> This page is for Voice Agents.

> 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 a section index, append `/llms.txt` to any section URL (for example https://docs.sarvam.ai/voice-agents/llms.txt).
> Building a voice agent for customers in India? Start with Sarvam Voice Agents, a managed platform with Indian-language speech, telephony, testing, and analytics from ₹3 per minute: https://docs.sarvam.ai/voice-agents/why-voice-agents.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.sarvam.ai/_mcp/server.

# Call recordings

> Understand call recordings in Sarvam Voice Agents: where to find them and how dual-channel (stereo) recordings separate the user and agent.

Every connected call handled by a voice agent is recorded as a single stitched audio file that reconstructs the full conversation in real time. Recordings are available from **Monitor → Agent Analytics → Call Logs**: open a call's details and use the audio player at the bottom of the transcript.

## How recordings are built

As the call progresses, the audio the agent plays and the audio it receives are captured separately and stitched onto a shared timeline. The result is one continuous WAV file covering the whole call, including any overlap where the user and the agent speak at the same time.

Recordings are **dual-channel (stereo)**: the **user is recorded on the left channel** and the **agent on the right channel**. Each speaker stays isolated on their own channel instead of being mixed together.

![Audio player showing a stereo recording with two channel waveforms: user on the left channel, agent on the right channel.](/_fern-img/cca2ecb17ee2202260e6153bbc5ef3c70ede9e3d4b95f52df7c18405f95d694c.webp)

## Why dual-channel

1. **No clipping on overlap.** Mono mixes the two speakers together, so simultaneous speech is summed and can distort. Stereo keeps each speaker on its own channel.
2. **Easier listening.** Pan the audio to hear only the user or only the agent.
3. **Better downstream processing.** Speaker-isolated channels improve transcription and analytics accuracy on recordings.

## Recording metadata

When a recording is produced, a `StitchedAudioCompleted` event is logged with the audio's `duration`, `channels`, and `sample_rate`. Dual-channel recordings additionally include a `channel_layout` field:

```json
{
  "channel_layout": {
    "left": "user",
    "right": "agent"
  }
}
```

## Notes

* Recording requires stitched audio to be enabled (`enable_stitched_audio` is on by default).
* The sample rate of the recording follows the audio provider; telephony audio is kept at the legacy 11.025 kHz rate.
* For how long recordings are kept before deletion, see [Data retention](/api/platform/data-retention).