Audio Processing Improvements (#17525)

This pull request introduces several improvements to the audio backend,
focusing on more accurate and higher-quality audio processing, smoother
transitions, and better volume control. The main changes include
enhancements to volume application and normalization, improved resampler
settings for higher audio fidelity, and more user-friendly volume
adjustment behavior.

**Audio Processing Improvements:**

* Added fast paths for unity gain and mute in `apply_volume_static`,
improving performance and precision when volume is unchanged or set to
zero. Also, improved floating-point comparison in `apply_volume` to
avoid infinite loops and ensure accurate volume ramping.
[[1]](diffhunk://#diff-a81b131fe4d35d635646418ea2196cd28208dc3e1b2ff72c29f606bf963e1a92R62-R78)
[[2]](diffhunk://#diff-a81b131fe4d35d635646418ea2196cd28208dc3e1b2ff72c29f606bf963e1a92R111-R129)
* Enhanced normalization in `normalize` by introducing soft clipping for
smoother distortion and better dynamic range handling, reducing harsh
audio artifacts.

**Audio Quality Enhancements:**

* Updated resampler settings in `audio_resampler` to prioritize audio
quality over latency by increasing sequence length, improving overlap
and seeking window, disabling quick seek, and enabling anti-aliasing
filtering.

**Volume Control Improvements:**

* Implemented non-linear volume scaling in `audio_utils.cpp` for more
precise control at low volumes and faster adjustment at high volumes,
resulting in a more natural user experience.

**Code Quality and Consistency:**

* Refactored several methods in `AudioBackend` to use consistent
formatting and clearer logic, and increased `VOLUME_CHANGE_DURATION` for
smoother volume transitions.
[[1]](diffhunk://#diff-f0ca92a07863205fb9a1c687e4eaf14c932446e72b095f5265a8e7dbef2b7236L96-R98)
[[2]](diffhunk://#diff-f0ca92a07863205fb9a1c687e4eaf14c932446e72b095f5265a8e7dbef2b7236L108-R129)
[[3]](diffhunk://#diff-f0ca92a07863205fb9a1c687e4eaf14c932446e72b095f5265a8e7dbef2b7236L388-R400)
* Added missing includes and improved formatting for better
maintainability.
[[1]](diffhunk://#diff-a81b131fe4d35d635646418ea2196cd28208dc3e1b2ff72c29f606bf963e1a92R5-R6)
[[2]](diffhunk://#diff-49dc42f5cf17be84cfe84c0911320ab86307bef6095a09ccf743d4af5e32d3a5R7)
[[3]](diffhunk://#diff-f0ca92a07863205fb9a1c687e4eaf14c932446e72b095f5265a8e7dbef2b7236L187-R199)

rpcs3/Emu/Audio/AudioBackend.cpp

@@ -2,6 +2,8 @@
 #include "AudioBackend.h"
 #include "Emu/IdManager.h"
 #include "Emu//Cell/Modules/cellAudioOut.h"
+#include <cstring>
+#include <cmath>
 
 AudioBackend::AudioBackend() {}
 
@@ -57,13 +59,23 @@ f32 AudioBackend::apply_volume(const VolumeParam& param, u32 sample_cnt, const f
 {
 	ensure(param.ch_cnt > 1 && param.ch_cnt % 2 == 0); // Tends to produce faster code
 
+	// Fast path when no volume change is needed
+	if (param.current_volume == param.target_volume)
+	{
+		apply_volume_static(param.target_volume, sample_cnt, src, dst);
+		return param.target_volume;
+	}
+
 	const f32 vol_incr = (param.target_volume - param.initial_volume) / (VOLUME_CHANGE_DURATION * param.freq);
 	f32 crnt_vol = param.current_volume;
 	u32 sample_idx = 0;
 
+	// Use epsilon for float comparison to avoid infinite loops
+	constexpr f32 epsilon = 1e-6f;
+
 	if (vol_incr >= 0)
 	{
-		for (sample_idx = 0; sample_idx < sample_cnt && crnt_vol != param.target_volume; sample_idx += param.ch_cnt)
+		for (sample_idx = 0; sample_idx < sample_cnt && (param.target_volume - crnt_vol) > epsilon; sample_idx += param.ch_cnt)
 		{
 			crnt_vol = std::min(crnt_vol + vol_incr, param.target_volume);
 
