android synthesis enhancements

jni/src/main/java/com/samtupy/nvgt/TTS.java

@@ -21,6 +21,14 @@
 import java.util.Set;
 import java.util.stream.Collectors;
 import org.libsdl.app.SDL;
+import android.os.storage.StorageManager;
+import android.os.ProxyFileDescriptorCallback;
+import android.os.Handler;
+import java.io.ByteArrayOutputStream;
+import java.io.File;
+import java.nio.file.Path;
+import android.os.ParcelFileDescriptor;
+import java.io.IOException;
 
 public class TTS {
 	// First the static screen reader methods.
@@ -161,8 +169,7 @@ public void setVolume(float volume) {
 		ttsVolume = volume;
 	}
 
-	@Override
-	public void finalize() {
+	public void shutdown() {
 		if (isActive()) {
 			tts.shutdown();
 		}
@@ -205,4 +212,58 @@ public float getVolume() {
 	public float getPan() {
 		return ttsPan;
 	}
+
+	public boolean speakToFile(String filename, String text) throws IOException {
+		Path p = Path.of(filename);
+		Path actual_path;
+		if (p.isAbsolute()) {
+			actual_path = p;
+		} else {
+			actual_path = Path.of(SDL.getContext().getFilesDir().getAbsolutePath(), filename);
+		}
+		File f = actual_path.toFile();
+		f.createNewFile();
+		if (!f.canWrite()) return false;
+		Bundle params = new Bundle();
+		params.putFloat(TextToSpeech.Engine.KEY_PARAM_VOLUME, ttsVolume);
+		params.putFloat(TextToSpeech.Engine.KEY_PARAM_PAN, ttsPan);
+		if (text.length() > tts.getMaxSpeechInputLength()) {
+			return false;
+		}
+		return tts.synthesizeToFile(text, params, f, null) == TextToSpeech.SUCCESS;
+	}
+
+	public byte[] speakToMemory(String text) throws IOException {
+		// It is much simpler to use the storage manager interface instead of memfd_create here
+		// To do: investigate whether it would be worth it creating a separate looper so we do not potentially risk blocking the main one
+		ByteArrayOutputStream stream = new ByteArrayOutputStream();
+		Handler handler = Handler.createAsync(SDL.getContext().getMainLooper());
+		Context context = SDL.getContext();
+		StorageManager manager = (StorageManager)context.getSystemService(context.STORAGE_SERVICE);
+		ParcelFileDescriptor pfd = manager.openProxyFileDescriptor(ParcelFileDescriptor.MODE_WRITE_ONLY, new ProxyFileDescriptorCallback() {
+			public void onFsync () { }
+
+			public long onGetSize () {
+				return stream.size();
+			}
+
+			public int onRead (long offset, int size, byte[] data) {
+				return 0;
+			}
+
+			public void onRelease () { }
+
+			public int onWrite (long offset, int size, byte[] data) {
+				stream.writeBytes(data);
+				return size;
+			}
+		}, handler);
+		Bundle params = new Bundle();
+		params.putFloat(TextToSpeech.Engine.KEY_PARAM_VOLUME, ttsVolume);
+		params.putFloat(TextToSpeech.Engine.KEY_PARAM_PAN, ttsPan);
+		if (tts.synthesizeToFile(text, params, pfd, null) == TextToSpeech.SUCCESS) {
+			return stream.toByteArray();
+		}
+		return null;
+	}
 }

src/tts.cpp

@@ -27,6 +27,7 @@
 #include <Poco/Exception.h>
 #include <Poco/Format.h>
 #include <SDL3/SDL.h>
+#include <stdexcept>
 #endif
 
