Low bandwidth mode

15 files changed, 251 insertions, 96 deletions

diff --git a/src/Message.h b/src/Message.h
index 2f4291c34..4eee97d69 100644
--- a/src/Message.h
+++ b/src/Message.h
@@ -58,7 +58,7 @@
 
 class MessageHandler {
 	public:
-		enum UDPMessageType { UDPVoice, UDPPing };
+		enum UDPMessageType { UDPVoiceCELT, UDPPing, UDPVoiceSpeex };
 
 #define MUMBLE_MH_MSG(x) x,
 		enum MessageType {
diff --git a/src/mumble/Audio.cpp b/src/mumble/Audio.cpp
index 6c266d788..767968e65 100644
--- a/src/mumble/Audio.cpp
+++ b/src/mumble/Audio.cpp
@@ -77,7 +77,8 @@ void LoopUser::addFrame(const QByteArray &packet) {
 	if (qtLastFetch.elapsed() > 100) {
 		AudioOutputPtr ao = g.ao;
 		if (ao) {
-			ao->addFrameToBuffer(this, QByteArray(), 0);
+			MessageHandler::UDPMessageType msgType = static_cast<MessageHandler::UDPMessageType>((packet.at(0) >> 5) & 0x7);
+			ao->addFrameToBuffer(this, QByteArray(), 0, msgType);
 		}
 	}
 
@@ -87,8 +88,9 @@ void LoopUser::fetchFrames() {
 	QMutexLocker l(&qmLock);
 
 	AudioOutputPtr ao = g.ao;
-	if (!ao || qmPackets.isEmpty())
+	if (!ao || qmPackets.isEmpty()) {
 		return;
+	}
 
 	double cmp = qtTicker.elapsed();
 
@@ -111,7 +113,9 @@ void LoopUser::fetchFrames() {
 		qba.append(static_cast<char>(msgFlags));
 		qba.append(pds.dataBlock(pds.left()));
 
-		ao->addFrameToBuffer(this, qba, iSeq);
+		MessageHandler::UDPMessageType msgType = static_cast<MessageHandler::UDPMessageType>((msgFlags >> 5) & 0x7);
+
+		ao->addFrameToBuffer(this, qba, iSeq, msgType);
 		i = qmPackets.erase(i);
 	}
 
diff --git a/src/mumble/AudioInput.cpp b/src/mumble/AudioInput.cpp
index 611046fa1..577f2276c 100644
--- a/src/mumble/AudioInput.cpp
+++ b/src/mumble/AudioInput.cpp
@@ -84,9 +84,51 @@ AudioInputPtr AudioInputRegistrar::newFromChoice(QString choice) {
 
 
 AudioInput::AudioInput() {
-	iFrameSize = SAMPLE_RATE / 100;
-	cmMode = celt_mode_create(SAMPLE_RATE, 1, iFrameSize, NULL);
-	ceEncoder = celt_encoder_create(cmMode);
+	adjustBandwidth(g.iMaxBandwidth, iAudioQuality, iAudioFrames);
+
+	if (preferCELT(iAudioQuality, iAudioFrames))
+		umtType = MessageHandler::UDPVoiceCELT;
+	else
+		umtType = MessageHandler::UDPVoiceSpeex;
+
+	if (umtType == MessageHandler::UDPVoiceCELT) {
+		iSampleRate = SAMPLE_RATE;
+		iFrameSize = SAMPLE_RATE / 100;
+
+		cmMode = celt_mode_create(SAMPLE_RATE, 1, iFrameSize, NULL);
+		ceEncoder = celt_encoder_create(cmMode);
+		esSpeex = NULL;
+		qWarning("AudioInput: %d bits/s, %d hz, %d sample CELT", iAudioQuality, iSampleRate, iFrameSize);
+	} else {
+		cmMode = NULL;
+		ceEncoder = NULL;
+
+		iAudioFrames /= 2;
+
+        speex_bits_init(&sbBits);
+        speex_bits_reset(&sbBits);
+		esSpeex = speex_encoder_init(&speex_uwb_mode);
+		speex_encoder_ctl(esSpeex,SPEEX_GET_FRAME_SIZE,&iFrameSize);
+		speex_encoder_ctl(esSpeex,SPEEX_GET_SAMPLING_RATE,&iSampleRate);
+
+		int iArg=1;
+		speex_encoder_ctl(esSpeex,SPEEX_SET_VBR, &iArg);
+
+		iArg = 0;
+		speex_encoder_ctl(esSpeex,SPEEX_SET_VAD, &iArg);
+		speex_encoder_ctl(esSpeex,SPEEX_SET_DTX, &iArg);
+
+		float fArg=8.0;
+		speex_encoder_ctl(esSpeex,SPEEX_SET_VBR_QUALITY, &fArg);
+
+		iArg = iAudioQuality;
+		speex_encoder_ctl(esSpeex, SPEEX_SET_VBR_MAX_BITRATE, &iArg);
+
+		iArg = 5;
+		speex_encoder_ctl(esSpeex,SPEEX_SET_COMPLEXITY, &iArg);
+		qWarning("AudioInput: %d bits/s, %d hz, %d sample Speex-UWB", iAudioQuality, iSampleRate, iFrameSize);
+	}
+	iEchoFreq = iMicFreq = iSampleRate;
 
 	mumble_drft_init(&fftTable, iFrameSize);
 
