A $30,000 line item is forcing 90% of indie studios to ship text-only dialogue. AI changes the math completely.Voice Acting Costs Are Killing Indie Game Development

A $30,000 line item is forcing 90% of indie studios to ship text-only dialogue. AI changes the math completely.

Here’s a number that stops most indie game developers cold: $250 to $350 per hour for union voice talent, with a 4-hour session minimum.

That’s $1,000–1,400 before a single line of NPC dialogue has been recorded. For a game with 15 distinct character types — heroes, villains, merchants, guards, companions, creatures — you’re booking 15 separate sessions. Even with non-union talent at $100–200/hour, the total budget for a modestly voiced RPG lands somewhere between $15,000 and $40,000.

For a studio like Naughty Dog or BioWare, that’s a rounding error. For a 2-person indie team working out of savings, it’s the entire budget.

So 90% of indie games ship with text-only dialogue. Not because the developers don’t want voice acting. Not because their games wouldn’t benefit from it. Because the economics make it impossible.

This is a problem worth solving. And I think AI is about to solve it — but not in the way most people expect.

The Current State of AI Voice in Games

Cloud TTS services like ElevenLabs have gotten remarkably good. The voice quality is there. A well-directed AI voice read can sound natural, emotional, and characterful. The “uncanny valley” objection from two years ago doesn’t hold up anymore.

But when you try to actually use these services for game production, you run into a different set of problems:

Per-character pricing kills iteration. Game dialogue direction is inherently iterative. “Try that line angrier.” “More sarcastic.” “Like she’s barely holding it together.” In a traditional voice session, once the actor is booked, additional takes are free. The marginal cost of “one more read” is zero. With per-character cloud TTS, every take costs money. Directors start self-censoring. They accept “good enough” instead of pushing for “exactly right.”

Generic voice libraries miss the point. Game developers don’t need “Female Voice 3” and “Male Voice 7.” They need archetypes. The grizzled mentor. The trickster companion. The alien creature with an otherworldly cadence. The ancient narrator who sounds like they’ve seen civilisations rise and fall. Most cloud TTS libraries are built for narration and content creation, not character work.

Consistency across sessions breaks. Generate the same character’s dialogue on Monday and Wednesday, and there’s no guarantee the voice will sound identical. Cloud models version-update. Session parameters drift. For a game where players interact with the same NPC across 40 hours, consistency is non-negotiable.

Unreleased scripts on third-party servers. Game studios are protective of their IP. Dialogue scripts for unannounced characters, plot twists, and story beats sitting on a cloud provider’s servers? That’s a leak risk. Even with NDAs and privacy policies, the dialogue exists outside the studio’s infrastructure.

What Game Developers Actually Need

I spent months talking to indie game developers about their voice production needs before building anything. The requirements that kept coming up:

Unlimited iteration at zero marginal cost. Try 10 readings of the same line. Try 5 different voices for one character. Regenerate all of a character’s dialogue after a script rewrite. Without thinking about cost.

Character archetypes, not generic voices. Voices designed for game contexts: heroes, villains, companions, NPCs, creatures, machines, narrators. Each with distinct personality and tone that makes casting decisions intuitive.

Local processing. Scripts stay on the studio’s machines. No uploads. No leak risk. This matters even more for studios working with publishers who have strict IP protection requirements.

Deterministic consistency. The same text, same voice, same settings should produce the same audio. Not “similar” — the same. Game audio needs to be reproducible.

Mastering for game engines. Audio that’s already normalised, properly formatted, and ready to drop into Unity or Unreal. Not raw output that needs separate processing.

Building for Game Dev First

When I designed the voice library for Vois, I started with game archetypes. Not as a marketing angle — because that’s literally what was missing from every TTS tool I tested.

