> For clean Markdown of any page, append `.md` to the page URL.
> For a complete documentation index, see https://docs.sarvam.ai/llms.txt.
> For full documentation content in one file, see https://docs.sarvam.ai/llms-full.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.sarvam.ai/_mcp/server.

# Transcribing a File

Upload or record audio and get back a speaker-labeled transcript. For an overview of Speech to Text, supported formats, and limits, see the [Speech to Text overview](/creative-speech-to-text).

## Upload and Transcribe

![Speech to Text screen. Header has Feedback and History buttons. The main panel is a dashed-border drop zone with a microphone icon reading Drop audio here, or browse files, with the caption listing supported formats and a 500 MB size limit. Below it, a second option reads or record audio with its own microphone icon. Below that, Try an example offers three colored example chips: Gujarati finance news, Marathi cricket commentary, and Tamil healthcare conversation.](https://files.buildwithfern.com/https://sarvam-api-docs.docs.buildwithfern.com/77448ece879c97deccf65102a90697facca80fef39b1cc712a233106e6d76c29/sarvam-creative/assets/stt-upload.png)

Drop a file, click **browse files**, or click **or record audio** to capture audio directly in the browser.

* Video files work directly - you don't need to extract the audio track yourself first.
* No file of your own handy? Click one of the **Try an example** chips to see a full transcription on a real sample clip.

Once you've provided a file, its name and size are confirmed (e.g. "Demo Video.mp4 · 35.0 MB"), and two dropdowns plus a toggle appear.

!\[Speech to Text screen with a file attached, showing Demo Video.mp4 at 35.0 MB with a remove button. Below it, a Language dropdown set to Auto Detect and a Mode dropdown set to Transcribe, alongside a Diarization toggle labeled Detect and label speakers. A black Transcribe button sits below, with an estimated cost of approximately 7.00 credits and an account balance reading 1,00,090 of 1,00,100 remaining.]\(../../assets/stt 2.png)

* **Language** - leave on **Auto Detect**, or pick the spoken language explicitly. Getting this right matters more for regional Indian languages and dialects than it might seem - an incorrect setting doesn't just mistranslate a few words, it can distort the model's expectations for the entire recording's phonetics.
* **Mode** - defaults to **Transcribe**.
* **Diarization** - toggle this on to detect and label individual speakers in the transcript. For a roundtable or panel recording, accurate diarization is often the single biggest factor in whether the transcript correctly separates who said what.
* The estimated credit cost (e.g. "\~7.00 credits") and your remaining account balance are both shown live before you commit.

Start transcription and wait for processing to complete.

## Reviewing and Exporting

* Once processing completes, the transcript opens in a slide-over **Transcript** panel.
* A playback bar at the bottom shows the source filename, a waveform, and elapsed/total duration.
* Icons beside the playback bar let you copy, download, or share the transcript.

Need this as a live API integration instead of a one-off upload? The same diarization pipeline is available via the Speech-to-Text API.