Speech to text without code: audio, video and voice messages

Speech to text in nana800 turns audio, video and voice messages into text from a link to the file: the Speech to text block sends the recording to OpenAI and returns the transcript. All you need is your own OpenAI API key, with no code. A block run costs $0.002, and OpenAI charges the transcription to your OpenAI account.

Build it in the builder
  • Block run: $0.002
  • Free every month: $3.50
  • No card needed

What you can do

  • Transcribe voice and video messages that people send to your Telegram bot
  • Turn recordings of calls, meetings and interviews into text: mp3, m4a, wav, ogg, webm, flac or mp4 video up to 25 MB
  • Give a language hint for better accuracy, or let the service detect the language
  • Give the model a list of names, brands and terms so it spells rare words correctly
  • Get text with timestamps per segment for subtitles (whisper-1 model only)
  • Pass the text further down the scheme: to ChatGPT for a short summary, to Google Sheets or back to Telegram

Blocks in the builder

  • Speech to text

    Action · $0.002 per run

    Downloads audio or video from a link (up to 25 MB), sends it to OpenAI and returns the text and the detected language, plus segments with start and end times in timestamp mode. The block detects the file format itself.

How to set it up

  1. 1

    Create an OpenAI API key

    In your OpenAI account open API keys, click Create new secret key and copy the key (it starts with sk-). Add funds to your OpenAI balance: transcription is paid from there, and with an empty balance the block returns an error.

  2. 2

    Add a trigger that brings the recording

    Create a scheme in the builder and add a trigger. For example, Telegram message from user to bot saves voice messages, audio and video to the nana800 storage and returns links to them. For a first test, the button trigger and any direct link to an audio file are enough.

  3. 3

    Set up the transcription block

    Add the Speech to text block and paste your OpenAI key. In the audio or video link field put the link from the trigger or your own link: it must open without signing in. Keep the gpt-transcribe model and type en or ru in the language field.

  4. 4

    Pass the text on

    Put the text field from the transcription result into the next block: Telegram message send to reply with the transcript, google sheet record to save it to a sheet, or Open AI request to get a short summary.

  5. 5

    Turn the scheme on and test it

    Turn the scheme on, press the button or send a voice message to the bot and open the block result. If there is no text, error_message explains why: a wrong key, a link that does not open, or a file over 25 MB. When it works, keep the scheme on. While it is on, it costs $0.0042 per hour.

Popular connections

  • Telegram + Speech to text

    Build a Telegram bot that turns voice messages into text with OpenAI speech recognition and replies with the transcript. No code, free balance every month.

Frequently asked questions

A block run in nana800 costs $0.002 and is charged only when the text is returned. OpenAI bills the transcription itself to your OpenAI account. An enabled scheme costs $0.0042 per hour. Your nana800 balance is topped up to $3.50 for free every month, no card needed. A scheme kept on around the clock for a month spends most of that on scheme hours.

Audio in mp3, m4a, wav, ogg, webm and flac, and mp4 video, up to 25 MB. The block detects the format from the file itself, so the link does not need an extension. The file must download from a direct public link without a password or sign-in.

Yes. The Telegram message from user to bot trigger saves voice messages, audio and video to the nana800 storage and returns links to them, and you put that link into the transcription block. Telegram gives bots files only up to 20 MB, larger files come without a link.

Yes, the block works only with your own OpenAI API key. You create it under API keys in your OpenAI account, and the balance there must not be empty. You pay OpenAI directly for transcription, and nana800 charges only the block run.

Set the language of the recording, for example en. If the recording has two languages, list the codes with a comma, for example en,de: only the gpt-transcribe model understands several languages. In the prompt field list the names, company names and terms that appear in the recording.

Yes. Choose with_timestamps in the output field and the whisper-1 model: the block returns segments with start and end times in seconds. Other models do not return timestamps. OpenAI will turn off every model except gpt-transcribe on February 26, 2027, and timestamp mode will stop working after that.

Build your first scheme for free

Sign-up is free and needs no card. Every month the balance is topped up to $3.50 for free, enough for about 3,500 runs of basic blocks.

Build it in the builder