@@ -115,7 +157,6 @@ AudioInput::AudioInput() {
 	psSpeaker = NULL;
 
 	iEchoChannels = iMicChannels = 0;
-	iEchoFreq = iMicFreq = SAMPLE_RATE;
 	iEchoFilled = iMicFilled = 0;
 	eMicFormat = eEchoFormat = SampleFloat;
 	iMicSampleSize = iEchoSampleSize = 0;
@@ -133,11 +174,6 @@ AudioInput::AudioInput() {
 
 	bRunning = false;
 
-	if (g.sh && g.sh->isRunning())
-		setMaxBandwidth(g.iMaxBandwidth);
-	else
-		iAudioQuality = g.s.iQuality;
-
 	connect(this, SIGNAL(doMute()), g.mw->qaAudioMute, SLOT(trigger()), Qt::QueuedConnection);
 }
 
@@ -145,8 +181,13 @@ AudioInput::~AudioInput() {
 	bRunning = false;
 	wait();
 
-	celt_encoder_destroy(ceEncoder);
-	celt_mode_destroy(cmMode);
+	if (umtType == MessageHandler::UDPVoiceCELT) {
+		celt_encoder_destroy(ceEncoder);
+		celt_mode_destroy(cmMode);
+	} else {
+		speex_bits_destroy(&sbBits);
+		speex_encoder_destroy(esSpeex);
+	}
 
 	mumble_drft_clear(&fftTable);
 	jitter_buffer_destroy(jb);
@@ -300,10 +341,10 @@ void AudioInput::initializeMixer() {
 	if (pfOutput)
 		delete [] pfOutput;
 
-	if (iMicFreq != SAMPLE_RATE)
-		srsMic = speex_resampler_init(1, iMicFreq, SAMPLE_RATE, 3, &err);
+	if (iMicFreq != iSampleRate)
+		srsMic = speex_resampler_init(1, iMicFreq, iSampleRate, 3, &err);
 
-	iMicLength = (iFrameSize * iMicFreq) / SAMPLE_RATE;
+	iMicLength = (iFrameSize * iMicFreq) / iSampleRate;
 
 	pfMicInput = new float[iMicLength];
 	pfOutput = new float[iFrameSize * qMax(1U,iEchoChannels)];
@@ -311,9 +352,9 @@ void AudioInput::initializeMixer() {
 	if (iEchoChannels > 0) {
 		psSpeaker = new short[iFrameSize * iEchoChannels];
 		bEchoMulti = g.s.bEchoMulti;
-		if (iEchoFreq != SAMPLE_RATE)
-			srsEcho = speex_resampler_init(bEchoMulti ? iEchoChannels : 1, iEchoFreq, SAMPLE_RATE, 3, &err);
-		iEchoLength = (iFrameSize * iEchoFreq) / SAMPLE_RATE;
+		if (iEchoFreq != iSampleRate)
+			srsEcho = speex_resampler_init(bEchoMulti ? iEchoChannels : 1, iEchoFreq, iSampleRate, 3, &err);
+		iEchoLength = (iFrameSize * iEchoFreq) / iSampleRate;
 		iEchoMCLength = bEchoMulti ? iEchoLength * iEchoChannels : iEchoLength;
 		iEchoFrameSize = bEchoMulti ? iFrameSize * iEchoChannels : iFrameSize;
 		pfEchoInput = new float[iEchoMCLength];
@@ -433,31 +474,69 @@ void AudioInput::addEcho(const void *data, unsigned int nsamp) {
 	}
 }
 
-int AudioInput::getMaxBandwidth() {
-	// Overhead
-	int overhead = 20 + 8 + 4 + 1 + 2;
-
-	if (g.s.bTransmitPosition)
-		overhead += 12;
+bool AudioInput::preferCELT(int bitrate, int frames) {
+	return ((bitrate >= 32000) || (frames == 1));
+}
 
-	if (NetworkConfig::TcpModeEnabled())
-		overhead += 12;
+void AudioInput::adjustBandwidth(int bitspersec, int &bitrate, int &frames) {
+	frames = g.s.iFramesPerPacket;
+	bitrate = g.s.iQuality;
 
