Merge pull request #5241 from MerryMage/stretch

Pitch-Preserving Audio Stretching

1 files changed, 117 insertions, 18 deletions

diff --git a/Source/Core/AudioCommon/Mixer.cpp b/Source/Core/AudioCommon/Mixer.cpp
index bdaaad812c..8149da088c 100644
--- a/Source/Core/AudioCommon/Mixer.cpp
+++ b/Source/Core/AudioCommon/Mixer.cpp
@@ -4,6 +4,7 @@
 
 #include "AudioCommon/Mixer.h"
 
+#include <cmath>
 #include <cstring>
 
 #include "Common/CommonTypes.h"
@@ -15,6 +16,15 @@
 CMixer::CMixer(unsigned int BackendSampleRate) : m_sampleRate(BackendSampleRate)
 {
   INFO_LOG(AUDIO_INTERFACE, "Mixer is initialized");
+
+  m_sound_touch.setChannels(2);
+  m_sound_touch.setSampleRate(BackendSampleRate);
+  m_sound_touch.setPitch(1.0);
+  m_sound_touch.setTempo(1.0);
+  m_sound_touch.setSetting(SETTING_USE_QUICKSEEK, 0);
+  m_sound_touch.setSetting(SETTING_SEQUENCE_MS, 62);
+  m_sound_touch.setSetting(SETTING_SEEKWINDOW_MS, 28);
+  m_sound_touch.setSetting(SETTING_OVERLAP_MS, 8);
 }
 
 CMixer::~CMixer()
@@ -37,26 +47,27 @@ unsigned int CMixer::MixerFifo::Mix(short* samples, unsigned int numSamples,
   u32 indexR = m_indexR.load();
   u32 indexW = m_indexW.load();
 
-  u32 low_waterwark = m_input_sample_rate * SConfig::GetInstance().iTimingVariance / 1000;
-  low_waterwark = std::min(low_waterwark, MAX_SAMPLES / 2);
-
-  float numLeft = (float)(((indexW - indexR) & INDEX_MASK) / 2);
-  m_numLeftI = (numLeft + m_numLeftI * (CONTROL_AVG - 1)) / CONTROL_AVG;
-  float offset = (m_numLeftI - low_waterwark) * CONTROL_FACTOR;
-  if (offset > MAX_FREQ_SHIFT)
-    offset = MAX_FREQ_SHIFT;
-  if (offset < -MAX_FREQ_SHIFT)
-    offset = -MAX_FREQ_SHIFT;
-
   // render numleft sample pairs to samples[]
   // advance indexR with sample position
   // remember fractional offset
 
   float emulationspeed = SConfig::GetInstance().m_EmulationSpeed;
-  float aid_sample_rate = m_input_sample_rate + offset;
+  float aid_sample_rate = static_cast<float>(m_input_sample_rate);
   if (consider_framelimit && emulationspeed > 0.0f)
   {
-    aid_sample_rate = aid_sample_rate * emulationspeed;
+    float numLeft = static_cast<float>(((indexW - indexR) & INDEX_MASK) / 2);
+
+    u32 low_waterwark = m_input_sample_rate * SConfig::GetInstance().iTimingVariance / 1000;
+    low_waterwark = std::min(low_waterwark, MAX_SAMPLES / 2);
+
+    m_numLeftI = (numLeft + m_numLeftI * (CONTROL_AVG - 1)) / CONTROL_AVG;
+    float offset = (m_numLeftI - low_waterwark) * CONTROL_FACTOR;
+    if (offset > MAX_FREQ_SHIFT)
+      offset = MAX_FREQ_SHIFT;
+    if (offset < -MAX_FREQ_SHIFT)
+      offset = -MAX_FREQ_SHIFT;
+
+    aid_sample_rate = (aid_sample_rate + offset) * emulationspeed;
   }
 
   const u32 ratio = (u32)(65536.0f * aid_sample_rate / (float)m_mixer->m_sampleRate);
@@ -88,6 +99,9 @@ unsigned int CMixer::MixerFifo::Mix(short* samples, unsigned int numSamples,
     m_frac &= 0xffff;
   }
 
+  // Actual number of samples written to the buffer without padding.
+  unsigned int actual_sample_count = currentSample / 2;
+
   // Padding
   short s[2];
   s[0] = Common::swap16(m_buffer[(indexR - 1) & INDEX_MASK]);
@@ -106,22 +120,99 @@ unsigned int CMixer::MixerFifo::Mix(short* samples, unsigned int numSamples,
   // Flush cached variable
   m_indexR.store(indexR);
 
-  return numSamples;
+  return actual_sample_count;
 }
 
-unsigned int CMixer::Mix(short* samples, unsigned int num_samples, bool consider_framelimit)
+unsigned int CMixer::Mix(short* samples, unsigned int num_samples)
 {
   if (!samples)
     return 0;
 
   memset(samples, 0, num_samples * 2 * sizeof(short));
 