The library has 63 voices organised across 15 character categories:

• Heroes (4 voices) — Confident protagonists. Determined, warm, occasionally vulnerable.
• Villains (5 voices) — Menacing to sophisticated. The kind of range that gives players chills.
• Companions (4 voices) — Supportive, witty, loyal. The characters players grow attached to.
• NPCs (5 voices) — Shopkeepers, innkeepers, villagers. The population of a world.
• Creatures (5 voices) — Otherworldly. Growling, ethereal, alien. Things that aren’t quite human.
• Machines (4 voices) — Robotic to near-human. AI assistants, drones, ship computers.
• Narrators (4 voices) — Authoritative, epic, intimate. The voice that frames the story.
• Announcers (4 voices) — Arena, sports, broadcast. High-energy, punchy.

Plus hosts, executives, educators, storytellers, broadcasters, mentors, and youth voices. Each designed voice went through a specific engineering process to ensure it has consistent character — not just a different pitch applied to the same base.

Beyond presets, voice cloning lets you create entirely custom characters. Record 15 seconds of a reference voice (with permission) and generate new dialogue in that voice. For studios that want a specific sound that doesn’t exist in the library, this is the path.

The Math That Changes

Let’s revisit the economics with local AI voice generation:

Traditional voice acting for a mid-sized indie RPG:

ItemCost15 character types × 4-hour sessions × $150/hr avg$9,000Director time (30 hours × $75/hr)$2,250Studio rental (5 days × $400/day)$2,000Editing and mastering (40 hours × $60/hr)$2,400Pick-up sessions for rewrites (2 days)$2,400Total~$18,000

Cloud AI voice generation:

ItemCost~5 million characters (with iteration)$150–750Mastering (separate tool/service)$200–500Total$350–1,250

Cheaper, but constrained by per-character costs and cloud dependency.

Local AI voice generation (Vois):

ItemCostAnnual subscription$108/yrUnlimited characters, unlimited iteration$0Built-in mastering$0Total$108

The difference isn’t just cost — it’s workflow. At $108 for unlimited everything, you stop thinking about voice production as a budget line item and start thinking about it as a creative tool. Want to voice every NPC in your game? Go ahead. Want to try 20 different readings of the villain’s monologue? No consequence. Want to re-voice an entire quest line because the script changed? It costs nothing.

The Counter-Argument: “AI Voices Aren’t as Good as Real Actors”

Fair point. The best human voice actors bring something that AI can’t replicate: genuine emotional range, improvisational character choices, and the ability to take direction in real-time in ways that surprise the director.

For AAA games with large budgets, human voice actors will remain the standard for lead characters. Nobody’s replacing the performance behind Ellie in The Last of Us with an AI voice. That level of emotional nuance and direction requires a human.

But here’s the thing: the choice for 90% of indie studios isn’t “AI voices vs. human actors.” It’s “AI voices vs. no voices at all.”

Text-only dialogue in a game that would benefit from voice isn’t a creative choice — it’s a financial constraint. If AI voices can bring those silent NPCs to life, even imperfectly, that’s a net gain for players and developers both.

And the quality gap is narrowing. Fast. What sounded obviously synthetic a year ago now passes casual listening tests. For background NPCs, barks, system announcements, and ambient dialogue, AI voices are already indistinguishable from professional recordings in context.

What’s Coming

I think voiced indie games are about to go from rare exception to default. The combination of high-quality local TTS, game-specific voice archetypes, and unlimited iteration at flat cost removes the barrier that’s kept voice out of indie budgets.

Within two years, I expect the conversation to shift from “can we afford voice acting?” to “which voice approach best serves our creative vision?” — with AI, human actors, and hybrid workflows all being valid choices based on the specific needs of the game, not the size of the budget.

Vois launches on Product Hunt on March 3. The free tier gives you access to all 63 voices and all three engines — 10 generations per day, no credit card. If you’re building a game and want to experiment with AI dialogue, that’s more than enough to test.

For indie game devs: what’s your current approach to character voices? Text-only? Placeholder recordings? Have you experimented with AI voice tools? I’m genuinely curious what’s working and what’s not.

I’m Praney. I’m building Vois — a desktop voice AI studio with 63 character voices designed for game development. Everything local, unlimited iteration.

vois.so