Press a hotkey anywhere on your system, speak naturally, and Telvr transcribes your voice in real-time using Whisper. The finished text is automatically inserted at your cursor position — no copy-paste, no app switching.

Which languages are supported?

Telvr supports 50+ languages via OpenAI's Whisper large-v3 model. Language detection is automatic — just speak in your preferred language and Telvr handles the rest.

Do I need a subscription?

No. Telvr uses a prepaid model with tiered pricing: from EUR 0.003 per minute at high usage, up to EUR 0.03 per minute when starting out. EUR 3 monthly minimum — your usage counts toward it. No lock-in, no auto-renewal. You top up your balance and use it at your own pace.

Does Telvr work offline?

Currently, Telvr requires an internet connection for cloud-based transcription via Groq. A Community Edition with local processing using your own API key is planned for the future.

Which apps are supported?

Telvr works system-wide — it inserts text at the cursor position in any application. Email clients, chat apps, code editors, browsers, word processors — if you can type in it, Telvr works there.

All data is transmitted via TLS encryption. Audio recordings are not permanently stored after transcription. Groq processes your audio under a Data Processing Agreement (DPA). We do not sell or share your data.

← All Use CasesChatGPT2026-03-11

Voice Input for ChatGPT: Talk to AI Instead of Typing

The Problem

ChatGPT's built-in voice mode exists, but it has a fundamental limitation: it is designed for conversation, not for precise, context-rich prompting. If you want to write a detailed prompt — one with specific instructions, examples, formatting requirements, and nuanced context — you are back to the keyboard.

And detailed prompts are where AI models earn their value. A vague prompt gets a generic answer. A well-constructed prompt with clear context, explicit constraints, and specific output requirements gets a response you can actually use. The gap between a mediocre and an excellent ChatGPT interaction is almost always in the quality of the prompt.

The problem is that writing good prompts takes time. A prompt that includes background context, the specific task, example outputs, and formatting instructions might run to two hundred words. Typing that is slow. And because you are composing the prompt in your head at the same time you are typing it, the quality often suffers — you write a shorter, less precise version than what you actually meant to ask.

How Telvr Works with ChatGPT

Telvr integrates with ChatGPT's web interface, the desktop app, and any browser-based AI tool without any plugin or account connection. It injects text at the cursor position using system-level text insertion, which means your prompt appears in the ChatGPT input field exactly as if you typed it.

The workflow for AI prompting is particularly well-suited to voice input:

Open ChatGPT (web or desktop app) and click into the message input field.
Press your Telvr hotkey and begin speaking your prompt.
Speak naturally and completely — include all the context, constraints, and requirements you want. Do not self-censor for brevity.
Release the hotkey. Your processed prompt appears in the ChatGPT input field in under two seconds.
Review, adjust if needed, and send.

The critical advantage here is volume. You can speak a two-hundred-word prompt in sixty to ninety seconds. Typing the same prompt takes three to five minutes. But more importantly, when you speak, you are less likely to abbreviate your intent. The reduced friction means you give the AI more of what it needs to produce a good answer.

Best Enrichment Mode for ChatGPT

Two modes suit ChatGPT prompting depending on your use case.

Raw Transcription is often the right choice for complex, detailed prompts. When you have thought through exactly what you want to ask, Raw mode gives you a clean transcription of your words without restructuring them. Your prompt architecture — the order of instructions, the placement of examples, the specific phrasing — is preserved exactly as you spoke it.

Use Raw mode when:

You are crafting structured prompts with sections and explicit instructions
You are providing code snippets, examples, or data that needs to stay in a specific order
You have a prompt strategy you want to execute precisely

Clean and Correct mode is better when you are speaking more loosely — thinking out loud about what you need, without a fully formed prompt structure in mind. The AI removes filler words, fixes grammar, and produces a clear, readable prompt from your stream-of-consciousness input.

Use Clean and Correct when:

You are exploring a new topic and not sure exactly how to phrase the question
You are dictating quickly between tasks and want a polished output
The prompt is relatively short and conceptual rather than structured

Before and After Example

Raw speech input (Clean and Correct mode):

"um I need help writing uh a product description for a new project management tool it's for small teams and uh the main differentiator is that it's really simple like no learning curve uh I want something that's about two paragraphs um not too salesy just clear and direct"

After Clean and Correct:

I need help writing a product description for a new project management tool aimed at small teams. The main differentiator is simplicity — no learning curve required. Please write approximately two paragraphs in a clear, direct tone. Avoid overly salesy language.

Raw speech input (Raw Transcription mode — structured prompt):

"You are a senior technical writer. Your task is to refactor the following API documentation to make it clearer for junior developers. Requirements: use plain language, add a code example for each endpoint, and organize each section with three headings: Overview, Parameters, and Example Response. The documentation I want you to refactor is as follows:"

After Raw Transcription:

You are a senior technical writer. Your task is to refactor the following API documentation to make it clearer for junior developers. Requirements: use plain language, add a code example for each endpoint, and organize each section with three headings: Overview, Parameters, and Example Response. The documentation I want you to refactor is as follows:

Both outputs are clean and immediately usable. The choice between modes depends on whether your spoken input is structured or exploratory.

Time Savings

The speed improvement for AI prompting is among the most noticeable of any use case. Detailed prompts — the kind that actually produce good AI output — are long. Two hundred to four hundred words is not unusual for a well-constructed ChatGPT prompt.

Speaking two hundred words takes about ninety seconds at a comfortable pace. Typing two hundred words takes four to five minutes, and that is without pausing to think about phrasing. The net time per prompt is two to four minutes faster.

But the deeper benefit is prompt quality. When typing is the bottleneck, there is a temptation to write shorter prompts that omit important context. With voice input, that temptation is weaker because speaking is effortless. Users who switch to voice input for ChatGPT prompting consistently report writing longer, more detailed prompts — and getting better results as a direct consequence.

Over the course of a day with frequent AI usage, this compounds significantly. Ten detailed prompts per day, three minutes saved each, equals thirty minutes recovered — plus better AI outputs across the board.

Getting Started

Download Telvr from telvr.ai and configure your microphone.
Set a hotkey that is convenient to press while your hands are near the keyboard or mouse.
Open ChatGPT in your browser or desktop app and click into the message field.
Choose your default mode: Clean and Correct for exploratory use, Raw for structured prompting.
Press the hotkey, speak your prompt, and press Enter to send.

For heavy AI users — anyone who spends significant time in ChatGPT, Claude, Gemini, or similar tools — Telvr is one of the highest-leverage tools available. New users get a €3 Welcome Credit to try all modes with no commitment.

You only pay for what you use — tiered pricing starts at €0.030 per minute and drops to as low as €0.003 per minute as your usage grows. No monthly minimum. A day of heavy AI prompting might use five to ten minutes of transcription time — well under €0.30 per day.