-	return (g.s.iQuality / 8)  + (overhead * 100/g.s.iFramesPerPacket);
+	if (bitspersec == -1) {
+		// No limit
+	} else {
+		if (getNetworkBandwidth(bitrate, frames) >= bitspersec) {
+			if ((frames <= 4) && (bitspersec <= 32000))
+				frames = 4;
+			else if ((frames == 1) && (bitspersec <= 64000))
+				frames = 2;
+			else if ((frames == 2) && (bitspersec <= 48000))
+				frames = 4;
+			if (getNetworkBandwidth(bitrate, frames) >= bitspersec) {
+				do {
+					bitrate -= 1000;
+				} while ((bitrate > 8000) && (getNetworkBandwidth(bitrate, frames) >= bitspersec));
+			}
+		}
+	}
 }
 
-void AudioInput::setMaxBandwidth(int bytespersec) {
-	int overhead = 20 + 8 + 4 + 1 + 2 + (g.s.bTransmitPosition ? 12 : 0) + (NetworkConfig::TcpModeEnabled() ? 12 : 0);
-	overhead *= (100 / g.s.iFramesPerPacket);
+void AudioInput::setMaxBandwidth(int bitspersec) {
+	if (bitspersec == g.iMaxBandwidth)
+		return;
 
-	bytespersec -= overhead;
+	int frames;
+	int bitrate;
+	adjustBandwidth(bitspersec, bitrate, frames);
+
+	g.iMaxBandwidth = bitspersec;
+	g.iAudioBandwidth = bitrate;
+
+	if (bitspersec != -1) {
+		if ((bitrate != g.s.iQuality) || (frames != g.s.iFramesPerPacket))
+			g.mw->msgBox(tr("Server maximum network bandwidth is only %1 kbit/s. Audio quality auto-adjusted to %2 kbit/s (%3ms)").arg(bitspersec / 1000).arg(bitrate / 10000).arg(frames*10));
+	}
+
+	AudioInputPtr ai = g.ai;
+	if (ai && (preferCELT(bitrate, frames) == (ai->umtType == MessageHandler::UDPVoiceCELT))) {
+		ai->iAudioQuality = bitrate;
+		ai->iAudioFrames = frames;
+		return;
+	}
+	g.ai.reset();
+	while (! ai.unique()) {}
+	ai.reset();
+	g.ai = AudioInputRegistrar::newFromChoice(g.s.qsAudioInput);
+	if (g.ai)
+		g.ai->start(QThread::HighestPriority);
+}
 
-	int target = qMin(bytespersec * 8, g.s.iQuality);
-	if (target < 16000)
-		target = 16000;
+int AudioInput::getNetworkBandwidth(int bitrate, int frames) {
+	int overhead = 20 + 8 + 4 + 1 + 2 + (g.s.bTransmitPosition ? 12 : 0) + (NetworkConfig::TcpModeEnabled() ? 12 : 0);
+	overhead *= (800 / frames);
+	int bw = overhead + bitrate;
 
-	g.iAudioBandwidth = (target/8 + overhead);
-	iAudioQuality = target;
+	return bw;
 }
 
 
@@ -558,7 +637,7 @@ void AudioInput::encodeAudioFrame() {
 		if (sesEcho)
 			speex_echo_state_destroy(sesEcho);
 
-		sppPreprocess = speex_preprocess_state_init(iFrameSize, SAMPLE_RATE);
+		sppPreprocess = speex_preprocess_state_init(iFrameSize, iSampleRate);
 
 		iArg = 1;
 		speex_preprocess_ctl(sppPreprocess, SPEEX_PREPROCESS_SET_VAD, &iArg);
@@ -578,7 +657,7 @@ void AudioInput::encodeAudioFrame() {
 
 		if (iEchoChannels > 0) {
 			sesEcho = speex_echo_state_init_mc(iFrameSize, iFrameSize*10, 1, bEchoMulti ? iEchoChannels : 1);
-			iArg = SAMPLE_RATE;
+			iArg = iSampleRate;
 			speex_echo_ctl(sesEcho, SPEEX_ECHO_SET_SAMPLING_RATE, &iArg);
 			speex_preprocess_ctl(sppPreprocess, SPEEX_PREPROCESS_SET_ECHO_STATE, sesEcho);
 
@@ -676,15 +755,27 @@ void AudioInput::encodeAudioFrame() {
 	*/
 
 	unsigned char buffer[512];
+	int len;
 
-	celt_encoder_ctl(ceEncoder,CELT_SET_VBR_RATE(iAudioQuality));
-
+	if (umtType == MessageHandler::UDPVoiceCELT) {
+		celt_encoder_ctl(ceEncoder,CELT_SET_VBR_RATE(iAudioQuality));
+		len = celt_encode(ceEncoder, psSource, NULL, buffer, qMin(iAudioQuality / 800, 127));
+	} else {
+		int vbr = 0;
+		speex_encoder_ctl(esSpeex, SPEEX_GET_VBR_MAX_BITRATE, &vbr);
+		if (vbr != iAudioQuality) {
+			vbr = iAudioQuality;
+			speex_encoder_ctl(esSpeex, SPEEX_SET_VBR_MAX_BITRATE, &vbr);
+		}
 
