> ## Documentation Index
> Fetch the complete documentation index at: https://developer.suki.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Send START_TIME Before Ambient Audio
> Send one START_TIME frame on /ws/stream before the first PCM chunk for Ambient and Form filling
export const CbRecipeMeta = ({items = []}) =>
;
};
}
sidebar={{
tutorialHref: "/documentation/tutorials/ambient-websocket-code-example",
apiHref: "/documentation/how-to/audio-streaming/websocket-streaming-wire-format-ambient",
apiLabel: "View Wire Format",
authSections: [
{
title: "Partner API",
text: "Authenticate the WebSocket first. Browser clients use Sec-WebSocket-Protocol. See",
href: "/documentation/cookbooks/browser-websocket-auth",
linkLabel: "Authenticate Browser WebSockets",
scopes: ["sdp_suki_token", "ambient_session_id"],
},
],
glance: [
{ label: "Product", value: "Ambient" },
{ label: "Surface", value: "Partner API" },
{ label: "Time", value: "~5 min" },
{ label: "Level", value: "Beginner" },
],
}}
>
## Problem
You open `/ws/stream` and send PCM right away. The stream misbehaves or the session never produces usable content.
Ambient and Form filling require one `START_TIME` message per stream segment before any `AUDIO` frames. Dictation on `/ws/transcribe` does not use `START_TIME`.
## Architecture
```mermaid actions={false} theme={"theme":{"light":"github-dark","dark":"material-theme-darker"}}
%%{init: {'theme':'base', 'themeVariables': { 'primaryColor':'#FFFBDE','primaryTextColor':'#1C1C1C','primaryBorderColor':'#FFD147','lineColor':'#6F5410','secondaryColor':'#FFFDF5','tertiaryColor':'#FFEB94','fontSize':'13px'}}}%%
flowchart LR
A[WebSocket open] --> B[START_TIME]
B --> C[AUDIO PCM chunks]
C --> D[RU9G]
```
## Prerequisites
Completed [Partner onboarding](/documentation/get-started/partner-onboarding) and have a valid `sdp_suki_token`.An open WebSocket on `wss://sdp.suki.ai/ws/stream` for an Ambient or Form filling session.You are not using Dictation `/ws/transcribe` for this recipe. Dictation skips `START_TIME`.
## Solution
Create the session and open `/ws/stream` ([Authenticate Browser WebSockets](/documentation/cookbooks/browser-websocket-auth) if you are in a browser).
Send one `START_TIME` frame before the first PCM chunk.
Stream PCM, then [End Ambient after Streaming](/documentation/cookbooks/end-ambient-after-streaming).
Send one JSON text frame:
```json theme={"theme":{"light":"github-dark","dark":"material-theme-darker"}}
{ "type": "START_TIME", "data": "" }
```
`data` is standard Base64 of the UTF-8 bytes of an RFC 3339 timestamp, for example `2026-04-25T12:34:56Z`. See [Ambient streaming wire format](/documentation/how-to/audio-streaming/websocket-streaming-wire-format-ambient).
```typescript theme={"theme":{"light":"github-dark","dark":"material-theme-darker"}}
// send-start-time-before-ambient-audio.ts
// Call once after the ambient (or Form filling) WebSocket opens, before PCM.
function toBase64Utf8(text: string): string {
if (typeof Buffer !== "undefined") {
return Buffer.from(text, "utf8").toString("base64");
}
return btoa(text);
}
/** RFC 3339 UTC without milliseconds, for example 2026-04-25T12:34:56Z */
function rfc3339Now(): string {
return new Date().toISOString().replace(/\.\d{3}Z$/, "Z");
}
function sendStartTime(ws: WebSocket): void {
const stamp = rfc3339Now();
ws.send(
JSON.stringify({
type: "START_TIME",
data: toBase64Utf8(stamp),
})
);
}
// Example after open:
// sendStartTime(ws);
// then send AUDIO frames with base64 PCM in the data field
```
```python theme={"theme":{"light":"github-dark","dark":"material-theme-darker"}}
# send_start_time_before_ambient_audio.py
# Call once after the ambient (or Form filling) WebSocket opens, before PCM.
import base64
import json
from datetime import datetime, timezone
def rfc3339_now() -> str:
# Example shape: 2026-04-25T12:34:56Z
return datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")
def send_start_time(ws) -> None:
stamp = rfc3339_now()
ws.send(
json.dumps(
{
"type": "START_TIME",
"data": base64.b64encode(stamp.encode("utf-8")).decode("ascii"),
}
)
)
# Example after open:
# send_start_time(ws)
# then send AUDIO frames with base64 PCM in the data field
```
Connect to `wss://sdp.suki.ai/ws/stream` with session auth. Form filling uses the same path and wire format.
Encode an RFC 3339 timestamp as UTF-8, Base64 it, and send `{ "type": "START_TIME", "data": "..." }` before any PCM.
Send `{ "type": "AUDIO", "data": "" }` in about 100 ms chunks. Do not send binary WebSocket frames.
Finish with `RU9G`, close the socket, and call REST end. See [End Ambient after Streaming](/documentation/cookbooks/end-ambient-after-streaming).
## Try it
* **START\_TIME is first**
After open, the first outbound message on `/ws/stream` should be `START_TIME`, then PCM `AUDIO` frames.
* **Form filling uses the same marker**
Form filling streams on `/ws/stream` with the same `START_TIME` + `data` PCM + `RU9G` contract.
* **Dictation does not**
On `/ws/transcribe`, skip `START_TIME` and end with `AUDIO_END` instead of `RU9G`.
## Common mistakes
* Send PCM before `START_TIME` on `/ws/stream`.
* Put the raw timestamp string in `data` without Base64.
* Send `START_TIME` on Dictation `/ws/transcribe`.
## Related cookbooks
Auth browser WebSocket with protocols.
Send RU9G, then end session.