|
| 1 | +package dev.agixt.agixt |
| 2 | + |
| 3 | +import android.content.Context |
| 4 | +import android.content.Intent |
| 5 | +import android.os.Bundle |
| 6 | +import android.speech.RecognitionListener |
| 7 | +import android.speech.RecognizerIntent |
| 8 | +import android.speech.SpeechRecognizer |
| 9 | +import android.util.Log |
| 10 | +import io.flutter.plugin.common.MethodChannel |
| 11 | +import io.flutter.plugin.common.BinaryMessenger |
| 12 | +import java.util.Locale |
| 13 | + |
| 14 | +/** |
| 15 | + * Handles wake word detection using Android's SpeechRecognizer |
| 16 | + * Listens for "computer" keyword to trigger voice input |
| 17 | + */ |
| 18 | +class WakeWordHandler( |
| 19 | + private val context: Context, |
| 20 | + private val binaryMessenger: BinaryMessenger |
| 21 | +) { |
| 22 | + private val TAG = "WakeWordHandler" |
| 23 | + private val CHANNEL = "dev.agixt.agixt/wake_word" |
| 24 | + |
| 25 | + private var speechRecognizer: SpeechRecognizer? = null |
| 26 | + private var isListening = false |
| 27 | + private var isPaused = false |
| 28 | + private var wakeWord = "computer" |
| 29 | + private var sensitivity = 0.5f |
| 30 | + private lateinit var methodChannel: MethodChannel |
| 31 | + |
| 32 | + fun initialize() { |
| 33 | + methodChannel = MethodChannel(binaryMessenger, CHANNEL) |
| 34 | + |
| 35 | + methodChannel.setMethodCallHandler { call, result -> |
| 36 | + when (call.method) { |
| 37 | + "initialize" -> { |
| 38 | + val word = call.argument<String>("wakeWord") ?: "computer" |
| 39 | + val sens = call.argument<Double>("sensitivity")?.toFloat() ?: 0.5f |
| 40 | + wakeWord = word.lowercase() |
| 41 | + sensitivity = sens |
| 42 | + setupSpeechRecognizer() |
| 43 | + result.success(true) |
| 44 | + } |
| 45 | + "start" -> { |
| 46 | + startListening() |
| 47 | + result.success(true) |
| 48 | + } |
| 49 | + "stop" -> { |
| 50 | + stopListening() |
| 51 | + result.success(true) |
| 52 | + } |
| 53 | + "pause" -> { |
| 54 | + pause() |
| 55 | + result.success(true) |
| 56 | + } |
| 57 | + "resume" -> { |
| 58 | + resume() |
| 59 | + result.success(true) |
| 60 | + } |
| 61 | + "setWakeWord" -> { |
| 62 | + wakeWord = (call.argument<String>("wakeWord") ?: "computer").lowercase() |
| 63 | + result.success(true) |
| 64 | + } |
| 65 | + "setSensitivity" -> { |
| 66 | + sensitivity = call.argument<Double>("sensitivity")?.toFloat() ?: 0.5f |
| 67 | + result.success(true) |
| 68 | + } |
| 69 | + "isAvailable" -> { |
| 70 | + result.success(SpeechRecognizer.isRecognitionAvailable(context)) |
| 71 | + } |
| 72 | + else -> result.notImplemented() |
| 73 | + } |
| 74 | + } |
| 75 | + } |
| 76 | + |
| 77 | + private fun setupSpeechRecognizer() { |
| 78 | + if (!SpeechRecognizer.isRecognitionAvailable(context)) { |
| 79 | + Log.e(TAG, "Speech recognition not available on this device") |
| 80 | + return |
| 81 | + } |
| 82 | + |
| 83 | + speechRecognizer?.destroy() |
| 84 | + speechRecognizer = SpeechRecognizer.createSpeechRecognizer(context) |
| 85 | + |
| 86 | + speechRecognizer?.setRecognitionListener(object : RecognitionListener { |