-	int len = celt_encode(ceEncoder, psSource, NULL, buffer, qMin(iAudioQuality / 800, 127));
+		speex_encode_int(esSpeex, psSource, &sbBits);
+		len = speex_bits_write(&sbBits, reinterpret_cast<char *>(buffer), 127);
+		speex_bits_reset(&sbBits);
+	}
 
 	iBitrate = len * 100 * 8;
-
 	flushCheck(QByteArray(reinterpret_cast<const char *>(buffer), len), ! iIsSpeech);
+
 	if (! iIsSpeech)
 		iBitrate = 0;
 
@@ -693,13 +784,15 @@ void AudioInput::encodeAudioFrame() {
 
 void AudioInput::flushCheck(const QByteArray &qba, bool terminator) {
 	qlFrames << qba;
-	if (! terminator && qlFrames.count() < g.s.iFramesPerPacket)
+	if (! terminator && qlFrames.count() < iAudioFrames)
 		return;
 
 	int flags = g.iTarget;
 	if (g.s.lmLoopMode == Settings::Server)
 		flags = 0x1f;
 
+	flags |= (umtType << 5);
+
 	char data[1024];
 	data[0] = static_cast<unsigned char>(flags);
 
diff --git a/src/mumble/AudioInput.h b/src/mumble/AudioInput.h
index 9740a78fd..a4375c202 100644
--- a/src/mumble/AudioInput.h
+++ b/src/mumble/AudioInput.h
@@ -34,6 +34,7 @@
 #include "Audio.h"
 #include "Settings.h"
 #include "Timer.h"
+#include "Message.h"
 #include "smallft.h"
 
 class AudioInput;
@@ -63,11 +64,11 @@ class AudioInput : public QThread {
 		friend class AudioEchoWidget;
 		friend class AudioStats;
 		friend class AudioInputDialog;
-		friend class FMODSystem;
 	private:
 		Q_OBJECT
 		Q_DISABLE_COPY(AudioInput)
 	protected:
+		typedef enum { CodecCELT, CodecSpeex } CodecFormat;
 		typedef enum { SampleShort, SampleFloat } SampleFormat;
 		typedef void (*inMixerFunc)(float * RESTRICT, const void * RESTRICT, unsigned int, unsigned int);
 	private:
@@ -78,8 +79,10 @@ class AudioInput : public QThread {
 		inMixerFunc imfMic, imfEcho;
 		inMixerFunc chooseMixer(const unsigned int nchan, SampleFormat sf);
 	protected:
+		MessageHandler::UDPMessageType umtType;
 		SampleFormat eMicFormat, eEchoFormat;
 
+		unsigned int iSampleRate;
 		unsigned int iMicChannels, iEchoChannels;
 		unsigned int iMicFreq, iEchoFreq;
 		unsigned int iMicLength, iEchoLength;
@@ -95,7 +98,12 @@ class AudioInput : public QThread {
 
 		CELTMode *cmMode;
 		CELTEncoder *ceEncoder;
+
+		SpeexBits sbBits;
+		void *esSpeex;
+
 		int iAudioQuality;
+		int iAudioFrames;
 
 		drft_lookup fftTable;
 
@@ -122,6 +130,9 @@ class AudioInput : public QThread {
 		void flushCheck(const QByteArray &, bool terminator);
 
 		void initializeMixer();
+
+		static bool preferCELT(int bitrate, int frames);
+		static void adjustBandwidth(int bitspersec, int &bitrate, int &frames);
 	signals:
 		void doMute();
 	public:
@@ -137,8 +148,8 @@ class AudioInput : public QThread {
 			return bPreviousVoice;
 		};
 
-		static int getMaxBandwidth();
-		void setMaxBandwidth(int bytespersec);
+		static int getNetworkBandwidth(int bitrate, int frames);
+		static void setMaxBandwidth(int bitspersec);
 
 		AudioInput();
 		~AudioInput();
diff --git a/src/mumble/AudioOutput.cpp b/src/mumble/AudioOutput.cpp
index 834f35d2a..e81b5509a 100644
--- a/src/mumble/AudioOutput.cpp
+++ b/src/mumble/AudioOutput.cpp
@@ -345,20 +345,39 @@ bool AudioOutputSample::needSamples(unsigned int snum) {
 	return !eof;
 }
 
-AudioOutputSpeech::AudioOutputSpeech(ClientUser *user, unsigned int freq) : AudioOutputUser(user->qsName) {
+AudioOutputSpeech::AudioOutputSpeech(ClientUser *user, unsigned int freq, MessageHandler::UDPMessageType type) : AudioOutputUser(user->qsName) {
 	int err;
 	p = user;
+	umtType = type;
 