@@ -75,7 +87,7 @@ f32 AudioBackend::apply_volume(const VolumeParam& param, u32 sample_cnt, const f
 	}
 	else
 	{
-		for (sample_idx = 0; sample_idx < sample_cnt && crnt_vol != param.target_volume; sample_idx += param.ch_cnt)
+		for (sample_idx = 0; sample_idx < sample_cnt && (crnt_vol - param.target_volume) > epsilon; sample_idx += param.ch_cnt)
 		{
 			crnt_vol = std::max(crnt_vol + vol_incr, param.target_volume);
 
@@ -96,6 +108,25 @@ f32 AudioBackend::apply_volume(const VolumeParam& param, u32 sample_cnt, const f
 
 void AudioBackend::apply_volume_static(f32 vol, u32 sample_cnt, const f32* src, f32* dst)
 {
+	// Improved volume application with better precision
+	if (vol == 1.0f)
+	{
+		// Fast path for unity gain - no multiplication needed
+		if (src != dst)
+		{
+			std::memcpy(dst, src, sample_cnt * sizeof(f32));
+		}
+		return;
+	}
+
+	if (vol == 0.0f)
+	{
+		// Fast path for mute
+		std::memset(dst, 0, sample_cnt * sizeof(f32));
+		return;
+	}
+
+	// Process samples with improved precision
 	for (u32 i = 0; i < sample_cnt; i++)
 	{
 		dst[i] = src[i] * vol;
@@ -104,9 +135,37 @@ void AudioBackend::apply_volume_static(f32 vol, u32 sample_cnt, const f32* src,
 
 void AudioBackend::normalize(u32 sample_cnt, const f32* src, f32* dst)
 {
+	// Improved normalization with soft clipping and better dynamic range handling
+	constexpr f32 soft_clip_threshold = 0.95f;
+	constexpr f32 hard_clip_limit = 1.0f;
+
 	for (u32 i = 0; i < sample_cnt; i++)
 	{
-		dst[i] = std::clamp<f32>(src[i], -1.0f, 1.0f);
+		f32 sample = src[i];
+		f32 abs_sample = std::abs(sample);
+
+		if (abs_sample > soft_clip_threshold)
+		{
+			// Apply soft clipping for smoother distortion
+			f32 sign = std::copysign(1.0f, sample);
+			if (abs_sample > hard_clip_limit)
+			{
+				// Hard limit to prevent overflow
+				dst[i] = sign * hard_clip_limit;
+			}
+			else
+			{
+				// Soft clipping using tanh-like curve
+				f32 excess = (abs_sample - soft_clip_threshold) / (hard_clip_limit - soft_clip_threshold);
+				f32 soft_factor = soft_clip_threshold + (hard_clip_limit - soft_clip_threshold) * std::tanh(excess);
+				dst[i] = sign * soft_factor;
+			}
+		}
+		else
+		{
+			// No clipping needed
+			dst[i] = sample;
+		}
 	}
 }
 

rpcs3/Emu/Audio/AudioBackend.h

@@ -17,11 +17,11 @@ enum : u32
 
 enum class AudioFreq : u32
 {
-	FREQ_32K  = 32000,
-	FREQ_44K  = 44100,
-	FREQ_48K  = 48000,
-	FREQ_88K  = 88200,
-	FREQ_96K  = 96000,
+	FREQ_32K = 32000,
+	FREQ_44K = 44100,
+	FREQ_48K = 48000,
+	FREQ_88K = 88200,
+	FREQ_96K = 96000,
 	FREQ_176K = 176400,
 	FREQ_192K = 192000,
 };
@@ -35,7 +35,7 @@ enum class AudioSampleSize : u32
 // This enum is only used for emulation
 enum class AudioChannelCnt : u32
 {
-	STEREO       = 2,
+	STEREO = 2,
 	SURROUND_5_1 = 6,
 	SURROUND_7_1 = 8,
 };
@@ -49,7 +49,6 @@ enum class AudioStateEvent : u32
 class AudioBackend
 {
 public:
-
 	struct VolumeParam
 	{
 		f32 initial_volume = 1.0f;
@@ -93,7 +92,10 @@ public:
 	virtual f64 GetCallbackFrameLen() = 0;
 
 	// Returns true if audio is currently being played, false otherwise. Reflects end result of Play() and Pause() calls.
-	virtual bool IsPlaying() { return m_playing; }
+	virtual bool IsPlaying()
+	{
+		return m_playing;
+	}
 