| 87 | + override fun onReadyForSpeech(params: Bundle?) { |
| 88 | + Log.d(TAG, "Ready for speech") |
| 89 | + } |
| 90 | + |
| 91 | + override fun onBeginningOfSpeech() { |
| 92 | + Log.d(TAG, "Beginning of speech") |
| 93 | + } |
| 94 | + |
| 95 | + override fun onRmsChanged(rmsdB: Float) { |
| 96 | + // Audio level changed - could be used for UI feedback |
| 97 | + } |
| 98 | + |
| 99 | + override fun onBufferReceived(buffer: ByteArray?) { |
| 100 | + // Not typically used |
| 101 | + } |
| 102 | + |
| 103 | + override fun onEndOfSpeech() { |
| 104 | + Log.d(TAG, "End of speech") |
| 105 | + } |
| 106 | + |
| 107 | + override fun onError(error: Int) { |
| 108 | + val errorMessage = when (error) { |
| 109 | + SpeechRecognizer.ERROR_AUDIO -> "Audio recording error" |
| 110 | + SpeechRecognizer.ERROR_CLIENT -> "Client side error" |
| 111 | + SpeechRecognizer.ERROR_INSUFFICIENT_PERMISSIONS -> "Insufficient permissions" |
| 112 | + SpeechRecognizer.ERROR_NETWORK -> "Network error" |
| 113 | + SpeechRecognizer.ERROR_NETWORK_TIMEOUT -> "Network timeout" |
| 114 | + SpeechRecognizer.ERROR_NO_MATCH -> "No match found" |
| 115 | + SpeechRecognizer.ERROR_RECOGNIZER_BUSY -> "Recognizer busy" |
| 116 | + SpeechRecognizer.ERROR_SERVER -> "Server error" |
| 117 | + SpeechRecognizer.ERROR_SPEECH_TIMEOUT -> "Speech timeout" |
| 118 | + else -> "Unknown error" |
| 119 | + } |
| 120 | + Log.d(TAG, "Error: $errorMessage ($error)") |
| 121 | + |
| 122 | + // Restart listening if still enabled and not paused |
| 123 | + if (isListening && !isPaused && error != SpeechRecognizer.ERROR_INSUFFICIENT_PERMISSIONS) { |
| 124 | + // Small delay before restarting to avoid rapid cycling |
| 125 | + android.os.Handler(context.mainLooper).postDelayed({ |
| 126 | + if (isListening && !isPaused) { |
| 127 | + startRecognition() |
| 128 | + } |
| 129 | + }, 500) |
| 130 | + } |
| 131 | + } |
| 132 | + |
| 133 | + override fun onResults(results: Bundle?) { |
| 134 | + val matches = results?.getStringArrayList(SpeechRecognizer.RESULTS_RECOGNITION) |
| 135 | + val confidences = results?.getFloatArray(SpeechRecognizer.CONFIDENCE_SCORES) |
| 136 | + |
| 137 | + Log.d(TAG, "Results: $matches") |
| 138 | + |
| 139 | + if (matches != null) { |
| 140 | + for ((index, match) in matches.withIndex()) { |
| 141 | + val confidence = confidences?.getOrNull(index) ?: 0.5f |
| 142 | + val matchLower = match.lowercase() |
| 143 | + |
| 144 | + // Check if wake word is detected with sufficient confidence |
| 145 | + // Adjust threshold based on sensitivity setting |
| 146 | + val threshold = 1.0f - sensitivity |
| 147 | + |
| 148 | + if (matchLower.contains(wakeWord) && confidence >= threshold) { |
| 149 | + Log.i(TAG, "Wake word '$wakeWord' detected! Confidence: $confidence") |
| 150 | + |
| 151 | + // Notify Flutter |
| 152 | + methodChannel.invokeMethod("onWakeWordDetected", mapOf( |
| 153 | + "wakeWord" to wakeWord, |
| 154 | + "confidence" to confidence.toDouble(), |
| 155 | + "transcript" to match |
| 156 | + )) |
| 157 | + |
| 158 | + // Pause listening while user is speaking |
| 159 | + pause() |
| 160 | + return |