-	iFrameSize = SAMPLE_RATE / 100;
-	cmMode = celt_mode_create(SAMPLE_RATE, 1, iFrameSize, NULL);
-	cdDecoder = celt_decoder_create(cmMode);
+	int srate = 0;
+
+	if (umtType == MessageHandler::UDPVoiceCELT) {
+		srate = SAMPLE_RATE;
+		iFrameSize = srate / 100;
+		cmMode = celt_mode_create(srate, 1, iFrameSize, NULL);
+		cdDecoder = celt_decoder_create(cmMode);
+
+		dsSpeex = NULL;
+	} else {
+		cmMode = NULL;
+		cdDecoder = NULL;
+
+		speex_bits_init(&sbBits);
+
+		dsSpeex = speex_decoder_init(&speex_uwb_mode);
+		int iArg=1;
+		speex_decoder_ctl(dsSpeex, SPEEX_SET_ENH, &iArg);
+		speex_decoder_ctl(dsSpeex, SPEEX_GET_FRAME_SIZE, &iFrameSize);
+		speex_decoder_ctl(dsSpeex, SPEEX_GET_SAMPLING_RATE, &srate);
+	}
 
-	if (freq != SAMPLE_RATE)
-		srs = speex_resampler_init(1, SAMPLE_RATE, freq, 3, &err);
+	if (freq != srate)
+		srs = speex_resampler_init(1, srate, freq, 3, &err);
 	else
 		srs = NULL;
 
-	iOutputSize = iroundf(ceilf(static_cast<float>(iFrameSize * freq) / static_cast<float>(SAMPLE_RATE)));
+	iOutputSize = iroundf(ceilf(static_cast<float>(iFrameSize * freq) / static_cast<float>(srate)));
 
 	iBufferOffset = iBufferFilled = iLastConsume = 0;
 	bLastAlive = true;
@@ -379,8 +398,13 @@ AudioOutputSpeech::AudioOutputSpeech(ClientUser *user, unsigned int freq) : Audi
 }
 
 AudioOutputSpeech::~AudioOutputSpeech() {
-	celt_decoder_destroy(cdDecoder);
-	celt_mode_destroy(cmMode);
+	if (umtType == MessageHandler::UDPVoiceCELT) {
+		celt_decoder_destroy(cdDecoder);
+		celt_mode_destroy(cmMode);
+	} else {
+		speex_bits_destroy(&sbBits);
+		speex_decoder_destroy(dsSpeex);
+	}
 
 	jitter_buffer_destroy(jbJitter);
 
@@ -484,18 +508,39 @@ bool AudioOutputSpeech::needSamples(unsigned int snum) {
 						nextalive = false;
 				}
 
-				jitter_buffer_update_delay(jbJitter, &jbp, NULL);
+				int activity;
+				if (umtType == MessageHandler::UDPVoiceCELT)
+					activity = 0;
+				else
+					speex_decoder_ctl(dsSpeex, SPEEX_GET_ACTIVITY, &activity);
+
+				if (activity < 30)
+					jitter_buffer_update_delay(jbJitter, &jbp, NULL);
 			}
 
 			jitter_buffer_tick(jbJitter);
 
 			if (! qlFrames.isEmpty()) {
 				const QByteArray &qba = qlFrames.takeFirst();
-				celt_decode_float(cdDecoder, reinterpret_cast<const unsigned char *>(qba.data()), qba.size(), pOut);
+				if (umtType == MessageHandler::UDPVoiceCELT)
+					celt_decode_float(cdDecoder, reinterpret_cast<const unsigned char *>(qba.constData()), qba.size(), pOut);
+				else {
+					speex_bits_read_from(&sbBits, qba.constData(), qba.size());
+					speex_decode(dsSpeex, &sbBits, pOut);
+					for(unsigned int i=0;i<iFrameSize;++i)
+						pOut[i] *= (1.0f / 32767.f);
+				}
+
 				if (qlFrames.isEmpty() && bHasTerminator)
 					nextalive = false;
 			} else {
-				celt_decode_float(cdDecoder, NULL, 0, pOut);
+				if (umtType == MessageHandler::UDPVoiceCELT)
+					celt_decode_float(cdDecoder, NULL, 0, pOut);
+				else {
+					speex_decode(dsSpeex, NULL, pOut);
+					for(unsigned int i=0;i<iFrameSize;++i)
+						pOut[i] *= (1.0f / 32767.f);
+				}
 			}
 
 			if (! nextalive) {
@@ -618,22 +663,25 @@ AudioOutputSample *AudioOutput::playSample(const QString &filename, bool loop) {
 
