CLI audio chat with ffmpeg and Azure AI realtime API

rt-chat.script.php

<?php namespace test;

use com\openai\realtime\RealtimeApi;
use util\cmd\Console;
use lang\Process;

const SPINNER= ['⠋', '⠙', '⠹', '⠸', '⠼', '⠴', '⠦', '⠧', '⠇', '⠏'];

// Wave file header
function wave($numChannels, $sampleRate, $bitsPerSample, $durationSeconds) {
  $numSamples= $sampleRate * $durationSeconds;
  $byteRate= $sampleRate * $numChannels * $bitsPerSample / 8;
  $blockAlign= $numChannels * $bitsPerSample / 8;
  $dataSize= $numSamples * $blockAlign;

  return pack(
    'a*Va*a*VvvVVvva*V',

    // Header
    'RIFF',
    44 + $dataSize - 8,
    'WAVE',

    // First chunk with audio format PCM (1)
    'fmt ',
    16,
    1,
    $numChannels,
    $sampleRate,
    $byteRate,
    $blockAlign,
    $bitsPerSample,

    // Second chunk
    'data',
    $dataSize
  );
}

$remote= 'wss://example.openai.azure.com/openai/realtime?api-version=2024-10-01-preview&deployment=gpt-4o-realtime-preview';
$auth= ['api-key' => getenv('AZUREAI_API_KEY')];

$api= new RealtimeApi($remote);
$api->connect($auth);

// 'amuch', 'dan', 'elan', 'marilyn', 'breeze', 'cove', 'ember', 'jupiter', 'alloy', 'echo', and 'shimmer'
Console::writeLine($api->transmit([
  'type'    => 'session.update',
  'session' => ['voice' => $argv[1] ?? 'alloy', 'turn_detection' => null],
]));

$record= new Process(
  'ffmpeg',
  ['-v', '0', '-f', 'pulse', '-sample_rate', '24000', '-channels', '1', '-i', 'default', '-f', 'wav', '-'],
  null,
  null,
  [0 => STDIN, 2 => STDERR]
);
$record->out->read(44); // Discard headers

Console::write("\033[37;1mListening, press 'q' to end \033[0m[ ");
$i= 0;
do {
  Console::write("\010", SPINNER[$i++ % sizeof(SPINNER)]);

  $chunk= $record->out->read();
  strlen($chunk) && $api->send([
    'type'  => 'input_audio_buffer.append',
    'audio' => base64_encode($chunk),
  ]);
} while (!$record->out->eof());

$record->out->close();
$record->close();
Console::writeLine("\010", $i, ' chunk(s) transmitted]');
$api->transmit(['type' => 'input_audio_buffer.commit']);

$api->transmit([
  'type'     => 'response.create',
  'response' => ['modalities'   => ['audio', 'text']],
]);

// 24kHz mono pcm16, 120 seconds (?!)
$proc= new Process('ffplay', ['-v', '0', '-nodisp', '-autoexit', '-'], null, null, [1 => STDOUT, 2 => STDERR]);
$proc->in->write(wave(1, 24_000, 16, 120));

Console::writeLine("\033[34m");
do {
  $event= $api->receive();

  switch ($event['type']) {
    case 'response.audio.delta':
      $proc->in->write(base64_decode($event['delta']));
      break;

    case 'response.audio_transcript.delta':
      Console::write($event['delta']);
      break;

    case 'error':
      Console::$err->writeLine();
      Console::$err->writeLine($event);
      break 2;
  }
} while ('response.done' !== $event['type']);

$proc->in->close();
$proc->close();

$api->close();
Console::writeLine("\033[0m");

Requires xp-forge/openai#15