ti-pote/apps/robot-client/src/services/audio.service.ts

import { ChildProcess, spawn } from 'node:child_process';
import { EventEmitter } from 'node:events';
import { type AudioConfig } from '../config/index.js';
import { createLogger, type Logger } from '../utils/index.js';

export interface AudioServiceEvents {
  /** Emitted when a raw PCM audio chunk is captured from the microphone */
  audio_chunk: (chunk: Buffer) => void;
  /** Emitted when playback of a response finishes */
  playback_done: () => void;
  /** Emitted on audio errors */
  error: (error: Error) => void;
}

/**
 * Audio service for Raspberry Pi.
 *
 * Uses ALSA tools (arecord/aplay) via child processes.
 * Works with any ALSA-compatible audio device:
 * - I2S (INMP441 mic, MAX98357 amp) connected directly to Pi GPIO
 * - USB audio devices
 * - Default system audio
 *
 * Audio format: PCM signed 16-bit little-endian, mono, 16kHz
 */
export class AudioService extends EventEmitter {
  private captureProcess: ChildProcess | null = null;
  private readonly logger: Logger;
  private _isCapturing = false;
  private _isPlaying = false;
  private _stoppedManually = false;

  constructor(private readonly config: AudioConfig) {
    super();
    this.logger = createLogger('audio', 'info');
  }

  get isCapturing(): boolean {
    return this._isCapturing;
  }

  get isPlaying(): boolean {
    return this._isPlaying;
  }

  /**
   * Start capturing audio from the microphone.
   * Emits 'audio_chunk' events with raw PCM buffers.
   */
  startCapture(): void {
    if (this._isCapturing) {
      this.logger.warn('Already capturing audio');
      return;
    }

    this.logger.info(
      { device: this.config.captureDevice, sampleRate: this.config.sampleRate },
      'Starting audio capture',
    );

    // arecord outputs raw PCM to stdout
    // -D: ALSA device
    // -f: format (S16_LE = signed 16-bit little-endian)
    // -r: sample rate
    // -c: channels
    // -t: type (raw = no header)
    // --buffer-size: in frames, controls latency
    const bufferFrames = Math.floor(this.config.sampleRate * (this.config.chunkDurationMs / 1000));

    this.captureProcess = spawn('arecord', [
      '-D', this.config.captureDevice,
      '-f', 'S16_LE',
      '-r', String(this.config.sampleRate),
      '-c', String(this.config.channels),
      '-t', 'raw',
      '--buffer-size', String(bufferFrames * 4),
      '--period-size', String(bufferFrames),
    ], {
      stdio: ['ignore', 'pipe', 'pipe'],
    });

    this._isCapturing = true;

    let chunkCount = 0;
    this.captureProcess.stdout?.on('data', (chunk: Buffer) => {
      chunkCount++;
      if (chunkCount === 1) {
        this.logger.info({ bytes: chunk.length }, 'First audio chunk received from arecord');
      }
      this.emit('audio_chunk', chunk);
    });

    this.captureProcess.stderr?.on('data', (data: Buffer) => {
      const msg = data.toString().trim();
      if (msg) {
        this.logger.debug({ msg }, 'arecord stderr');
      }
    });

    this.captureProcess.on('error', (err) => {
      this.logger.error({ err }, 'arecord process error');
      this._isCapturing = false;
      this.emit('error', new Error(`Audio capture failed: ${err.message}`));
    });

    this.captureProcess.on('exit', (code) => {
      this._isCapturing = false;
      if (code !== null && code !== 0 && !this._stoppedManually) {
        this.logger.warn({ code }, 'arecord exited with non-zero code');
      }
      this._stoppedManually = false;
    });
  }

  /**
   * Stop capturing audio from the microphone.
   */
  stopCapture(): void {
    if (!this.captureProcess) return;

    this.logger.info('Stopping audio capture');
    this._stoppedManually = true;
    this.captureProcess.kill('SIGTERM');
    this.captureProcess = null;
    this._isCapturing = false;
  }

  /**
   * Play audio through the speaker.
   * Accepts either raw PCM or WAV (with RIFF header) data.
   *
   * @returns Promise that resolves when playback is complete
   */
  async play(audioBuffer: Buffer): Promise<void> {
    if (this._isPlaying) {
      this.logger.warn('Already playing audio, queueing...');
    }

    this._isPlaying = true;

    const isWav = audioBuffer.length > 4 && audioBuffer.toString('ascii', 0, 4) === 'RIFF';

    return new Promise((resolve, reject) => {
      const args = isWav
        ? ['-D', this.config.playbackDevice, '-t', 'wav', '-']
        : [
            '-D', this.config.playbackDevice,
            '-f', 'S16_LE',
            '-r', String(this.config.sampleRate),
            '-c', String(this.config.channels),
            '-t', 'raw',
            '-',
          ];

      const playProcess = spawn('aplay', args, {
        stdio: ['pipe', 'ignore', 'pipe'],
      });

      playProcess.stderr?.on('data', (data: Buffer) => {
        const msg = data.toString().trim();
        if (msg && !msg.startsWith('Playing') && !msg.startsWith('Warning')) {
          this.logger.error({ msg }, 'aplay stderr');
        }
      });

      playProcess.on('error', (err) => {
        this._isPlaying = false;
        reject(new Error(`Audio playback failed: ${err.message}`));
      });

      playProcess.on('exit', (code) => {
        this._isPlaying = false;
        if (code === 0 || code === null) {
          this.emit('playback_done');
          resolve();
        } else {
          reject(new Error(`aplay exited with code ${code}`));
        }
      });

      // Write audio data to aplay's stdin and close it
      playProcess.stdin?.write(audioBuffer);
      playProcess.stdin?.end();
    });
  }

  /**
   * Stop any currently playing audio.
   */
  stopPlayback(): void {
    // aplay is spawned per-play, so we can't easily stop it here
    // For interrupt support, we'd track the play process
    this._isPlaying = false;
  }

  /**
   * Clean up resources.
   */
  async destroy(): Promise<void> {
    this.stopCapture();
    this.removeAllListeners();
  }
}