Recording Audio

The MicdropRecorder class enables recording of audio messages during voice calls, capturing both user and assistant audio with their corresponding conversation messages.

Basic Usage

import { MicdropServer, MicdropRecorder, Logger } from '@micdrop/server'

const server = new MicdropServer(socket, { agent, stt, tts })

// Create recorder
const recorder = new MicdropRecorder(server)
recorder.logger = new Logger('MicdropRecorder')

// Listen for audio messages
recorder.on('AudioMessage', (audioMessage) => {
  console.log('Audio recorded:', {
    role: audioMessage.role,
    messageIndex: audioMessage.messageIndex,
    bufferSize: audioMessage.buffer.length,
  })
})

// Get all recordings when call ends
recorder.on('Complete', (audioMessages) => {
  console.log(`Call complete with ${audioMessages.length} audio messages`)
})

AudioMessage Structure

Each AudioMessage contains:

interface AudioMessage {
  buffer: Buffer // Raw audio data
  messageIndex: number // Index in agent.conversation
  message: string // The message content
  role: 'user' | 'assistant' // Speaker role
}

Saving to File

Save audio messages incrementally as they arrive:

import { writeFileSync } from 'fs'

const audioMessages: object[] = []

recorder.on('AudioMessage', (audioMessage) => {
  const data = {
    messageIndex: audioMessage.messageIndex,
    message: audioMessage.message,
    role: audioMessage.role,
    buffer: audioMessage.buffer.toString('base64'),
    timestamp: new Date().toISOString(),
  }
  audioMessages.push(data)
  writeFileSync('recording.json', JSON.stringify(audioMessages, null, 2))
})

Recorder Events

The recorder emits the following events:

Event	Payload	Description
AudioMessage	AudioMessage	Emitted when an audio message is complete
Complete	AudioMessage[]	Emitted when the call ends with all recordings

Server Events

MicdropRecorder relies on events from MicdropServer:

Event	Payload	Description
UserAudio	Buffer	Raw audio chunk from user
AssistantAudio	Buffer	Raw audio chunk from TTS
End	MicdropCallSummary	Call ended

You can also listen to these events directly on MicdropServer:

server.on('UserAudio', (chunk) => {
  console.log('User audio chunk:', chunk.length, 'bytes')
})

server.on('AssistantAudio', (chunk) => {
  console.log('Assistant audio chunk:', chunk.length, 'bytes')
})

server.on('End', (summary) => {
  console.log('Call ended:', summary)
})

Cleanup

Call destroy() to clean up listeners when done:

recorder.destroy()

Technical Notes

• User audio chunks arrive before the user message is added to the conversation
• The recorder buffers chunks and associates them with messages when available
• Assistant audio is finalized when the next user speech begins or when the call ends
• Audio is stored as raw PCM buffers (16kHz, 16-bit, mono)