Type Alias: MediaStreamIncoming

MediaStreamIncoming = { payload: Buffer; sequenceNumber?: number; streamSid: string; type: "audio"; } | { digit: string; durationMs?: number; streamSid: string; type: "dtmf"; } | { callSid: string; metadata?: Record<string, unknown>; streamSid: string; type: "start"; } | { streamSid: string; type: "stop"; } | { name: string; streamSid: string; type: "mark"; }

Defined in: packages/agentos/src/channels/telephony/MediaStreamParser.ts:126

Discriminated union of all normalised events that can arrive on a media stream WebSocket connection, regardless of the underlying telephony provider.

Variant summary

type	When it fires	Key payload fields
audio	Each inbound audio chunk (~20ms intervals)	payload (mu-law Buffer)
dtmf	Caller presses a phone keypad button	digit, durationMs?
start	Stream session begins (metadata available)	callSid, metadata?
stop	Stream session ends / call disconnects	(none beyond streamSid)
mark	Named sync point injected into audio stream	name

All variants carry a streamSid field to identify which stream the event belongs to (important when a single server handles multiple concurrent calls).

Type Declaration

{ payload: Buffer; sequenceNumber?: number; streamSid: string; type: "audio"; }

payload

payload: Buffer

Raw mu-law bytes decoded from whatever encoding the provider uses.

sequenceNumber?

optional sequenceNumber: number

Monotonically increasing sequence number, when provided.

streamSid

streamSid: string

Provider-assigned stream identifier.

type

type: "audio"

Inbound audio chunk encoded as mu-law 8-bit 8 kHz PCM.

Audio arrives as small chunks (typically 20ms / 160 bytes) at regular intervals for the duration of the call. The pipeline must decode mu-law -> PCM Int16 -> resample -> Float32 before feeding to STT/VAD.

{ digit: string; durationMs?: number; streamSid: string; type: "dtmf"; }

digit

digit: string

Single character digit pressed by the caller (0-9, *, #, A-D).

durationMs?

optional durationMs: number

Duration the key was held, in milliseconds, when reported.

streamSid

streamSid: string

Provider-assigned stream identifier.

type

type: "dtmf"

DTMF tone detected by the provider during the call.

Not all providers relay DTMF over the media stream -- Telnyx, for example, only delivers DTMF via HTTP webhooks. Check the provider's parser documentation for availability.

{ callSid: string; metadata?: Record<string, unknown>; streamSid: string; type: "start"; }

callSid

callSid: string

Provider call-leg identifier (e.g. Twilio CallSid, Telnyx call_control_id).

metadata?

optional metadata: Record<string, unknown>

Additional provider-specific metadata attached to the start event.

streamSid

streamSid: string

Provider-assigned stream identifier.

type

type: "start"

Stream successfully started; metadata about the call is available.

This is always the first meaningful event on a new stream connection. The TelephonyStreamTransport transitions from connecting to open upon receiving this event and sends the optional MediaStreamParser.formatConnected acknowledgment.

{ streamSid: string; type: "stop"; }

streamSid

streamSid: string

Provider-assigned stream identifier.

type

type: "stop"

Call ended or stream was explicitly stopped.

The TelephonyStreamTransport transitions to closed and emits a 'close' event upon receiving this.

{ name: string; streamSid: string; type: "mark"; }

name

name: string

The label assigned to this mark point.

streamSid

streamSid: string

Provider-assigned stream identifier.

type

type: "mark"

Named marker injected into the audio stream for synchronisation.

Marks are used to correlate outbound audio playback completion with application logic (e.g., knowing when a TTS utterance finished playing so the agent can transition from speaking to listening).