> ## Documentation Index
> Fetch the complete documentation index at: https://docs.corti.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Audio archive
> Persist the dictation microphone audio you stream to Corti, and finalize it into a downloadable recording
**Follow along with the runnable example.** This guide walks through the [Audio archive example](https://github.com/corticph/corti-examples/tree/main/sdk/typescript/applets/src/applets/dictation-audio-archive) in `corti-examples` — bootstrap your build from it or use it as reference.
Some workflows need to keep the audio a user dictated — for playback, quality review, or later re-transcription. This guide shows where to capture that audio and how to persist it *while* streaming to Corti.
It builds on the raw-SDK path from the [Microphone](/stt/guides/dictation-microphone) guide, where your app owns `MediaRecorder` directly — that's the point at which each audio chunk is available to both send and keep.
***
## Tee each chunk
The only change from a plain raw-SDK session is what you do in `ondataavailable`: append the chunk to a local archive **and** send it to Corti. Both consume the same blob.
```ts title="JavaScript" theme={null}
recorder.ondataavailable = async (e) => {
if (e.data.size === 0) return;
await appendChunk(e.data); // persist locally
// pass e.data.arrayBuffer() to socket.sendAudio() to stream to Corti
};
```
***
## Session lifecycle
Model recording as a session with three controls, so a user can pause and resume without dropping the connection:
| Control | Effect |
| :-------------- | :-------------------------------------------------------------------- |
| **Record** | Open a new session (or resume a paused one) and start capturing |
| **Stop** | Flush buffered audio, then pause the recorder — the socket stays open |
| **End session** | Flush, finalize the archive, and close the socket |
The key detail is **flushing before you pause or end**: calling `socket.sendFlush()` tells the server to return transcripts for audio it has already received, so nothing is lost at the boundary.
```ts title="JavaScript" theme={null}
// Stop: emit a final chunk, then pause (keep the socket open).
async function stop() {
recorder.requestData(); // force a last ondataavailable
recorder.pause();
}
```
Call `socket.sendFlush()` after pausing — the server returns any buffered transcripts and then a `flushed` message signalling completion. See the [audio archive example](https://github.com/corticph/corti-examples/tree/main/sdk/typescript/applets/src/applets/dictation-audio-archive) in `corti-examples` for the full socket flush implementation.
***
## Finalize the archive
On **End session**, assemble the collected chunks into a single `Blob` using the recorder's actual MIME type, then build a playback/download URL.
```ts title="JavaScript" theme={null}
const mimeType = recorder.mimeType || "audio/webm"; // what MediaRecorder actually produced
const blob = new Blob(chunks, { type: mimeType });
const url = URL.createObjectURL(blob); //