WIP: multi-slot WOV arbiter with VAD gate and D0I3/S0iX support

app/boards/intel_adsp_ace30_ptl.conf

@@ -16,6 +16,15 @@ CONFIG_COMP_GOOGLE_RTC_AUDIO_PROCESSING=m
 CONFIG_GOOGLE_RTC_AUDIO_PROCESSING_MOCK=y
 CONFIG_COMP_STFT_PROCESS=y
 
+# WOV Arbiter, KPB, Mixin/Mixout and Sample Detector
+CONFIG_COMP_MIXIN_MIXOUT=y
+CONFIG_COMP_WOV_ARBITER=y
+CONFIG_COMP_KPB=y
+CONFIG_KPB_MAX_BUFF_TIME=2000
+CONFIG_COMP_ECNS=y
+CONFIG_SAMPLES=y
+CONFIG_SAMPLE_KEYPHRASE=y
+
 # SOF / infrastructure
 CONFIG_KCPS_DYNAMIC_CLOCK_CONTROL=n
 CONFIG_PROBE=y

app/boards/intel_adsp_ace30_wcl.conf

@@ -12,6 +12,12 @@ CONFIG_FORMAT_CONVERT_HIFI3=n
 # SOF / audio modules / mocks
 # This mock is part of official sof-bin releases because the CI that
 # tests it can't use extra CONFIGs. See #9410, #8722 and #9386
+# Google hotword needs the proprietary hotword_dsp_api.h and an ACE30 library build
+CONFIG_COMP_GOOGLE_HOTWORD_DETECT=n
+CONFIG_COMP_WOV_ARBITER=y
+CONFIG_COMP_ECNS=y
+# Post-ECNS KPB is mono 16-bit: 2000 ms needs 64000 B, well within the DSP heap
+CONFIG_KPB_MAX_BUFF_TIME=2000
 CONFIG_COMP_GOOGLE_RTC_AUDIO_PROCESSING=m
 CONFIG_GOOGLE_RTC_AUDIO_PROCESSING_MOCK=y
 CONFIG_COMP_STFT_PROCESS=y

app/boards/intel_adsp_cavs25.conf

@@ -11,6 +11,10 @@ CONFIG_COMP_DRC=y
 CONFIG_COMP_MFCC=y
 CONFIG_COMP_MULTIBAND_DRC=y
 CONFIG_COMP_VOLUME_WINDOWS_FADE=y
+CONFIG_COMP_WOV_ARBITER=y
+CONFIG_COMP_KPB=y
+CONFIG_SAMPLES=y
+CONFIG_SAMPLE_KEYPHRASE=y
 CONFIG_FORMAT_CONVERT_HIFI3=n
 CONFIG_PCM_CONVERTER_FORMAT_S16LE=y
 CONFIG_PCM_CONVERTER_FORMAT_S24LE=y

app/configs/wcl/modules.conf

@@ -0,0 +1,2 @@
+CONFIG_COMP_TDFB=m
+CONFIG_COMP_VOLUME=m

src/audio/CMakeLists.txt

@@ -41,6 +41,9 @@ if(NOT CONFIG_COMP_MODULE_SHARED_LIBRARY_BUILD)
 	if(CONFIG_COMP_DRC)
 		add_subdirectory(drc)
 	endif()
+	if(CONFIG_COMP_ECNS)
+		add_subdirectory(ecns)
+	endif()
 	if(CONFIG_COMP_FIR)
 		add_subdirectory(eq_fir)
 	endif()
@@ -107,6 +110,9 @@ if(NOT CONFIG_COMP_MODULE_SHARED_LIBRARY_BUILD)
 	if(CONFIG_COMP_VOLUME)
 		add_subdirectory(volume)
 	endif()
+	if(CONFIG_COMP_WOV_ARBITER)
+		add_subdirectory(wov_arbiter)
+	endif()
 	if(CONFIG_DTS_CODEC)
 		add_subdirectory(codec)
 	endif()

src/audio/Kconfig

@@ -102,13 +102,48 @@ config COMP_STUBS
 	  Select to force all 3P blocks to link against stubs rather than their libraries. This
 	  should only be used in testing environments like fuzzers or CI.
 
+config COMP_WOV_ARBITER
+	bool "WOV arbiter component"
+	depends on COMP_KPB
+	depends on IPC_MAJOR_4
+	help
+	  Select to build the WOV (Wake-on-Voice) arbiter.  The arbiter sits
+	  between multiple KPB host-drain outputs and a single host PCM copier.
+	  When a keyword is detected by one of the WOV detectors the arbiter
+	  routes that KPB's drain stream to the host and instructs the remaining
+	  detectors to pause.
+
+config COMP_ECNS
+	bool "ECNS data processing component"
+	default y
+	depends on IPC_MAJOR_4
+	help
+	  Select to build the ECNS (Echo Cancellation and Noise Suppression)
+	  Data Processing (DP) component. Operates with a 20ms period to process
+	  4 DMIC channels (2 mics + 2 echo references) and outputs clean mono
+	  to KPB and clean stereo to host capture.
+
 config COMP_KPB
 	bool "KPB component"
 	default y
 	help
 	  Select for KPB component
 if COMP_KPB
 