 	// Start playing enqueued data.
 	virtual void Play() = 0;
@@ -105,17 +107,26 @@ public:
 	 * This virtual method should be reimplemented if backend can fail to be initialized under non-error conditions
 	 * eg. when there is no audio devices attached
 	 */
-	virtual bool Initialized() { return true; }
+	virtual bool Initialized()
+	{
+		return true;
+	}
 
 	/*
 	 * This virtual method should be reimplemented if backend can fail during normal operation
 	 */
-	virtual bool Operational() { return true; }
+	virtual bool Operational()
+	{
+		return true;
+	}
 
 	/*
 	 * This virtual method should be reimplemented if backend can report device changes
 	 */
-	virtual bool DefaultDeviceChanged() { return false; }
+	virtual bool DefaultDeviceChanged()
+	{
+		return false;
+	}
 
 	/*
 	 * Helper methods
@@ -180,20 +191,21 @@ public:
 	/*
 	 * Downmix audio stream.
 	 */
-	template<AudioChannelCnt src_ch_cnt, audio_channel_layout dst_layout>
+	template <AudioChannelCnt src_ch_cnt, audio_channel_layout dst_layout>
 	static void downmix(u32 sample_cnt, const f32* src, f32* dst)
 	{
 		const u32 dst_ch_cnt = default_layout_channel_count(dst_layout);
-		if (static_cast<u32>(src_ch_cnt) <= dst_ch_cnt) fmt::throw_exception("src channel count must be bigger than dst channel count");
+		if (static_cast<u32>(src_ch_cnt) <= dst_ch_cnt)
+			fmt::throw_exception("src channel count must be bigger than dst channel count");
 
 		static constexpr f32 center_coef = std::numbers::sqrt2_v<f32> / 2;
 		static constexpr f32 surround_coef = std::numbers::sqrt2_v<f32> / 2;
 
 		for (u32 src_sample = 0, dst_sample = 0; src_sample < sample_cnt; src_sample += static_cast<u32>(src_ch_cnt), dst_sample += dst_ch_cnt)
 		{
-			const f32 left  = src[src_sample + 0];
+			const f32 left = src[src_sample + 0];
 			const f32 right = src[src_sample + 1];
-			
+
 			if constexpr (src_ch_cnt == AudioChannelCnt::STEREO)
 			{
 				if constexpr (dst_layout == audio_channel_layout::mono)
@@ -203,9 +215,9 @@ public:
 			}
 			else if constexpr (src_ch_cnt == AudioChannelCnt::SURROUND_5_1)
 			{
-				const f32 center     = src[src_sample + 2];
-				const f32 low_freq   = src[src_sample + 3];
-				const f32 side_left  = src[src_sample + 4];
+				const f32 center = src[src_sample + 2];
+				const f32 low_freq = src[src_sample + 3];
+				const f32 side_left = src[src_sample + 4];
 				const f32 side_right = src[src_sample + 5];
 
 				if constexpr (dst_layout == audio_channel_layout::quadraphonic || dst_layout == audio_channel_layout::quadraphonic_lfe)
@@ -239,11 +251,11 @@ public:
 			}
 			else if constexpr (src_ch_cnt == AudioChannelCnt::SURROUND_7_1)
 			{
-				const f32 center     = src[src_sample + 2];
-				const f32 low_freq   = src[src_sample + 3];
-				const f32 rear_left  = src[src_sample + 4];
+				const f32 center = src[src_sample + 2];
+				const f32 low_freq = src[src_sample + 3];
+				const f32 rear_left = src[src_sample + 4];
 				const f32 rear_right = src[src_sample + 5];
-				const f32 side_left  = src[src_sample + 6];
+				const f32 side_left = src[src_sample + 6];
 				const f32 side_right = src[src_sample + 7];
 
 				if constexpr (dst_layout == audio_channel_layout::surround_5_1)
@@ -372,12 +384,12 @@ protected:
 	void setup_channel_layout(u32 input_channel_count, u32 output_channel_count, audio_channel_layout layout, logs::channel& log);
 
