5 years ago… so I was thinking resources now would be much improved… no?

It sounds raspy and wavery, which is a side effect of its age. However, it sounds flat and directionless, which is a problem all these generators still have. If I heard something like this in a game, the best case scenario is that I turn voices off. The worst case scenario is that I refund the game entirely because I can’t trust the developer cared enough to put effort into the game anywhere else.

Really poor, definitely not. Would say it’s okay for an indie game. However careful where you use what voice.
Not sure where Murgilod hears raspyness, what I hear is “breathiness”, almost a seducing voice! It may work well for a bar maid in silent environments - that sort of thing. As a tutorial voice or random quest NPC instructions, I’d go with something more neutral or up beat.

Just FYI Murgilod made his name here lately with being anti-AI - and I’m the opposite, to take all that with this grain of salt :wink:

Guiding the intention/intonation (is that “direction”?) of the voice is indeed recommended though. Some services allow for “meta input” or “audio tags” for exactly that purpose. https://elevenlabs.io/ was already mentioned and may be the best one around.

Personally I’ve had fun with this one: https://www.weights.com/ However they are primarily voice-cloning which does have a greater potential to alienate players, I’d assume.

You might be able to hire real voice actors from high school theater or local colleges for much less than older professional voice actors.

They’re likely to be the only ones left that are decent. Replicate Studios shut down back in the summer. That said I would be hesistant to use AI voices until I determined the way my audience would receive them.

well this is going good, tried to use their sample, errored out, see below…

Then tried to sign up for their website, and that errored out too…

Guess I’ll try later… thanks for the suggestions though…

V

Depends what you’re building. For pre-recorded voiced NPCs where you need high-quality TTS to batch-generate dialogue lines, ElevenLabs has the strongest voice quality right now and their Unity SDK makes offline audio generation straightforward. You render the lines, ship the audio files, done.

If you want NPCs that actually talk back to players in real time ie. voice input, LLM response, synthesized reply, lip sync, ElevenLabs alone won’t cut it because you’re also coordinating STT, an LLM, and blend shape animation. That’s a three-or-four plugin stack if you assemble it yourself, with all the version compatibility maintenance that entails. We provide a Unity integration for the full conversational NPC loop with Convai’s Unity Asset Store plugin. It handles dialog, long-term memory, actions, voice, and lipsync as one package. The free tier is genuinely usable for prototyping before you commit to anything, and the GitHub examples cover the common setup cases.