 }
 
-void AudioOutput::addFrameToBuffer(ClientUser *user, const QByteArray &qbaPacket, unsigned int iSeq) {
+void AudioOutput::addFrameToBuffer(ClientUser *user, const QByteArray &qbaPacket, unsigned int iSeq, MessageHandler::UDPMessageType type) {
 	if (iChannels == 0)
 		return;
 	qrwlOutputs.lockForRead();
 	AudioOutputSpeech *aop = dynamic_cast<AudioOutputSpeech *>(qmOutputs.value(user));
 
-	if (! aop) {
+	if (! aop || (aop->umtType != type)) {
 		qrwlOutputs.unlock();
 
+		if (aop)
+			removeBuffer(aop);
+
 		while ((iMixerFreq == 0) && bRunning) {}
 
 		if (! iMixerFreq)
 			return;
 
 		qrwlOutputs.lockForWrite();
-		aop = new AudioOutputSpeech(user, iMixerFreq);
+		aop = new AudioOutputSpeech(user, iMixerFreq, type);
 		qmOutputs.replace(user,aop);
 	}
 
diff --git a/src/mumble/AudioOutput.h b/src/mumble/AudioOutput.h
index 94feb3e9f..9d2317afa 100644
--- a/src/mumble/AudioOutput.h
+++ b/src/mumble/AudioOutput.h
@@ -60,6 +60,7 @@
 
 #include "Audio.h"
 #include "Settings.h"
+#include "Message.h"
 #include "smallft.h"
 
 class AudioOutput;
@@ -111,6 +112,7 @@ class AudioOutputSpeech : public AudioOutputUser {
 		Q_OBJECT
 		Q_DISABLE_COPY(AudioOutputSpeech)
 	protected:
+		MessageHandler::UDPMessageType umtType;
 		unsigned int iBufferOffset;
 		unsigned int iBufferFilled;
 		unsigned int iOutputSize;
@@ -131,6 +133,9 @@ class AudioOutputSpeech : public AudioOutputUser {
 		CELTMode *cmMode;
 		CELTDecoder *cdDecoder;
 
+		SpeexBits sbBits;
+		void *dsSpeex;
+
 		QList<QByteArray> qlFrames;
 
 		unsigned char ucFlags;
@@ -141,7 +146,7 @@ class AudioOutputSpeech : public AudioOutputUser {
 		virtual bool needSamples(unsigned int snum);
 
 		void addFrameToBuffer(const QByteArray &, unsigned int iBaseSeq);
-		AudioOutputSpeech(ClientUser *, unsigned int freq);
+		AudioOutputSpeech(ClientUser *, unsigned int freq, MessageHandler::UDPMessageType type);
 		~AudioOutputSpeech();
 };
 
@@ -220,7 +225,7 @@ class AudioOutput : public QThread {
 
 		AudioOutput();
 		~AudioOutput();
-		void addFrameToBuffer(ClientUser *, const QByteArray &, unsigned int iSeq);
+		void addFrameToBuffer(ClientUser *, const QByteArray &, unsigned int iSeq, MessageHandler::UDPMessageType type);
 		void removeBuffer(const ClientUser *);
 		AudioOutputSample *playSample(const QString &filename, bool loop = false);
 		void run() = 0;
diff --git a/src/mumble/Global.cpp b/src/mumble/Global.cpp
index c9c1700e5..a02f2928d 100644
--- a/src/mumble/Global.cpp
+++ b/src/mumble/Global.cpp
@@ -46,7 +46,7 @@ Global::Global() {
 	bCenterPosition = false;
 	bPosTest = false;
 	iAudioPathTime = 0;
-	iMaxBandwidth = 0;
+	iMaxBandwidth = -1;
 
 	QString apppath = QCoreApplication::instance()->applicationDirPath();
 	QFile inifile(QString::fromLatin1("%1/mumble.ini").arg(apppath));
diff --git a/src/mumble/MainWindow.cpp b/src/mumble/MainWindow.cpp
index 4c8f3e88c..a0e7f6b6b 100644
--- a/src/mumble/MainWindow.cpp
+++ b/src/mumble/MainWindow.cpp
@@ -693,7 +693,7 @@ void MainWindow::on_qaServerInformation_triggered() {
 		          .arg(cs.uiRemoteGood).arg(cs.uiRemoteLate).arg(cs.uiRemoteLost).arg(cs.uiRemoteResync)
 		          .arg(cs.uiGood).arg(cs.uiLate).arg(cs.uiLost).arg(cs.uiResync);
 	}
-	qsAudio=tr("<h2>Audio bandwidth</h2><p>Maximum %1 kbit/s<br />Current %2 kbit/s</p>").arg(g.iMaxBandwidth / 125.0,0,'f',1).arg(g.iAudioBandwidth / 125.0,0,'f',1);
+	qsAudio=tr("<h2>Audio bandwidth</h2><p>Maximum %1 kbit/s<br />Current %2 kbit/s</p>").arg(g.iMaxBandwidth / 1000.0,0,'f',1).arg(g.iAudioBandwidth / 1000.0,0,'f',1);
 
