For decades, players have customized how they look in virtual worlds through avatars, character creators, cosmetics, and skins. The next frontier may be how they sound.

Advances in AI-powered voice technology are rapidly transforming voice from a simple communication tool into a new layer of online identity. Whether it's roleplaying in multiplayer games, building a streaming persona, improving accessibility, or simply expressing individuality, real-time voice transformation is becoming an increasingly important part of how people present themselves in digital spaces.

At the center of that shift are companies like Voicemod that have evolved from traditional voice-changing software into platforms focused on AI-powered voice experiences running directly on consumer hardware. In this interview, Voicemod CEO Jaime Bosch discusses the technical foundations behind modern AI voice transformation, the growing importance of local AI processing, and how partnerships with companies such as Qualcomm are helping transform sophisticated voice models.

Voicemod is often described as a real-time AI voice platform. From a technical perspective, how are these voice transformations actually achieved? What role does AI play versus more traditional audio processing?

Jaime Bosch, Voicemod CEO: Our journey started with digital signal processing (DSP), which is very effective for applying filters and modifying an existing voice in real time. AI takes that much further by allowing us to change the timbre, the underlying character or DNA of a voice, rather than simply layering effects on top of it. That gives users much more control over their audio identity. DSP still plays an important role in our sound design layer because it is lightweight, reliable, and well-suited for many real-time use cases.

You’ve emphasized running voice transformation directly on NPUs and local hardware. For developers who may not be familiar, what does “on-device AI” actually mean in practice, and how is it different from cloud-based voice processing?

Jaime Bosch: On-device AI means all inference is performed locally on the user’s machine rather than sent to the cloud. In practice, that removes round-trip latency, avoids dependency on connectivity, and keeps voice data private. For real-time systems like voice, those differences are critical because even small delays or performance spikes break the experience.

Real-time voice transformation introduces strict constraints around latency and performance. How does the system maintain low latency while still delivering expressive, high-quality voice changes?

Jaime Bosch: We design the entire pipeline around real-time constraints, from model architecture to how audio is buffered and processed. Models are optimized to run on very small audio frames with strict timing budgets. The result is that transformations feel immediate while still preserving clarity and expressiveness.

With partnerships like Qualcomm, how are you leveraging NPUs specifically? What kinds of workloads are being offloaded there, and how does that change what’s possible compared to CPU or GPU processing?

Jaime Bosch: NPUs are particularly well-suited for the inference workloads behind AI voice transformation. By running those models on the NPU, we can keep the CPU and GPU focused on gameplay and streaming while maintaining consistent performance. That allows us to run more demanding AI voice models without compromise, which is especially important in gaming environments where latency and system overhead matter. Over time, that makes high-quality AI voice experiences more practical on mainstream consumer devices.

With the recent integration into Elgato’s Wave Link as native VST effects, Voicemod is moving closer to the core audio pipeline rather than sitting as a separate app. What does that shift mean from a technical and user experience standpoint?

Jaime Bosch: Moving closer to the native audio pipeline is important because it makes voice transformation feel much more natural inside the tools creators already use. From a technical standpoint, it simplifies routing and makes the experience more direct. From a user perspective, it means creators can access voice effects as part of their existing workflow, instead of treating voice as something separate they have to manage on the side.

Historically, tools like Voicemod required audio routing between multiple applications, but now effects can run natively in existing pipelines. How important is reducing that friction for broader adoption?

Jaime Bosch: Reducing friction is one of the most important drivers of broader adoption, because people want these experiences to feel native to where they already communicate. For streamers, that means less setup inside their production workflow. More broadly, it points toward a future where voice transformation can exist directly inside in-game voice chat and wherever online communication happens. That is a big part of why we have built our SDK, so voice can live inside the platforms and environments where people already speak.

For game developers, what would it take to integrate real-time voice transformation directly into games or engines? Are we moving toward voice as a native gameplay system?

Jaime Bosch: We are now at the point where Voicemod is being integrated directly into games through our SDK, which is an important shift. That means developers can start treating voice transformation as part of the native player experience rather than an external layer. Over time, I do think voice will become a more natural part of gameplay systems, especially in social and multiplayer environments where communication is already central to how players interact.

Have you seen emergent use cases where players are using voice transformation in ways you didn’t initially anticipate?

Jaime Bosch: One of the most interesting trends is how players use voice to shape identity and social dynamics, not just for entertainment. We see people using it to feel more confident, to roleplay more deeply, or to participate in communities where they might otherwise hesitate. We’ve even had one user tell us they used Voicemod to rebuild their voice after a serious health event, which speaks to how personal this layer can be. It has become a tool for presence and participation, not just effects.

AI voice technology inevitably raises concerns around privacy, impersonation, and misuse. How does Voicemod approach these challenges, especially when processing happens locally on-device?

Jaime Bosch: We approach this from both a technical and ethical standpoint. Running processing locally reduces exposure of sensitive voice data, which is an important baseline for privacy. Just as important, all of our AI models are built using data with permission, and Voicemod is proud to have earned Fairly Trained certification for its AI speech and singing models. We believe creative voice technology should expand expression while respecting the rights of the people whose voices make that technology possible.

Does on-device processing change the data privacy model compared to cloud-based AI systems? What data, if any, leaves the user’s machine?

Jaime Bosch: On-device processing does change the privacy model because, in our case, real-time voice transformation does not require sending live audio to the cloud for inference. That gives users more control and reduces the risks that come with centralized data handling. At the same time, I would not say on-device is automatically better in every situation. Both on-device and cloud-based systems can be appropriate if they are designed responsibly and with real respect for user privacy.

 At Voicemod, that means being very clear about what the product does and what it does not. The app accesses the microphone to apply the voice effects the user selects, and we do not listen to users’ conversations. We also have a service improvement program that is strictly opt-in. In those cases, only very short clips are shared, never full conversations, and that audio is not used to train our models. For us, the principle is simple: users should understand what is happening, stay in control, and trust that their data is being handled responsibly.

As this space evolves, what responsibilities do platforms like Voicemod have in shaping ethical standards for voice AI in gaming?

Jaime Bosch: Companies like Voicemod carry a real responsibility — not just toward our users, but toward everyone in the AI value chain: the artists, the developers, the broader ecosystem. That responsibility becomes especially acute with cutting-edge technology, where legislation inevitably arrives later than the technology itself. When companies act thoughtfully from the start — protecting users, respecting artists, designing for trust — they are not just complying with future rules. They are shaping what those rules will look like, and setting the standard that users and regulators will eventually recognize as their baseline expectation.

At Voicemod, that philosophy started with a deliberate choice when we began training our voice AI models: instead of scraping the internet for data the way so many others have done, we hired voice talent and sourced licensed data. That decision made us the first voice AI model of our kind to obtain the Fairly Trained certification — a recognition that our training data meets the highest ethical standards for consent and compensation.

But the standard doesn't stop at how the model was built. It runs through everything we do: how we design our tools, how we approach privacy by design, how we think about the experience of every user. The goal is simple — people should be able to supercharge their voice, build that layer of their identity in online spaces, and have fun doing it, without ever having to wonder what we're doing with their actual biological voice. That trust is not a feature. It is the foundation.

Looking forward, how do you see AI voice evolving over the next 3–5 years, especially as on-device AI hardware becomes more powerful?

Jaime Bosch: Voice will become a standard layer of digital identity across games and interactive platforms. As hardware improves, transformations will feel more natural, more personalized, and always available in real time. Over time, controlling how you sound will become as standard as choosing an avatar or a skin.

Voicemod CEO Jaime Bosch