Aydın Tiryaki and Google Gemini

Tech giants have been marketing the vision from famous sci-fi movies for years: “The keyboard era is over, now you will handle everything just by talking to the air.” However, when it comes to practice, especially building architectures that require fine-tuning and surgical precision like the “Gem Factory”, systems condemn us back to those physical keys they want us to leave behind. If a keyboardless artificial intelligence vision is being offered, it is imperative that this job is done properly, not half-baked.

Four Different Devices, Four Different Experiences

To understand the quality of speech-to-text (STT) technologies, it is necessary to test the issue directly on a hardware and platform basis. Trials conducted in four different environments reveal the massive gap between systems:

• Samsung S25 Ultra: The power of the mobile ecosystem. STT performance on this Android-based phone is near perfect. The device’s deep audio processing algorithms and neural processing units (NPU) offer the infrastructure that best captures context and Turkish phonetics.
• Samsung Tab S9: Although it falls a step behind the S25 Ultra, it exhibits a not-bad, satisfactory performance because it still benefits from the optimization of the same Android ecosystem.
• i9 Processor Windows 11 Desktop: Despite being a hardware monster, Google’s STT running over Chrome is practically a disaster here. It swallows words at the beginning of sentences, forcing you to enter the same words over and over again to make sure they are understood.
• i7 Processor Windows 10 Desktop: Although the voice input infrastructure on this veteran device is different from Windows 11, its comprehension ability is again quite weak and old-generation.

The Problem is Not the Human Voice, But the Listening Engine!

So, is the reason for this “lack of comprehension” and swallowing of words on the desktop the acoustic difficulties of the human voice? Absolutely not. When voice input is made to the Claude application on the same S25 Ultra device, with the same microphone and the same voice, a perfect and smooth text emerges without missing a single thing. Similarly, ChatGPT works orders of magnitude better than Google’s STT on all devices.

This flawless “A/B test” proves that the problem is not in the voice, but in the architecture of the listening engine. Google’s STT infrastructure takes its foundations from years of search engine habits (short and fast queries). However, Claude (Anthropic) and ChatGPT (OpenAI) built their voice recognition systems from scratch, directly to feed large language models (LLMs) and grasp those long contexts without breaking them. Google must drop this stubbornness, completely trash the old-generation STT engine, and build a new, context-aware STT.

Claude’s Achilles Heel: The Windows and Turkish Conundrum

Claude, which works wonders in perceiving voice commands, harbors serious internal handicaps when looking at other dimensions of the business:

• The Abandonment of the Desktop Ecosystem: Claude does not have its own voice input (STT) support working in the Windows environment. To cover this gap, you have to use Windows’ built-in dictation tool (Win + H), but this tool is at least as clunky and bad as Google’s desktop STT.
• Lack of Turkish Nuance: Despite its smooth context ability, Claude treats Turkish almost as a second-class language (low-resource language). Although it is very capable in text construction, it sometimes takes on a very mechanical, translation-like tone in the subtle cultural nuances of Turkish, wordplays (in concepts like tertip, tefriş), and the proper use of idioms.

Conclusion

While companies shout “keyboardless future,” pushing users to struggle either with Google’s stubbornly unchanged clunky voice engine or Claude’s deficiencies that ignore the desktop world and the intricacies of Turkish is a massive disappointment. For a truly seamless experience, the audio listening “ears” need to catch up at the same speed to that massive level reached by LLM (Large Language Model) intelligence. Until then, our keyboard will continue to be our safest harbor for those precise commands.

| aydintiryaki.org | YouTube | Aydın Tiryaki’nin Yazıları ve Videoları │Articles and Videos by Aydın Tiryaki | Bilgi Merkezi│Knowledge Hub  | ░ Virgülüne Dokunmadan │ Verbatim ░ | ░ Bir Kelimenin Peşinden Yapay Zekanın Derinliklerine: “Tebdil-i Mekan”dan “Zırvalama”ya Uzanan Bir Söyleşinin Anatomisi │From a Single Word to the Depths of AI:The Anatomy of a Dialogue Spanning from “Tebdil-i Mekan” to “Spouting Nonsense” ░ 02.07.2026

Credits: The subject, scope, and editorial framework of this article series were determined by Aydın Tiryaki. Gemini (Google, Advanced / Pro mode) assisted during the initial 35-stage interactive dialogue that evolved from the concept of “tebdili mekan”; while NotebookLM assisted in analyzing this dynamic conversation, expanding it into comprehensive articles, and executing the bilingual writing and translation process.