Huge quaility improvement with soxr by priming with 5ms of silence

lib/audiodev/AudioVoice.cpp

@@ -83,7 +83,8 @@ void AudioVoiceMono::_resetSampleRate(double sampleRate)
     soxr_delete(m_src);
 
     double rateOut = m_parent.mixInfo().m_sampleRate;
-    soxr_io_spec_t ioSpec = soxr_io_spec(SOXR_INT16_I, m_parent.mixInfo().m_sampleFormat);
+    soxr_datatype_t formatOut = m_parent.mixInfo().m_sampleFormat;
+    soxr_io_spec_t ioSpec = soxr_io_spec(SOXR_INT16_I, formatOut);
     soxr_quality_spec_t qSpec = soxr_quality_spec(SOXR_20_BITQ, m_dynamicRate ? SOXR_VR : 0);
 
     soxr_error_t err;
@@ -102,6 +103,37 @@ void AudioVoiceMono::_resetSampleRate(double sampleRate)
     soxr_set_input_fn(m_src, soxr_input_fn_t(SRCCallback), this, 0);
     _setPitchRatio(m_pitchRatio, false);
     m_resetSampleRate = false;
+
+    m_silentOut = true;
+    switch (formatOut)
+    {
+    case SOXR_INT16_I:
+    {
+        std::vector<int16_t>& scratch16 = m_root.m_scratch16;
+        if (scratch16.size() < m_root.m_5msFrames)
+            scratch16.resize(m_root.m_5msFrames);
+        soxr_output(m_src, scratch16.data(), m_root.m_5msFrames);
+        break;
+    }
+    case SOXR_INT32_I:
+    {
+        std::vector<int32_t>& scratch32 = m_root.m_scratch32;
+        if (scratch32.size() < m_root.m_5msFrames)
+            scratch32.resize(m_root.m_5msFrames);
+        soxr_output(m_src, scratch32.data(), m_root.m_5msFrames);
+        break;
+    }
+    case SOXR_FLOAT32_I:
+    {
+        std::vector<float>& scratchFlt = m_root.m_scratchFlt;
+        if (scratchFlt.size() < m_root.m_5msFrames)
+            scratchFlt.resize(m_root.m_5msFrames);
+        soxr_output(m_src, scratchFlt.data(), m_root.m_5msFrames);
+        break;
+    }
+    default: break;
+    }
+    m_silentOut = false;
 }
 
 size_t AudioVoiceMono::SRCCallback(AudioVoiceMono* ctx, int16_t** data, size_t frames)
@@ -110,7 +142,13 @@ size_t AudioVoiceMono::SRCCallback(AudioVoiceMono* ctx, int16_t** data, size_t f
     if (scratchIn.size() < frames)
         scratchIn.resize(frames);
     *data = scratchIn.data();
-    return ctx->m_cb->supplyAudio(*ctx, frames, scratchIn.data());
+    if (ctx->m_silentOut)
+    {
+        memset(*data, 0, frames * 2);
+        return frames;
+    }
+    else
+        return ctx->m_cb->supplyAudio(*ctx, frames, scratchIn.data());
 }
 
 size_t AudioVoiceMono::pumpAndMix(const AudioVoiceEngineMixInfo& mixInfo,
@@ -234,7 +272,8 @@ void AudioVoiceStereo::_resetSampleRate(double sampleRate)
     soxr_delete(m_src);
 
     double rateOut = m_parent.mixInfo().m_sampleRate;
-    soxr_io_spec_t ioSpec = soxr_io_spec(SOXR_INT16_I, m_parent.mixInfo().m_sampleFormat);
+    soxr_datatype_t formatOut = m_parent.mixInfo().m_sampleFormat;
+    soxr_io_spec_t ioSpec = soxr_io_spec(SOXR_INT16_I, formatOut);
     soxr_quality_spec_t qSpec = soxr_quality_spec(SOXR_20_BITQ, m_dynamicRate ? SOXR_VR : 0);
 
     soxr_error_t err;
@@ -253,6 +292,37 @@ void AudioVoiceStereo::_resetSampleRate(double sampleRate)
     soxr_set_input_fn(m_src, soxr_input_fn_t(SRCCallback), this, 0);
     _setPitchRatio(m_pitchRatio, false);
     m_resetSampleRate = false;
+
+    m_silentOut = true;
+    switch (formatOut)
+    {
+    case SOXR_INT16_I:
+    {
+        std::vector<int16_t>& scratch16 = m_root.m_scratch16;
+        if (scratch16.size() < m_root.m_5msFrames * 2)
+            scratch16.resize(m_root.m_5msFrames * 2);
+        soxr_output(m_src, scratch16.data(), m_root.m_5msFrames);
+        break;
+    }
+    case SOXR_INT32_I:
+    {
+        std::vector<int32_t>& scratch32 = m_root.m_scratch32;
+        if (scratch32.size() < m_root.m_5msFrames * 2)
+            scratch32.resize(m_root.m_5msFrames * 2);
+        soxr_output(m_src, scratch32.data(), m_root.m_5msFrames);
+        break;
+    }
+    case SOXR_FLOAT32_I:
+    {
+        std::vector<float>& scratchFlt = m_root.m_scratchFlt;
+        if (scratchFlt.size() < m_root.m_5msFrames * 2)
+            scratchFlt.resize(m_root.m_5msFrames * 2);
+        soxr_output(m_src, scratchFlt.data(), m_root.m_5msFrames);
+        break;
+    }
+    default: break;
+    }
+    m_silentOut = false;
 }
 
