Building with ChatGPT Voice | OpenAI

TL;DR: A demonstration of using ChatGPT Voice on desktop to manage tasks—from drafting blog posts and suggesting features to debugging code and coordinating launch preparations—all through natural speech. ## Voice Interaction for Desktop Workflows The video opens with a user speaking commands to ChatGPT: “嘿Chat，投到投影仪上。播我的daylist。” (Hey ChatGPT, cast to the projector. Play my daylist.) After a brief pause, the user shifts to a work task: preparing launch materials for the upcoming voice feature on desktop. The assistant immediately begins drafting a blog post. > “当然。先来个简单初稿。在桌面端应用里认识一下语音功能。语音让工作推进更自然。你觉得怎么样？” (Sure. Here’s a simple first draft: Meet voice on the desktop app. Voice keeps your workflow smooth. What do you think?) The user approves: “使用电脑时保持流畅体验。就这么定了。” (A smooth experience while using the computer. That’s settled.) ## Collaborative Feature Development The user then proposes a feature idea for last‑minute polish: “我有个功能想法，可以在最后关头做些打磨。我会先随便聊聊，然后希望你能问我一些问题，反驳我的观点。” (I have a feature idea for some last‑minute polish. I’ll start by chatting casually, and then I want you to ask me questions, challenge my points.) The assistant highlights a promising technical observation: “看起来仅靠着色器内部就能取得很大进展，因为它已经有音频和相位输入了。Codex，你能去检查一下功能开关，确保配置正确吗？” (It seems like a lot can be done inside the shader alone, since it already has audio and phase inputs. Codex, can you check the feature flag and make sure it’s configured correctly?) ## Handling Bug Reports and Root‑Cause Analysis During the conversation, a new bug report appears in the feedback channel. The user instructs: “等等，反馈频道里还有一个新的bug报告。你能开一个新线程，提交一个拉取请求，查找根本原因吗？” (Wait, there’s a new bug report in the feedback channel. Can you open a new thread, submit a pull request, and look for the root cause?) The assistant quickly locates the issue: “让我看看。找到了。是那个功能开关。完全配置错误。常有的事。” (Let me check. Found it. It’s that feature flag. Completely misconfigured. Happens all the time.) ## Launch Readiness and Final Coordination The user asks: “启动所需的一切都已准备就绪，对吧？” (Everything we need for launch is ready, right?) The assistant confirms: “是的，我觉得可以发布了。” (Yes, I think it’s good to release.) The user then outlines a follow‑up process: “明白了吧，只需设定一个目标，然后标记团队稍后审核。我会用手机和你确认。收到。我会为此创建一个后续目标。然后，嗯，别出错。” (Got it, just set a goal and tag the team for review later. I’ll confirm with you on my phone. Received. I’ll create a follow‑up goal for this. And, um, no mistakes.) The assistant acknowledges: “收到。我会为此创建一个后续目标。然后，嗯，别出错。” (Received. I’ll create a follow‑up goal for this. And, um, no mistakes.) ## Key Takeaways - **Voice as a productivity tool** – From projecting content to drafting copy and managing bugs, voice commands streamline desktop workflows. - **Iterative collaboration** – The assistant can be asked to question or challenge ideas, acting as a critical thinking partner. - **Real‑time debugging** – The assistant can inspect code, check feature flags, and open pull requests based on voice instructions. - **Launch coordination** – Voice interactions support both high‑level confidence checks and low‑level task creation (e.g., follow‑up goals). --- Source: [Building with ChatGPT Voice | OpenAI](https://www.youtube.com/watch?v=E0ZMOschrTU)