Have you heard? The NPCs have more voice options now! It's hard to miss when Elysia's British accent seems to hit hard with many players. ❤️🔥
Curious why you can choose between Legacy and Advanced text-to-speech? Read on to learn how our team yandere'd their way into this one—the emotional journey's end result is characters that are more dynamic, emotive, and dare we say alive?
Why Voice Options?
Originally, we used Azure's Speech Services for our demo builds — it was stable and fast to set up for a demo project. But Azure came with some limitations. We wanted our girls to sound consistent across languages, yet few of Azure's voices could handle that well. Plus, Azure's TTS costs always ate up a big chunk of our budget. 💸 We're always on the hunt for ways to reduce cost without sacrificing quality.
At the same time, other TTS services like ElevenLabs and MiniMax came out, offering more advanced options, including different emotional tones. After many months of testing, we chose MiniMax, which supports multiple languages, delivers more expressive voices, and is significantly more cost-friendly.
We first rolled it out in the Steam China region, then in Russian and German, and finally implemented it across all languages for the global version this September.
Pros & Cons
MiniMax isn't the only service we tested. Our team compared multiple providers on price, stability, scalability, and technical support. Here's how they stack up for our case:
Azure: Pricey
ElevenLabs: Widely used and good quality, but the most expensive.
OpenAI: Also pricey, with no available technical support.
Inworld: ❌ Not fitting for our use case.
MiniMax: Cost-effective, with voice output that can reflect different emotions, plus excellent technical support from their customer service team.
Here are some samples comparing familiar voices with our test ones. We use one of Eddie's iconic CUTscene lines—let us know which of these voices truly speak to you!
View game page
We're also careful not to rely on smaller or short-term projects. Voice consistency matters — you don’t want to wake up one day and hear Eddie with a completely different voice, never able to change back. That's why we still keep both Legacy and Enhanced voice models in our game, so players who prefer the older voices can stick with them.
Scalability is another key factor. We currently support six languages, including English, Chinese, and Japanese, and we want to expand further. We're surveying our global community to decide which language to add next, so it's important to have a service that can reliably cover most of our needs.
Technical Challenges
Switching pipelines is never easy:
One major challenge was picking tones that sound natural across all languages, then testing them through various game scenarios. The tone needs to match the original, since players are already familiar with how each girl should sound — it's part of their identity.
We also changed our audio format — a technical adjustment that doesn't impact gameplay — moving from .wav (our early demo's quick-prototyping choice) to .mp3, which is lighter in size. This was partly because the .wav files from MiniMax didn't fully fit our previous project's structure.
These details may stay behind the curtain, but they're what make sure you hear the right voice, with the right emotion, at exactly the right moment. 🎧
It Never Ends
We're still experimenting — balancing cost, stability, and immersion. The goal stays the same: voices that feel alive, connect with our characters, and make your time with them unforgettable. ✨
Based on the AI2U community's vocal feedback, fans are split between Eddie's original sweet voice and her new "anime dub" one. Meanwhile, Elysia's new British voice has enchanted many longtime and returning players, while Estelle's Legacy voice is preferred by her dedicated engineers.
Are you a fan of the Enhanced voices or a Legacy TTS enjoyer? We'd love to know if you like the new options or if you're committed to the NPCs' original voices (Eddie approves either way 💖🔪)