After a lot more research and playing around, I found a good workaround solution.
First of all I think this is an iOS bug. When all below conditions are true I found that the voice instruction itself is also ducked (or at least it sounds ducked) resulting in the voice instruction playing at the same volume as the DUCKED music (thus way too soft to hear well).
- Playing music in the background
- Ducking this background music through
the
.duckOther audioSessionCategory
- Playing a voiceUtterance through AVSpeechSynthesizer
- Playing audio over a connected bluetooth
device (like bluetooth headset or bluetooth car speakers)
The workaround solution I found is to feed the speechUtterance to an AVAudioEngine. This can only be done on iOS13 or above, since that adds the .write method to AVSpeechSynthesizer
In short I use AVAudioEngine, AVAudioUnitEQ and AVAudioPlayerNode, setting the globalGain property of the AVAudioUnitEQ to about 10 dB. There are also a few quirks with this, but they can be worked around (see code comments).
Here's the complete code:
import UIKit
import AVFoundation
import MediaPlayer
class ViewController: UIViewController {
// MARK: AVAudio properties
var engine = AVAudioEngine()
var player = AVAudioPlayerNode()
var eqEffect = AVAudioUnitEQ()
var converter = AVAudioConverter(from: AVAudioFormat(commonFormat: AVAudioCommonFormat.pcmFormatInt16, sampleRate: 22050, channels: 1, interleaved: false)!, to: AVAudioFormat(commonFormat: AVAudioCommonFormat.pcmFormatFloat32, sampleRate: 22050, channels: 1, interleaved: false)!)
let synthesizer = AVSpeechSynthesizer()
var bufferCounter: Int = 0
let audioSession = AVAudioSession.sharedInstance()
override func viewDidLoad() {
super.viewDidLoad()
let outputFormat = AVAudioFormat(commonFormat: AVAudioCommonFormat.pcmFormatFloat32, sampleRate: 22050, channels: 1, interleaved: false)!
setupAudio(format: outputFormat, globalGain: 0)
}
func activateAudioSession() {
do {
try audioSession.setCategory(.playback, mode: .voicePrompt, options: [.mixWithOthers, .duckOthers])
try audioSession.setActive(true, options: .notifyOthersOnDeactivation)
} catch {
print("An error has occurred while setting the AVAudioSession.")
}
}
@IBAction func tappedPlayButton(_ sender: Any) {
eqEffect.globalGain = 0
play()
}
@IBAction func tappedPlayLoudButton(_ sender: Any) {
eqEffect.globalGain = 10
play()
}
func play() {
let path = Bundle.main.path(forResource: "voiceStart", ofType: "wav")!
let file = try! AVAudioFile(forReading: URL(fileURLWithPath: path))
self.player.scheduleFile(file, at: nil, completionHandler: nil)
let utterance = AVSpeechUtterance(string: "This is to test if iOS is able to boost the voice output above the 100% limit.")
synthesizer.write(utterance) { buffer in
guard let pcmBuffer = buffer as? AVAudioPCMBuffer, pcmBuffer.frameLength > 0 else {
print("could not create buffer or buffer empty")
return
}
// QUIRCK Need to convert the buffer to different format because AVAudioEngine does not support the format returned from AVSpeechSynthesizer
let convertedBuffer = AVAudioPCMBuffer(pcmFormat: AVAudioFormat(commonFormat: AVAudioCommonFormat.pcmFormatFloat32, sampleRate: pcmBuffer.format.sampleRate, channels: pcmBuffer.format.channelCount, interleaved: false)!, frameCapacity: pcmBuffer.frameCapacity)!
do {
try self.converter!.convert(to: convertedBuffer, from: pcmBuffer)
self.bufferCounter += 1
self.player.scheduleBuffer(convertedBuffer, completionCallbackType: .dataPlayedBack, completionHandler: { (type) -> Void in
DispatchQueue.main.async {
self.bufferCounter -= 1
print(self.bufferCounter)
if self.bufferCounter == 0 {
self.player.stop()
self.engine.stop()
try! self.audioSession.setActive(false, options: [])
}
}
})
self.converter!.reset()
//self.player.prepare(withFrameCount: convertedBuffer.frameLength)
}
catch let error {
print(error.localizedDescription)
}
}
activateAudioSession()
if !self.engine.isRunning {
try! self.engine.start()
}
if !self.player.isPlaying {
self.player.play()
}
}
func setupAudio(format: AVAudioFormat, globalGain: Float) {
// QUIRCK: Connecting the equalizer to the engine somehow starts the shared audioSession, and if that audiosession is not configured with .mixWithOthers and if it's not deactivated afterwards, this will stop any background music that was already playing. So first configure the audio session, then setup the engine and then deactivate the session again.
try? self.audioSession.setCategory(.playback, options: .mixWithOthers)
eqEffect.globalGain = globalGain
engine.attach(player)
engine.attach(eqEffect)
engine.connect(player, to: eqEffect, format: format)
engine.connect(eqEffect, to: engine.mainMixerNode, format: format)
engine.prepare()
try? self.audioSession.setActive(false)
}
}
