assemblyai/dist/services/streaming/browser/dual-channel-capture.d.ts

The AssemblyAI JavaScript SDK provides an easy-to-use interface for interacting with the AssemblyAI API, which supports async and real-time transcription, as well as the latest LeMUR models.

import { StreamingTranscriber } from "../service";
export { BrowserOnlyError } from "../../../types/streaming/dual-channel";
type ErrorListener = (err: Error) => void;
export type DualChannelCaptureParams = {
    /** Microphone MediaStream. Caller should set `echoCancellation: true` at `getUserMedia` time. */
    micStream: MediaStream;
    /** System-audio MediaStream (e.g. `getDisplayMedia({ audio: true })`). */
    systemStream: MediaStream;
    /**
     * The transcriber to push tagged PCM into. MUST be constructed with
     * `channels: [{ name: "mic" }, { name: "system" }]` so the per-channel
     * `sendAudio` calls succeed.
     */
    transcriber: StreamingTranscriber;
    /**
     * Target sample rate sent to the transcriber. Defaults to 16000. The
     * AudioContext runs at the device's native rate; resampling happens inside
     * the encoder worklet (forcing `AudioContext({ sampleRate })` is unreliable
     * across browsers).
     */
    targetSampleRate?: number;
};
/**
 * Browser-only adapter that pumps two `MediaStream`s into a `StreamingTranscriber`
 * configured for dual-channel mode. Each `MediaStream` runs through its own
 * `pcm16-encoder` AudioWorklet (resample to `targetSampleRate`, encode to Int16
 * PCM); each PCM chunk is forwarded via `transcriber.sendAudio(pcm, { channel })`.
 *
 * All dual-channel orchestration (mixing, VAD, per-word attribution) lives inside
 * `StreamingTranscriber` — this class is a pure I/O adapter. Non-browser runtimes
 * can replicate its job by pushing tagged PCM into `transcriber.sendAudio` directly.
 *
 * Caller responsibilities:
 * - **Echo cancellation** is set at `getUserMedia` time (`audio: { echoCancellation: true }`).
 * - **System-audio capture** is platform-dependent. Chrome's `getDisplayMedia({ audio: true })`
 *   captures tab audio (and on Windows, full system audio when sharing the whole screen).
 *   macOS requires a virtual loopback driver (e.g. BlackHole) to expose system audio at all.
 * - **Token auth.** Construct the transcriber with `token` — API-key auth is unsupported in browsers.
 * - **Stream ownership.** `stop()` tears down the AudioContext but does NOT stop the
 *   `MediaStreamTrack`s passed in — callers own those.
 */
export declare class DualChannelCapture {
    private readonly params;
    private errorListener?;
    private context?;
    private micSource?;
    private sysSource?;
    private micEncoder?;
    private sysEncoder?;
    private running;
    constructor(params: DualChannelCaptureParams);
    on(event: "error", listener: ErrorListener): void;
    /**
     * Wire the capture pipeline and start pumping tagged PCM into the transcriber.
     * The transcriber must already be connected. Returns once the worklet is
     * registered and the audio graph is live.
     */
    start(): Promise<void>;
    private makeEncoder;
    /**
     * Tear down internal nodes and close the AudioContext. Does NOT stop the
     * caller-provided MediaStream tracks — they remain available for preview UI,
     * recording, etc. Idempotent.
     */
    stop(): Promise<void>;
}