【これ便利】OpenAIの『音声 AI 3 モデル』が 登場！GPT-Realtime-2 / Translate / Whisper を徹底解説

In this video, I explained the three new voice AI siblings announced by OpenAI: GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper. GPT-Realtime-2 boasts GPT-5 class inference, expandable context to 128K, and supports parallel tool calls and tiered reasoning effort settings, making the experience of having the AI ​​think and call tools while continuing the conversation a realistic possibility. I regularly use Aqua Voice for voice input, and I'd be incredibly happy if voice-to-voice, voice-to-action, and systems-to-voice automation experiences became more widespread. This latest advancement feels like a significant step in that direction. The cost is roughly 14 yen per minute, or about 850 yen per hour. For transcription alone, a local Whisper v3 Turbo might suffice in some cases, so it's best to choose based on your needs. What use cases would you like to see this real-time voice AI in? Please let us know in the comments! Timestamp: 00:00 Opening - The 3 Voice AI Brothers Appear 00:27 The Significance of GPT-5 Grade Inference in Voice 02:41 Division of Roles in the 3 Brothers System 03:11 Trying the Official GPT-Realtime-2 Demo 03:58 3 Voice AI Patterns (Voice-to-X) 05:12 Enhancements in Realtime-2 and 128K 06:37 Real-Time Simultaneous Translation Demo 07:17 Whisper Transcription and 10 Voice Types 07:42 Realtime API Connection Method and Pricing: 14 Yen per Minute 08:56 Differentiation from ElevenLabs / Gemini Live 09:19 Experiencing the Operation of an AI Agent Based on Voice 10:33 Summary and Membership Announcement --- Thank you for watching! Welcome to [Masao AI In-Depth Explanation Channel]! I, Masao, an engineer involved in AI service development, will thoroughly explain "truly useful ways to use and develop AI" while incorporating actual development work and know-how. ▼ Subscribe to the channel here ▼ https://www.youtube.com/@ai_masaou?su... ✅ AI-Driven Development Research Department (Note Community) ▼ Membership here ▼ https://note.com/masa_wunder/membership ----- 🎁 Free Gift from Official LINE 🎁 We are giving away source code and other materials introduced in videos and articles for "free" to those who register on our official LINE account! By registering, you can get sample code and usage tips that even those with no development experience can try immediately! You'll receive a download site and password immediately! ✨ ▼ Register here https://lin.ee/cJtJ8Y3 ----- ■ Official Website https://masao-ai.web.app/ ■ X (Former Twitter) https://x.com/ai_masaou ■ note (Blog) https://note.com/masa_wunder ■ zenn (Technical Blog) https://zenn.dev/aimasaou ■ Business Inquiries/Contact: Please contact [email protected] ----- ■ Masao's Profile ・14 years of development experience (8th year as a self-employed individual) ・Education: Former Imperial University, research on pixel interpolation ・Developed a service used by over 30 million people and an app with over 4 million downloads through personal development ・Also runs an SNS marketing support business and a web media business ・Specialties: Frontend and mobile apps, image processing ----- ■ Materials Used VOICEVOX: Shikoku Metan, Aoyama Ryusei BGMER: https://bgmer.net/ Sound Effect Lab: https://soundeffect-lab.info/