 	QMessageBox qmb(QMessageBox::Information, tr("Mumble Server Information"), qsVersion + qsControl + qsVoice + qsCrypt + qsAudio, QMessageBox::Ok, this);
 	qmb.setDefaultButton(QMessageBox::Ok);
@@ -1739,6 +1739,7 @@ void MainWindow::serverDisconnected(QString reason) {
 			qtReconnect->start();
 		}
 	}
+	AudioInput::setMaxBandwidth(-1);
 }
 
 void MainWindow::on_Icon_activated(QSystemTrayIcon::ActivationReason reason) {
@@ -1757,14 +1758,13 @@ void MainWindow::customEvent(QEvent *evt) {
 	if (evt->type() == TI_QEVENT) {
 		hide();
 		return;
-	}
-	if (evt->type() == MB_QEVENT) {
+	} else if (evt->type() == MB_QEVENT) {
 		MessageBoxEvent *mbe=static_cast<MessageBoxEvent *>(evt);
 		g.l->log(Log::Information, mbe->msg);
 		return;
-	}
-	if (evt->type() != SERVERSEND_EVENT)
+	} else if (evt->type() != SERVERSEND_EVENT) {
 		return;
+	}
 
 	ServerHandlerMessageEvent *shme=static_cast<ServerHandlerMessageEvent *>(evt);
 
diff --git a/src/mumble/Messages.cpp b/src/mumble/Messages.cpp
index f80b35831..6e58f16bb 100644
--- a/src/mumble/Messages.cpp
+++ b/src/mumble/Messages.cpp
@@ -88,7 +88,6 @@ void MainWindow::msgReject(const MumbleProto::Reject &msg) {
 }
 
 void MainWindow::msgServerSync(const MumbleProto::ServerSync &msg) {
-	g.iMaxBandwidth = msg.max_bandwidth();
 	g.uiSession = msg.session();
 	g.pPermissions = static_cast<ChanACL::Permissions>(msg.permissions());
 	g.l->clearIgnore();
@@ -103,14 +102,7 @@ void MainWindow::msgServerSync(const MumbleProto::ServerSync &msg) {
 	}
 	iTargetCounter = 100;
 
-	AudioInputPtr ai = g.ai;
-	if (ai) {
-		int bw = ai->getMaxBandwidth();
-		if (bw > g.iMaxBandwidth) {
-			g.l->log(Log::Information, tr("Server maximum bandwidth is only %1 kbit/s. Quality auto-adjusted.").arg(g.iMaxBandwidth / 125));
-		}
-		ai->setMaxBandwidth(g.iMaxBandwidth);
-	}
+	AudioInput::setMaxBandwidth(msg.max_bandwidth());
 
 	findDesiredChannel();
 
diff --git a/src/mumble/ServerHandler.cpp b/src/mumble/ServerHandler.cpp
index a083845ea..1f7beff68 100644
--- a/src/mumble/ServerHandler.cpp
+++ b/src/mumble/ServerHandler.cpp
@@ -150,20 +150,22 @@ void ServerHandler::udpReady() {
 
 		PacketDataStream pds(buffer + 1, buflen-5);
 
-		unsigned int msgType = (buffer[0] >> 5) & 0x7;
+		MessageHandler::UDPMessageType msgType = static_cast<MessageHandler::UDPMessageType>((buffer[0] >> 5) & 0x7);
 		unsigned int msgFlags = buffer[0] & 0x1f;
 
 		if (msgType == MessageHandler::UDPPing) {
 			quint64 t;
 			pds >> t;
 			accUDP(static_cast<double>(tTimestamp.elapsed() - t) / 1000.0);
-		} else if (msgType == MessageHandler::UDPVoice) {
-			handleVoicePacket(msgFlags, pds);
+		} else if (msgType == MessageHandler::UDPVoiceCELT) {
+			handleVoicePacket(msgFlags, pds, msgType);
+		} else if (msgType == MessageHandler::UDPVoiceSpeex) {
+			handleVoicePacket(msgFlags, pds, msgType);
 		}
 	}
 }
 
-void ServerHandler::handleVoicePacket(unsigned int msgFlags, PacketDataStream &pds) {
+void ServerHandler::handleVoicePacket(unsigned int msgFlags, PacketDataStream &pds, MessageHandler::UDPMessageType type) {
 	unsigned int uiSession;
 	pds >> uiSession;
 	ClientUser *p = ClientUser::get(uiSession);
@@ -176,7 +178,7 @@ void ServerHandler::handleVoicePacket(unsigned int msgFlags, PacketDataStream &p
 		qba.append(static_cast<char>(msgFlags));
 		qba.append(pds.dataBlock(pds.left()));
 
