---
title: "Voice Input"
id: "1822"
type: "page"
slug: "speech"
published_at: "2026-09-20T22:09:00+00:00"
modified_at: "2026-09-21T01:36:29+00:00"
url: "https://xedant.com/agents/telegram/docs/speech"
markdown_url: "https://xedant.com/agents/telegram/docs/speech.md"
excerpt: "Voice input is dictating tasks straight into the agent chat. Press the mic button in…"
---

# Voice Input

[https://xedant.com/agents/telegram/docs/speech.md](https://xedant.com/agents/telegram/docs/speech.md)

**Voice input is dictating tasks straight into the agent chat.** Press the mic button in the input box, say the task — the recognized speech appears as text in the box and goes to the agent as an ordinary message. Typing a long task is optional.

## What it is and why

Tasks for the agent are convenient to dictate rather than type: “look at what came out in the sources this week and suggest three posts” — say it out loud and the text is already in the input box. It helps most on a phone or tablet, where typing takes long, and whenever a thought needs to be caught quickly.

## How to enable it

Nothing to configure: the mic button is already in the message input box. It appears when speech recognition is enabled in the settings of your Xedant Agent — Telegram Agent uses its capabilities and has no speech settings of its own. If there is no mic button, enable speech recognition in the agent’s settings. Voice input is part of the agent chat, so it requires a configured and working Xedant Agent.

## How it works

Press the mic button (“Start voice recording”) — the button starts pulsing and the recording begins. Say the task and press the button again (“Stop recording”): the recognized text appears in the box right where you were typing — it can be fixed up before sending. No need to stop after every phrase.

While you speak, the preliminary text shows right above the input box, and finished phrases insert into the box immediately. It works in any agent chat — both in the chat panel and in the prompts queue rows: a task can be dictated and queued without typing.

## Recognition methods

The recognition method is chosen in the agent’s settings, and Telegram Agent adapts to it by itself. Two families work:

- **In the browser** — the browser’s built-in speech recognition. Works free and without keys, in Chrome, Edge and Safari; the audio never leaves your browser, but internet access is required;
- **On the agent’s side** — a local Whisper model on your agent server or cloud recognition services. Telegram Agent records the voice from the microphone, passes the recording to the agent, and the agent transcribes it with its own engine — so recognition can be kept entirely on your side, without external services.

## What can go wrong

Problems are reported with clear hints:

- the first run of recognition on your server takes time: the engine installs ffmpeg and downloads the model, and the mic button shows the “installing ffmpeg” and “loading model” hints. This is one-time preparation — afterwards recognition starts instantly;
- the browser does not support built-in speech recognition — then pick recognition on the agent’s side in the agent’s settings;
- no microphone access — allow access in the browser settings; if the microphone is not detected, check the device settings;
- the recording came out empty or too quiet — the hint “Check your microphone or speak louder” appears.

## Where it is available

The mic button sits in the input box of the chat panel and in every row of the prompts queue. The recognized text goes to the agent as an ordinary message — there is no separate “voice” channel, and it can only be sent with a valid license (like any other message into the chat). More about the chat is in the [Agent](/agents/telegram/docs/agent)
 section, and about the task queue in the same place, in “Prompts and auto-send”.

[← Back to the documentation index](/agents/telegram/docs)