-  m_dma_mixer.Mix(samples, num_samples, consider_framelimit);
-  m_streaming_mixer.Mix(samples, num_samples, consider_framelimit);
-  m_wiimote_speaker_mixer.Mix(samples, num_samples, consider_framelimit);
+  if (SConfig::GetInstance().m_audio_stretch)
+  {
+    unsigned int available_samples =
+        std::min(m_dma_mixer.AvailableSamples(), m_streaming_mixer.AvailableSamples());
+
+    m_stretch_buffer.fill(0);
+
+    m_dma_mixer.Mix(m_stretch_buffer.data(), available_samples, false);
+    m_streaming_mixer.Mix(m_stretch_buffer.data(), available_samples, false);
+    m_wiimote_speaker_mixer.Mix(m_stretch_buffer.data(), available_samples, false);
+
+    if (!m_is_stretching)
+    {
+      m_sound_touch.clear();
+      m_is_stretching = true;
+    }
+    StretchAudio(m_stretch_buffer.data(), available_samples, samples, num_samples);
+  }
+  else
+  {
+    m_dma_mixer.Mix(samples, num_samples, true);
+    m_streaming_mixer.Mix(samples, num_samples, true);
+    m_wiimote_speaker_mixer.Mix(samples, num_samples, true);
+    m_is_stretching = false;
+  }
+
   return num_samples;
 }
 
+void CMixer::StretchAudio(const short* in, unsigned int num_in, short* out, unsigned int num_out)
+{
+  const double time_delta = static_cast<double>(num_out) / m_sampleRate;  // seconds
+
+  // We were given actual_samples number of samples, and num_samples were requested from us.
+  double current_ratio = static_cast<double>(num_in) / static_cast<double>(num_out);
+
+  const double max_latency = SConfig::GetInstance().m_audio_stretch_max_latency;
+  const double max_backlog = m_sampleRate * max_latency / 1000.0 / m_stretch_ratio;
+  const double backlog_fullness = m_sound_touch.numSamples() / max_backlog;
+  if (backlog_fullness > 5.0)
+  {
+    // Too many samples in backlog: Don't push anymore on
+    num_in = 0;
+  }
+
+  // We ideally want the backlog to be about 50% full.
+  // This gives some headroom both ways to prevent underflow and overflow.
+  // We tweak current_ratio to encourage this.
+  constexpr double tweak_time_scale = 0.5;  // seconds
+  current_ratio *= 1.0 + 2.0 * (backlog_fullness - 0.5) * (time_delta / tweak_time_scale);
+
+  // This low-pass filter smoothes out variance in the calculated stretch ratio.
+  // The time-scale determines how responsive this filter is.
+  constexpr double lpf_time_scale = 1.0;  // seconds
+  const double m_lpf_gain = 1.0 - std::exp(-time_delta / lpf_time_scale);
+  m_stretch_ratio += m_lpf_gain * (current_ratio - m_stretch_ratio);
+
+  // Place a lower limit of 10% speed.  When a game boots up, there will be
+  // many silence samples.  These do not need to be timestretched.
+  m_stretch_ratio = std::max(m_stretch_ratio, 0.1);
+  m_sound_touch.setTempo(m_stretch_ratio);
+
+  DEBUG_LOG(AUDIO, "Audio stretching: samples:%u/%u ratio:%f backlog:%f gain: %f", num_in, num_out,
+            m_stretch_ratio, backlog_fullness, m_lpf_gain);
+
+  m_sound_touch.putSamples(in, num_in);
+
+  const size_t samples_received = m_sound_touch.receiveSamples(out, num_out);
+
+  if (samples_received != 0)
+  {
+    m_last_stretched_sample[0] = out[samples_received * 2 - 2];
+    m_last_stretched_sample[1] = out[samples_received * 2 - 1];
+  }
+
+  // Preform padding if we've run out of samples.
+  for (size_t i = samples_received; i < num_out; i++)
+  {
+    out[i * 2 + 0] = m_last_stretched_sample[0];
+    out[i * 2 + 1] = m_last_stretched_sample[1];
+  }
+}
+
 void CMixer::MixerFifo::PushSamples(const short* samples, unsigned int num_samples)
 {
   // Cache access in non-volatile variable
@@ -295,3 +386,11 @@ void CMixer::MixerFifo::SetVolume(unsigned int lvolume, unsigned int rvolume)
   m_LVolume.store(lvolume + (lvolume >> 7));
   m_RVolume.store(rvolume + (rvolume >> 7));
 }
+
+unsigned int CMixer::MixerFifo::AvailableSamples() const
+{
+  unsigned int samples_in_fifo = ((m_indexW.load() - m_indexR.load()) & INDEX_MASK) / 2;
+  if (samples_in_fifo <= 1)
+    return 0;  // CMixer::MixerFifo::Mix always keeps one sample in the buffer.
+  return (samples_in_fifo - 1) * m_mixer->m_sampleRate / m_input_sample_rate;
+}