-		ao->addFrameToBuffer(p, qba, iSeq);
+		ao->addFrameToBuffer(p, qba, iSeq, type);
 	}
 }
 
@@ -330,13 +332,14 @@ void ServerHandler::message(unsigned int msgType, const QByteArray &qbaMsg) {
 		if (qbaMsg.length() < 1)
 			return;
 
-		unsigned int msgUDPType = (ptr[0] >> 5) & 0x7;
+		MessageHandler::UDPMessageType msgType = static_cast<MessageHandler::UDPMessageType>((ptr[0] >> 5) & 0x7);
 		unsigned int msgFlags = ptr[0] & 0x1f;
 		PacketDataStream pds(qbaMsg.constData() + 1, qbaMsg.size());
 
-		if (msgUDPType == MessageHandler::UDPVoice) {
-			handleVoicePacket(msgFlags, pds);
-		}
+		if (msgType == MessageHandler::UDPVoiceCELT)
+			handleVoicePacket(msgFlags, pds, msgType);
+		else if (msgType == MessageHandler::UDPVoiceSpeex)
+			handleVoicePacket(msgFlags, pds, msgType);
 	} else if (msgType == MessageHandler::Ping) {
 		MumbleProto::Ping msg;
 		if (msg.ParseFromArray(qbaMsg.constData(), qbaMsg.size())) {
diff --git a/src/mumble/ServerHandler.h b/src/mumble/ServerHandler.h
index 822e129a7..4ca7b8810 100644
--- a/src/mumble/ServerHandler.h
+++ b/src/mumble/ServerHandler.h
@@ -72,7 +72,7 @@ class ServerHandler : public QThread {
 		QUdpSocket *qusUdp;
 		QMutex qmUdp;
 
-		void handleVoicePacket(unsigned int msgFlags, PacketDataStream &pds);
+		void handleVoicePacket(unsigned int msgFlags, PacketDataStream &pds, MessageHandler::UDPMessageType type);
 	public:
 		Timer tTimestamp;
 		QList<QSslError> qlErrors;
diff --git a/src/mumble/Settings.cpp b/src/mumble/Settings.cpp
index 722d01056..bde7125de 100644
--- a/src/mumble/Settings.cpp
+++ b/src/mumble/Settings.cpp
@@ -217,7 +217,7 @@ Settings::Settings() {
 	iMaxImageHeight = 1024;
 	bSuppressIdentity = false;
 
-#if defined(AUDIO_TEST)
+#if defined(AUDIO_TEST) || 1
 	lmLoopMode = Local;
 #else
 	lmLoopMode = None;
diff --git a/src/mumble/WASAPI.cpp b/src/mumble/WASAPI.cpp
index 2eaab8073..6bd36e409 100644
--- a/src/mumble/WASAPI.cpp
+++ b/src/mumble/WASAPI.cpp
@@ -547,7 +547,6 @@ void WASAPIOutput::setVolumes(IMMDevice *pDevice, bool talking) {
 			if (SUCCEEDED(hr = pEnumerator->GetCount(&max))) {
 				for (int i=0;i<max;++i) {
 					IAudioSessionControl *pControl = NULL;
-					IUnknown *pUnknown = NULL;
 					if (SUCCEEDED(hr = pEnumerator->GetSession(i, &pControl))) {
 						IAudioSessionControl2 *pControl2 = NULL;
 						if (SUCCEEDED(hr = pControl->QueryInterface(version ? __uuidof(IAudioSessionControl2) : __uuidof(IVistaAudioSessionControl2), (void **) &pControl2)))  {
diff --git a/src/mumble/mumble.pro b/src/mumble/mumble.pro
index 97ca61a9d..e510a7241 100644
--- a/src/mumble/mumble.pro
+++ b/src/mumble/mumble.pro
@@ -23,7 +23,7 @@ unix:!CONFIG(bundled-speex):system(pkg-config --atleast-version=1.2 speexdsp) {
 }
 
 CONFIG(no-bundled-speex) {
-  PKGCONFIG	*= speexdsp
+  PKGCONFIG	*= speex speexdsp
 }
 
 !CONFIG(no-bundled-speex) {
diff --git a/src/mumble/mumble_pch.hpp b/src/mumble/mumble_pch.hpp
index 030d2c530..18c1bc2c6 100644
--- a/src/mumble/mumble_pch.hpp
+++ b/src/mumble/mumble_pch.hpp
@@ -51,10 +51,10 @@
 #include <celt_header.h>
 #else
 #include <ogg/ogg.h>
-#include <speex/speex.h>
-#include <speex/speex_header.h>
 #include <speex/speex_callbacks.h>
 #endif
+#include <speex/speex.h>
+#include <speex/speex_header.h>
 #include <speex/speex_jitter.h>
 #include <speex/speex_preprocess.h>
 #include <speex/speex_echo.h>