 size_t AudioVoiceStereo::SRCCallback(AudioVoiceStereo* ctx, int16_t** data, size_t frames)
@@ -262,7 +332,13 @@ size_t AudioVoiceStereo::SRCCallback(AudioVoiceStereo* ctx, int16_t** data, size
     if (scratchIn.size() < samples)
         scratchIn.resize(samples);
     *data = scratchIn.data();
-    return ctx->m_cb->supplyAudio(*ctx, frames, scratchIn.data());
+    if (ctx->m_silentOut)
+    {
+        memset(*data, 0, samples * 2);
+        return frames;
+    }
+    else
+        return ctx->m_cb->supplyAudio(*ctx, frames, scratchIn.data());
 }
 
 size_t AudioVoiceStereo::pumpAndMix(const AudioVoiceEngineMixInfo& mixInfo,

lib/audiodev/AudioVoice.hpp

@@ -81,6 +81,7 @@ class AudioVoiceMono : public AudioVoice
 {
     AudioMatrixMono m_matrix;
     AudioMatrixMono m_subMatrix;
+    bool m_silentOut = false;
     void _resetSampleRate(double sampleRate);
 
     static size_t SRCCallback(AudioVoiceMono* ctx,
@@ -104,6 +105,7 @@ class AudioVoiceStereo : public AudioVoice
 {
     AudioMatrixStereo m_matrix;
     AudioMatrixStereo m_subMatrix;
+    bool m_silentOut = false;
     void _resetSampleRate(double sampleRate);
 
     static size_t SRCCallback(AudioVoiceStereo* ctx,

lib/audiodev/WASAPI.cpp

@@ -28,6 +28,9 @@ struct WASAPIAudioVoiceEngine : BaseAudioVoiceEngine
     ComPtr<IAudioClient> m_audClient;
     ComPtr<IAudioRenderClient> m_renderClient;
 
+    size_t m_curBufFrame = 0;
+    std::vector<float> m_5msBuffer;
+
     struct NotificationClient : public IMMNotificationClient
     {
         WASAPIAudioVoiceEngine& m_parent;
@@ -229,6 +232,8 @@ struct WASAPIAudioVoiceEngine : BaseAudioVoiceEngine
         }
         m_mixInfo.m_sampleRate = pwfx->Format.nSamplesPerSec;
         m_5msFrames = (m_mixInfo.m_sampleRate * 5 / 500 + 1) / 2;
+        m_curBufFrame = m_5msFrames;
+        m_5msBuffer.resize(m_5msFrames * chMapOut.m_channelCount);
 
         if (pwfx->Format.wFormatTag == WAVE_FORMAT_PCM ||
             (pwfx->Format.wFormatTag == WAVE_FORMAT_EXTENSIBLE && pwfx->SubFormat == KSDATAFORMAT_SUBTYPE_PCM))
@@ -373,24 +378,26 @@ struct WASAPIAudioVoiceEngine : BaseAudioVoiceEngine
                 continue;
             }
 
-            DWORD flags = 0;
+            for (size_t f=0 ; f<frames ;)
-            switch (m_mixInfo.m_sampleFormat)
             {
-            case SOXR_INT16_I:
+                if (m_curBufFrame == m_5msFrames)
-                _pumpAndMixVoices(frames, reinterpret_cast<int16_t*>(bufOut));
+                {
-                break;
+                    _pumpAndMixVoices(m_5msFrames, m_5msBuffer.data());
-            case SOXR_INT32_I:
+                    m_curBufFrame = 0;
-                _pumpAndMixVoices(frames, reinterpret_cast<int32_t*>(bufOut));
+                }
-                break;
+
-            case SOXR_FLOAT32_I:
+                size_t remRenderFrames = std::min(frames - f, m_5msFrames - m_curBufFrame);
-                _pumpAndMixVoices(frames, reinterpret_cast<float*>(bufOut));
+                if (remRenderFrames)
-                break;
+                {
-            default:
+                    memmove(reinterpret_cast<float*>(bufOut) + m_mixInfo.m_channelMap.m_channelCount * f,
-                flags = AUDCLNT_BUFFERFLAGS_SILENT;
+                            &m_5msBuffer[m_curBufFrame * m_mixInfo.m_channelMap.m_channelCount],
-                break;
+                            remRenderFrames * m_mixInfo.m_channelMap.m_channelCount * sizeof(float));
+                    m_curBufFrame += remRenderFrames;
+                    f += remRenderFrames;
+                }
             }
 
-            res = m_renderClient->ReleaseBuffer(frames, flags);
+            res = m_renderClient->ReleaseBuffer(frames, 0);
             if (FAILED(res))
             {
                 m_rebuild = true;