Fixed audio artifacts for text to speech

Author: widavies

src/main/java/example/ExampleBot.java

@@ -32,7 +32,7 @@ public ExampleBot() {
 
         // Windows (with closed captioning instead of wake detection)
         cord = VocalCord.newConfig(this).withClosedCaptioning().withTTS(SsmlVoiceGender.MALE,
-                true).build();
+                false).build();
 
         // Linux (using WSL)
 //        cord = VocalCord.newConfig(this).withWakeDetection("/mnt/c/Users/wdavi/IdeaProjects/VocalCord/native/linux/libjni_porcupine.so",

src/main/java/vocalcord/TTSEngine.java

@@ -48,8 +48,10 @@ byte[] tts(String text) throws Exception {
             byte[] pcm = audioContents.toByteArray();
 
             // Three things need to happen - big endian, stereo, pad to a multiple of 3840
-            byte[] converted = new byte[pcm.length * 2 + (AUDIO_FRAME - pcm.length * 2 % AUDIO_FRAME)]; // ensures converted is a multiple of AUDIO_FRAME
-            for(int i = 0; i < pcm.length; i += 2) {
+            // Add a frame of silence at the beginning so that the sound doesn't clip weirdly
+            byte[] converted = new byte[AUDIO_FRAME + pcm.length * 2 + (AUDIO_FRAME - pcm.length * 2 % AUDIO_FRAME)];
+            // ensures converted is a multiple of AUDIO_FRAME
+            for(int i = AUDIO_FRAME; i < pcm.length; i += 2) {
                 short reversed = Short.reverseBytes((short) ((pcm[i] << 8) | (pcm[i + 1] & 0xFF)));
                 byte low = (byte) (reversed >> 8);
                 byte high = (byte) (reversed & 0x00FF);
@@ -66,6 +68,8 @@ byte[] tts(String text) throws Exception {
     }
 
     void say(String phrase) throws Exception {
+        this.index = Integer.MAX_VALUE;
+
         if(ttsCache != null) {
             TTSCache.CacheResponse response = ttsCache.checkCache(phrase);