Voice & audio - Eka Developer Platform APIs

The core SDK supports two audio paths: recording an audio message and sending it like any other message, and a dedicated voice mode for real-time spoken conversations.

Recording audio

Start recording to receive audio chunks via a callback, then stop when the user is done. Each chunk is an AudioMetaData object.

await sdk.startRecording({
  onChunks: (chunk) => {
    // chunk: { audio (base64), format, duration, timestamp }
    bufferChunk(chunk);
  },
  onError: (error) => console.error("recording error:", error),
});

// later, when the user stops:
sdk.endRecording();

AudioMetaData

interface AudioMetaData {
  audio: string;    // base64-encoded audio chunk
  format: string;   // MIME type, e.g. "audio/mp3"
  duration: number; // milliseconds
  timestamp: number;
}

Sending audio

Send a recorded clip with sendMessage using the audio field:

await sdk.sendMessage({
  audio: {
    audio: base64Clip,
    format: "audio/mp3",
    duration: 4200,
    timestamp: 1700000000000,
  },
});

Transcripts

When audio is transcribed, the SDK emits an AUDIO_TRANSCRIPT event. Use it to show the recognized text:

import { SYNAPSE_REALTIME_EVENTS } from "@eka-care/medassist-core";

sdk.on(SYNAPSE_REALTIME_EVENTS.AUDIO_TRANSCRIPT, (evt) => {
  showTranscript(evt.data.text);
});

Voice mode

For continuous spoken conversation, use the voice getter, which returns a VoiceAgent:

const voice = sdk.voice;
// drive a real-time voice conversation through the VoiceAgent

Voice mode is a distinct, real-time conversational interface — separate from one-off audio messages. Check that your agent has voice enabled in console.eka.care.

​Recording audio

​AudioMetaData

​Sending audio

​Transcripts

​Voice mode

Recording audio

AudioMetaData

Sending audio

Transcripts

Voice mode