But it works in Internet Explorer!

What’s going wrong? If there are issues with Windows, I urge you to refund my subscription costs. There are plenty of excellent AI tools that can translate to a high standard. I no longer have time to fiddle around with this app.

Please excuse these remarks, but I’m asking for a refund of the costs and that will be that.

Thank you

🎤 Upcoming: Deep Dive into the DeepL Voice API

Our next DevTalk is coming up on May 12th, 5–6pm Berlin time with Christian Meißner Technical Lead for the DeepL Voice API.

We’ll focus on real-world implementation: how to use the Voice API in projects, real-time translation workflows, architecture decisions, SDK usage, and practical questions.

👉 If you’re building with the Voice API, exploring what’s possible, or just curious, feel free to drop questions in advance.

🛠️ Made with DeepL: VoiceBuddy

Tim shared VoiceBuddy, an open-source Windows app that provides real-time translated captions for any audio playing on your system: YouTube, Twitch, Discord, game voice chat, Zoom, podcasts, and more.

It’s a great example of what can be built with the newly released DeepL Voice API.

👉 Check it out, test it, and share feedback or questions in the thread.

💻 DeepL API Prompts: feedback wanted

Tim also shared a new GitHub repository with sample prompts, schemas, and examples for using the DeepL API with coding agents.

Since this approach is still fairly new, feedback would be very helpful, especially from anyone already using AI-assisted development tools.

💬 Ongoing API & product discussions

There’s still ongoing discussion around the Style Rules endpoint, where the API team is gathering feedback on endpoint structure, language-specific handling, formatting preferences, and developer workflows.

👀 Looking ahead

• DevTalk: Deep Dive into the DeepL Voice API → May 12

• More builder-focused discussions and project showcases coming soon

• Keep sharing what you’re building, testing, or experimenting with!

Particularly curious about if you'd consider this approach actually helpful or not 🙂

https://github.com/DeepLcom/deepl-api-prompts

🎤 Upcoming: Deep Dive into the DeepL Voice API (AMA)

Our next DevTalk is coming up! Now scheduled for May 12th, 5–6pm (Berlin time) with Christian Meißner (Technical Lead, Voice API).

We’ll focus on real-world implementation:

• How to actually use the Voice API in projects

• Architecture decisions and trade-offs

• Real-time translation workflows

👉 If you haven’t yet, feel free to drop questions in advance, this helps make the session more relevant for everyone.

🛠️ Recent sessions: building and showcasing

Over the past weeks, we’ve hosted hands-on formats:

• “Let’s Build! – DeepL Voice”
  → live prototyping session exploring what can be built with the Voice API in a relaxed, collaborative format

• Show & Tell – TranslateSheet
  → first community showcase with real project insights (recordings now available)

These sessions are a great starting point if you’re thinking about building something yourself.

💬 Product & API discussions

There’s still ongoing discussion around the Style Rules endpoint, where the API team is actively gathering feedback.

We’ve already seen:

• concrete suggestions on API structure and endpoint design

• discussion around language-specific handling and developer workflows

👉 If you’re working with the API, your input here can directly shape what gets built.

🐛 Community feedback & issues

We’ve also seen a few smaller issues and usability topics come up (e.g. session handling / UI quirks).

These are being picked up and forwarded internally. Keep them coming, they’re valuable signals for the team.

👀 Looking ahead

• DevTalk (Voice API AMA) → May 12

• More hands-on and builder-focused formats coming up! Stay tuned.

After I click "Weiter" it just logs me in again. This usually happens when I use DeepL on another PC/Browser and then go back.

Then I'm logged in again.

Tim Cadenbach told me to post it here - so Imma just mention you here.

Specs: Browser is usually Brave, OS doesn't matter (Win11 and macOS)

Another note: While creating this post and selecting tags, the select menu clips out 😅 (this is at the bottom of my screen)

This Thursday, April 23, 5-6pm (Berlin time) May 12th 2026, 5-6pm join us for a live AMA with Christian Meißner, the Technical Lead behind the DeepL Voice API.

(call was moved to a new date unfortunately)

This session is focused on real-world implementation:

• How to actually use the Voice API in your projects

• Architecture decisions and trade-offs

• Real-time translation workflows

• Lessons learned from building with the API

Whether you’re already building with the API or just exploring what’s possible, this is your chance to go straight to the source.

We’ll keep it highly technical, bring your questions, your ideas, or even your code. Feel free to ask any questions in advance in this thread!

🗓 Date: May 12th 2026 5-6pm
🎤 Format: Live AMA
👨‍💻 Audience: Developers & technical builders

Ask anything. Bring code.

>> Event Link

To the best of my knowledge, any service that utilizes the Transformer architecture to probabilistically output sequences generally falls under the overarching umbrella of Generative AI. While AI translation is unique in its specific application of Transformer and seq2seq models, I have inferred that DeepL's specialized approach might be characterized by:

• Training on meticulously curated, high-quality datasets.

• A more deterministic approach maintained through rigorous hyperparameter control.

I am very interested to know if there are additional architectural distinctions beyond these points. For example, I would be fascinated to learn if there are unique features, such as the integration of classical linguistic models into the framework.

I also have two more specific questions.

• First, regarding DeepL Write: is it built on the same underlying model as the translation engine, differentiated primarily through hyperparameter tuning (e.g., temperature or beam search constraints), or does it employ a distinct architecture altogether?

• Second, would you characterize DeepL's outputs as more deterministic compared to other LLM-based services? If so, is this a deliberate design philosophy, or more of a natural consequence of optimizing for translation-specific tasks?

If there are any technical definitions or papers that could be shared within the bounds of company confidentiality, I would be eager to study them to deepen my understanding. Thank you for your time and for sharing your expertise.