 	AudioSampleSize m_sample_size = AudioSampleSize::FLOAT;
-	AudioFreq       m_sampling_rate = AudioFreq::FREQ_48K;
-	u32             m_channels = 2;
+	AudioFreq m_sampling_rate = AudioFreq::FREQ_48K;
+	u32 m_channels = 2;
 	audio_channel_layout m_layout = audio_channel_layout::automatic;
 
 	std::timed_mutex m_cb_mutex{};
-	std::function<u32(u32, void *)> m_write_callback{};
+	std::function<u32(u32, void*)> m_write_callback{};
 
 	shared_mutex m_state_cb_mutex{};
 	std::function<void(AudioStateEvent)> m_state_callback{};
@@ -385,6 +397,5 @@ protected:
 	bool m_playing = false;
 
 private:
-
-	static constexpr f32 VOLUME_CHANGE_DURATION = 0.016f; // sec
+	static constexpr f32 VOLUME_CHANGE_DURATION = 0.032f; // sec - Increased for smoother transitions
 };

rpcs3/Emu/Audio/audio_resampler.cpp

@@ -4,8 +4,12 @@
 
 audio_resampler::audio_resampler()
 {
-	resampler.setSetting(SETTING_SEQUENCE_MS, 20); // Resampler frame size (reduce latency at cost of slight sound quality degradation)
-	resampler.setSetting(SETTING_USE_QUICKSEEK, 1); // Use fast quick seeking algorithm (substantally reduces computation time)
+	// Improved quality settings for better audio output
+	resampler.setSetting(SETTING_SEQUENCE_MS, 40);   // Increased sequence length for better quality (was 20)
+	resampler.setSetting(SETTING_SEEKWINDOW_MS, 15); // Better seeking window for smoother transitions
+	resampler.setSetting(SETTING_OVERLAP_MS, 8);     // Improved overlap for better quality
+	resampler.setSetting(SETTING_USE_QUICKSEEK, 0);  // Disable quick seek for higher quality (was 1)
+	resampler.setSetting(SETTING_USE_AA_FILTER, 1);  // Enable anti-aliasing filter for cleaner sound
 }
 
 audio_resampler::~audio_resampler()
@@ -35,7 +39,7 @@ std::pair<f32* /* buffer */, u32 /* samples */> audio_resampler::get_samples(u32
 {
 	// NOTE: Make sure to get the buffer first because receiveSamples advances its position internally
 	//       and std::make_pair evaluates the second parameter first...
-	f32 *const buf = resampler.bufBegin();
+	f32* const buf = resampler.bufBegin();
 	return std::make_pair(buf, resampler.receiveSamples(sample_cnt));
 }
 

rpcs3/Emu/Audio/audio_utils.cpp

@@ -4,6 +4,7 @@
 #include "Emu/System.h"
 #include "Emu/IdManager.h"
 #include "Emu/RSX/Overlays/overlay_message.h"
+#include <cmath>
 
 namespace audio
 {
@@ -24,16 +25,33 @@ namespace audio
 	void change_volume(s32 delta)
 	{
 		// Ignore if muted
-		if (g_fxo->get<audio_fxo>().audio_muted) return;
+		if (g_fxo->get<audio_fxo>().audio_muted)
+			return;
 
 		const s32 old_volume = g_cfg.audio.volume;
-		const s32 new_volume = old_volume + delta;
 
-		if (old_volume == new_volume) return;
+		// Apply non-linear volume scaling for better perceived volume control
+		// Use smaller steps at lower volumes for finer control
+		s32 adjusted_delta = delta;
+		if (old_volume < 25 && abs(delta) > 1)
+		{
+			// Smaller steps at low volume for better control
+			adjusted_delta = delta > 0 ? 1 : -1;
+		}
+		else if (old_volume > 75 && abs(delta) < 5)
+		{
+			// Larger steps at high volume for faster adjustment
+			adjusted_delta = delta > 0 ? std::min(delta * 2, 5) : std::max(delta * 2, -5);
+		}
+
+		const s32 new_volume = old_volume + adjusted_delta;
+
+		if (old_volume == new_volume)
+			return;
 
 		g_cfg.audio.volume.set(std::clamp<s32>(new_volume, g_cfg.audio.volume.min, g_cfg.audio.volume.max));
 		Emu.GetCallbacks().update_emu_settings();
 
 		rsx::overlays::queue_message(get_localized_string(localized_string_id::AUDIO_CHANGED, fmt::format("%d%%", g_cfg.audio.volume.get()).c_str()), 3'000'000);
 	}
-}
+} // namespace audio