Telegram Desktop is the simplest route. Right-click the voice message, pick Save As and choose a folder; the file is saved with the .ogg extension. If you receive many voice messages, the app also keeps downloaded ones in its Telegram Desktop folder inside Downloads, so they may already be on your disk.
On Android, tap the three dots on the message, or press and hold it, and choose Save to Downloads; the file then appears in the Telegram folder of your downloads or music directory. On iPhone, press and hold the message, tap Select, then the Share icon, and choose Save to Files. In Telegram Web, the message menu has a Download option that places the file in the browser's download folder.
For a whole conversation, Telegram Desktop has an export function in the chat menu: Export chat history, with Voice messages ticked. It writes every voice message from the chat into one folder as separate files, which is far quicker than saving them one at a time.
Telegram can turn voice messages into text inside the app, but the feature belongs to the Premium subscription, with only an occasional free use for everyone else. The text also stays inside the chat bubble: it is awkward to copy out of a long message and it is not something you can file, search later or hand to a colleague.
This page is for the cases around that. You do not have Premium, or you have used the free attempts. The message is fifteen minutes long and you want the text in a document. The audio was posted in a channel as a file rather than a voice note. Or you have a hundred messages from an exported chat and want all of them as text you can search.
Telegram voice messages tend to be longer than those in other messengers, because the app puts no practical ceiling on them. People dictate meeting notes, teachers send explanations to a class group, and channel owners publish spoken updates and whole audio episodes. A message of that length is a real recording, and the transcript comes back with timestamps so you can jump to a point in the original.
Round video messages are saved as small .mp4 files. Only the sound is used, so they are transcribed exactly like a voice note. Audio posted to a channel as a music file, usually MP3 or M4A, is uploaded as it is. The spoken language is detected automatically from the audio, and more than fifty languages are supported.
A voice message is sent to one person or one group, and the transcript should not travel further than that. The speech models run on our own servers, and the audio is not forwarded to an external speech service. That is true for guest uploads as well as for accounts.
As a guest you get a free preview of each transcript to read and copy, and a short message is usually shown in full. A free account keeps the full transcripts in a history and lets you fix a misheard name; what you transcribed as a guest moves into it when you sign up. The free allowance is 5 hours across 10 recordings. Subtitle export, AI chat, summaries and export to PDF or Word are part of the paid plans.