| 161 | + } |
| 162 | + } |
| 163 | + } |
| 164 | + |
| 165 | + // Continue listening if wake word wasn't detected |
| 166 | + if (isListening && !isPaused) { |
| 167 | + startRecognition() |
| 168 | + } |
| 169 | + } |
| 170 | + |
| 171 | + override fun onPartialResults(partialResults: Bundle?) { |
| 172 | + val matches = partialResults?.getStringArrayList(SpeechRecognizer.RESULTS_RECOGNITION) |
| 173 | + |
| 174 | + if (matches != null) { |
| 175 | + for (match in matches) { |
| 176 | + val matchLower = match.lowercase() |
| 177 | + if (matchLower.contains(wakeWord)) { |
| 178 | + Log.d(TAG, "Partial wake word detection: $match") |
| 179 | + // Could send partial detection event if needed |
| 180 | + } |
| 181 | + } |
| 182 | + } |
| 183 | + } |
| 184 | + |
| 185 | + override fun onEvent(eventType: Int, params: Bundle?) { |
| 186 | + Log.d(TAG, "Event: $eventType") |
| 187 | + } |
| 188 | + }) |
| 189 | + } |
| 190 | + |
| 191 | + private fun startListening() { |
| 192 | + if (!SpeechRecognizer.isRecognitionAvailable(context)) { |
| 193 | + Log.e(TAG, "Speech recognition not available") |
| 194 | + return |
| 195 | + } |
| 196 | + |
| 197 | + isListening = true |
| 198 | + isPaused = false |
| 199 | + startRecognition() |
| 200 | + } |
| 201 | + |
| 202 | + private fun startRecognition() { |
| 203 | + if (speechRecognizer == null) { |
| 204 | + setupSpeechRecognizer() |
| 205 | + } |
| 206 | + |
| 207 | + val intent = Intent(RecognizerIntent.ACTION_RECOGNIZE_SPEECH).apply { |
| 208 | + putExtra(RecognizerIntent.EXTRA_LANGUAGE_MODEL, RecognizerIntent.LANGUAGE_MODEL_FREE_FORM) |
| 209 | + putExtra(RecognizerIntent.EXTRA_LANGUAGE, Locale.getDefault()) |
| 210 | + putExtra(RecognizerIntent.EXTRA_PARTIAL_RESULTS, true) |
| 211 | + putExtra(RecognizerIntent.EXTRA_MAX_RESULTS, 5) |
| 212 | + // Shorter silence timeouts for wake word detection |
| 213 | + putExtra(RecognizerIntent.EXTRA_SPEECH_INPUT_MINIMUM_LENGTH_MILLIS, 1000) |
| 214 | + putExtra(RecognizerIntent.EXTRA_SPEECH_INPUT_COMPLETE_SILENCE_LENGTH_MILLIS, 1500) |
| 215 | + putExtra(RecognizerIntent.EXTRA_SPEECH_INPUT_POSSIBLY_COMPLETE_SILENCE_LENGTH_MILLIS, 1000) |
| 216 | + } |
| 217 | + |
| 218 | + try { |
| 219 | + speechRecognizer?.startListening(intent) |
| 220 | + } catch (e: Exception) { |
| 221 | + Log.e(TAG, "Error starting recognition: ${e.message}") |
| 222 | + } |
| 223 | + } |
| 224 | + |
| 225 | + private fun stopListening() { |
| 226 | + isListening = false |
| 227 | + isPaused = false |
| 228 | + speechRecognizer?.stopListening() |
| 229 | + speechRecognizer?.cancel() |
| 230 | + } |
| 231 | + |
| 232 | + private fun pause() { |
| 233 | + isPaused = true |
| 234 | + speechRecognizer?.stopListening() |
| 235 | + speechRecognizer?.cancel() |
| 236 | + } |
| 237 | + |
| 238 | + private fun resume() { |
| 239 | + if (isListening) { |
| 240 | + isPaused = false |
| 241 | + startRecognition() |
| 242 | + } |
| 243 | + } |
| 244 | + |
| 245 | + fun destroy() { |
| 246 | + stopListening() |
| 247 | + speechRecognizer?.destroy() |
| 248 | + speechRecognizer = null |
| 249 | + } |
| 250 | +} |
0 commit comments