 char* minitrim(char* data, unsigned long* bufsize, int bitrate, int channels) {
@@ -66,7 +67,12 @@ tts_voice::tts_voice(const std::string& builtin_voice_name) {
 	audioout = 0;
 }
 tts_voice::~tts_voice() {
-	//destroy();
+	#ifdef __ANDROID__
+	// Manually clean up here
+	if (env != NULL && TTSObj != NULL && midShutdown != NULL) {
+		env->CallVoidMethod(TTSObj, midShutdown);
+	}
+	#endif
 }
 void tts_voice::setup() {
 	#ifdef _WIN32
@@ -106,7 +112,10 @@ void tts_voice::setup() {
 	midGetPitch = env->GetMethodID(TTSClass, "getPitch", "()F");
 	midGetPan = env->GetMethodID(TTSClass, "getPan", "()F");
 	midGetVolume = env->GetMethodID(TTSClass, "getVolume", "()F");
-	if (!midIsActive || !midIsSpeaking || !midSpeak || !midSilence || !midGetVoice || !midSetRate || !midSetPitch || !midSetPan || !midSetVolume || !midGetVoices || !midSetVoice || !midGetMaxSpeechInputLength || !midGetPitch || !midGetPan || !midGetRate || !midGetVolume) throw Poco::Exception("One or more methods on the TTS class could not be retrieved from JNI!");
+	midSpeakToFile = env->GetMethodID(TTSClass, "speakToFile", "(Ljava/lang/String;Ljava/lang/String)Z");
+	midSpeakToMemory = env->GetMethodID(TTSClass, "speakToMemory", "(Ljava/lang/String;)[B");
+	midShutdown = env->GetMethodID(TTSClass, "shutdown", "()V");
+	if (!midIsActive || !midIsSpeaking || !midSpeak || !midSilence || !midGetVoice || !midSetRate || !midSetPitch || !midSetPan || !midSetVolume || !midGetVoices || !midSetVoice || !midGetMaxSpeechInputLength || !midGetPitch || !midGetPan || !midGetRate || !midGetVolume || !midSpeakToFile || !midSpeakToMemory || !midShutdown) throw Poco::Exception("One or more methods on the TTS class could not be retrieved from JNI!");
 	if (!env->CallBooleanMethod(TTSObj, midIsActive)) throw Poco::Exception("TTS engine could not be initialized!");
 	voice_index = 1;
 	#else
@@ -237,7 +246,26 @@ bool tts_voice::speak_to_file(const std::string& filename, const std::string& te
 	}
 	#elif defined(__ANDROID__)
 	else {
-		return false;
+		jstring jtext = env->NewStringUTF(text.c_str());
+		jstring jfile = env->NewStringUTF(filename.c_str());
+		auto res = env->CallBooleanMethod(TTSObj, midSpeakToFile, jfile, jtext);
+		if (auto exc = env->ExceptionOccurred(); exc) {
+			// Translate into C++ exception
+			jclass ThrowableClass = env->FindClass("java/lang/Throwable");
+			if (!ThrowableClass) throw Poco::Exception("This JVM implementation is broken: could not translate Java exception!");
+			jmethodID midGetMessage = env->GetMethodID(ThrowableClass, "getMessage", "()Ljava/lang/String;");
+			if (!midGetMessage) throw Poco::Exception("This JVM implementation is broken: Throwable does not have getMessage method!");
+			jstring message = static_cast<jstring>(env->CallObjectMethod(exc, midGetMessage));
+			const char* msg_chars = env->GetStringUTFChars(message, 0);
+			std::string msg(msg_chars);
+			env->ReleaseStringUTFChars(message, msg_chars);
+			env->DeleteLocalRef(jfile);
+			env->DeleteLocalRef(jtext);
+			throw std::runtime_error(msg);
+		}
+		env->DeleteLocalRef(jfile);
+		env->DeleteLocalRef(jtext);
+		return res;
 	}
 	#endif
 	char* ptr = minitrim(data, &bufsize, bitrate, channels);
@@ -254,6 +282,7 @@ std::string tts_voice::speak_to_memory(const std::string& text) {
 	if (text.empty()) return "";
 	unsigned long bufsize;
 	char* data;
+	std::string data_res;
 	if (voice_index == builtin_index) {
 		if (samprate != 48000 || bitrate != 16 || channels != 2) {
 			samprate = 48000;
@@ -284,6 +313,31 @@ std::string tts_voice::speak_to_memory(const std::string& text) {
 	else {
 		return ""; // Not implemented yet.
 	}
+	#elif defined(__ANDROID__)
+	else {
+		jstring jtext = env->NewStringUTF(text.c_str());
+		jarray jbytes = env->CallObjectMethod(TTSObj, midSpeakToMemory, jtext);
+		if (auto exc = env->ExceptionOccurred(); exc) {
+			// Translate into C++ exception
+			jclass ThrowableClass = env->FindClass("java/lang/Throwable");
+			if (!ThrowableClass) throw Poco::Exception("This JVM implementation is broken: could not translate Java exception!");
+			jmethodID midGetMessage = env->GetMethodID(ThrowableClass, "getMessage", "()Ljava/lang/String;");
+			if (!midGetMessage) throw Poco::exception("This JVM implementation is broken: Throwable does not have getMessage method!");
+			jstring message = env->CallObjectMethod(exc, midGetMessage);
+			const char* msg_chars = env->GetStringUTFChars(message, 0);
+			std::string msg(msg_chars);
+			env->ReleaseStringUTFChars(message, msg_chars);
+			throw std::runtime_error(msg);
+		}
+		env->DeleteLocalRef(jtext);
+		data_res.reserve(env->GetArrayLength(jbytes));
+		jbyte* bytes = env->GetByteArrayElements(jbytes, NULL);
+		for (auto i = 0; i < env->GetArrayLength(jbytes); ++i) {
+			data_res.append(static_cast<char>(bytes[i]));
+		}
+		env->ReleaseByteArrayElements(jbytes, bytes, JNI_ABORT);
+		data = data_res.c_str();
+	}
 	#endif
 	if (!data) return "";
 	char* ptr = minitrim(data, &bufsize, bitrate, channels);

src/tts.h

@@ -32,7 +32,7 @@ class tts_voice {
 	AVTTSVoice* inst;
 	#elifdef __ANDROID__
 	jclass TTSClass;
-	jmethodID constructor, midIsActive, midIsSpeaking, midSpeak, midSilence, midGetVoice, midSetRate, midSetPitch, midSetPan, midSetVolume, midGetVoices, midSetVoice, midGetMaxSpeechInputLength, midGetPitch, midGetPan, midGetRate, midGetVolume;
+	jmethodID constructor, midIsActive, midIsSpeaking, midSpeak, midSilence, midGetVoice, midSetRate, midSetPitch, midSetPan, midSetVolume, midGetVoices, midSetVoice, midGetMaxSpeechInputLength, midGetPitch, midGetPan, midGetRate, midGetVolume, midSpeakToFile, midSpeakToMemory, midShutdown;
 	JNIEnv* env;
 	jobject TTSObj;
 	#endif