+config KPB_MAX_NO_OF_CLIENTS
+	int "Maximum number of KPB clients"
+	default 4
+	help
+	  Maximum number of simultaneous KPB drain clients (host + WOV detectors).
+	  Increase if more than 4 clients need concurrent KPB access.
+
+config KPB_MAX_BUFF_TIME
+	int "KPB history buffer time (ms)"
+	default 6000 if XCHAL_HW_VERSION_MAJOR >= 300
+	default 2100
+	help
+	  KPB history ring-buffer length in milliseconds.
+
 config KPB_FORCE_COPY_TYPE_NORMAL
 	bool "KPB force copy type normal"
 	default y

src/audio/copier/copier_dai.c

@@ -415,13 +415,21 @@ __cold void copier_dai_free(struct processing_module *mod)
 	assert_can_be_cold();
 
 	for (int i = 0; i < cd->endpoint_num; i++) {
-		dai_common_free(cd->dd[i]);
-		mod_free(mod, cd->dd[i]->gain_data);
-		mod_free(mod, cd->dd[i]);
+		if (cd->dd[i]) {
+			if (cd->dd[i]->gain_data) {
+				mod_free(mod, cd->dd[i]->gain_data);
+				cd->dd[i]->gain_data = NULL;
+			}
+			dai_common_free(cd->dd[i]);
+			mod_free(mod, cd->dd[i]);
+			cd->dd[i] = NULL;
+		}
 	}
 	/* only dai have multi endpoint case */
-	if (cd->multi_endpoint_buffer)
+	if (cd->multi_endpoint_buffer) {
 		buffer_free(cd->multi_endpoint_buffer);
+		cd->multi_endpoint_buffer = NULL;
+	}
 }
 
 int copier_dai_prepare(struct comp_dev *dev, struct copier_data *cd)

src/audio/dai-zephyr.c

@@ -683,25 +683,40 @@ __cold void dai_common_free(struct dai_data *dd)
 {
 	assert_can_be_cold();
 
+	if (!dd)
+		return;
+
 #ifdef CONFIG_SOF_TELEMETRY_IO_PERFORMANCE_MEASUREMENTS
 	io_perf_monitor_release_slot(dd->io_perf_dai_byte_count);
 #endif
 
-	if (dd->group)
+	if (dd->group) {
 		dai_group_put(dd->group);
+		dd->group = NULL;
+	}
 
 	if (dd->chan_index >= 0) {
-		sof_dma_release_channel(dd->dma, dd->chan_index);
+		if (dd->dma)
+			sof_dma_release_channel(dd->dma, dd->chan_index);
 		dd->chan_index = -EINVAL;
 	}
 
-	sof_dma_put(dd->dma);
+	if (dd->dma) {
+		sof_dma_put(dd->dma);
+		dd->dma = NULL;
+	}
 
 	dai_release_llp_slot(dd);
 
-	dai_put(dd->dai);
+	if (dd->dai) {
+		dai_put(dd->dai);
+		dd->dai = NULL;
+	}
 
-	sof_heap_free(dd->alloc_ctx.heap, dd->dai_spec_config);
+	if (dd->dai_spec_config) {
+		sof_heap_free(dd->alloc_ctx.heap, dd->dai_spec_config);
+		dd->dai_spec_config = NULL;
+	}
 }
 
 __cold static void dai_free(struct comp_dev *dev)
@@ -710,6 +725,11 @@ __cold static void dai_free(struct comp_dev *dev)
 
 	assert_can_be_cold();
 
+	if (!dd) {
+		comp_free_device(dev);
+		return;
+	}
+
 	if (dd->group)
 		notifier_unregister(dev, dd->group, NOTIFIER_ID_DAI_TRIGGER);
 

src/audio/ecns/CMakeLists.txt

@@ -0,0 +1,3 @@
+# SPDX-License-Identifier: BSD-3-Clause
+
+add_local_sources(sof ecns.c)

src/audio/ecns/README.md

@@ -0,0 +1,77 @@
+# ECNS (Echo Cancellation & Noise Suppression) Processing Module
+
+## Overview
+
+The `ecns` component is an IPC4 Data Processing (DP) module operating at a 20 ms period (320 samples at 16 kHz). It consumes a 4-channel audio stream comprising:
+- **Channels 0 & 1**: Primary microphone inputs.
+- **Channels 2 & 3**: Far-end echo reference inputs.
+
+The module performs echo cancellation and noise suppression, delivering two distinct output streams via separate output pins:
+1. **Output Pin 0 (Mono Clean Mic)**: Provides a single-channel, echo-canceled and noise-suppressed speech stream intended for keyword detection and pre-roll buffering in the Key Phrase Buffer (`kpb`).
+2. **Output Pin 1 (Stereo Clean Mic)**: Provides a two-channel clean speech stream routed directly or via a host mixin/mixout bridge to a standard ALSA capture device (e.g. `hw:0,10`).
+
+---
+
+## Topology Architecture
+
+```mermaid
+graph LR
+    subgraph Capture["Pipeline 100 — 4ch DMIC"]
+        DAI["DAI Copier (4ch 16kHz)"] --> M100["mixin 100.1"]
+    end
+
+    subgraph ECNS_DP["Pipeline 105 — ECNS DP Module (20ms / 320 frames)"]
+        M100 --> MO105["mixout 105.1"]
+        MO105 --> ECNS["ecns.105.1\n(UUID: 214f8a6c-493a-4b1e-8b1e0d3b6f8a2c15)"]
+        ECNS -- "Pin 0 (1ch Mono)" --> MIX105_1["mixin 105.1 (KPB Mixin)"]
+        ECNS -- "Pin 1 (2ch Stereo)" --> MIX105_2["mixin 105.2 (Host Mixin)"]
+    end
+
+    subgraph WOV_Path["Pipeline 106 & 104 — WOV / KPB Path"]
+        MIX105_1 --> MO106["mixout 106.1"]
+        MO106 --> KPB["kpb.106.1\n(2.0s mono / 64 KB)"]
+        KPB --> WOV["WOV Slots 0..2"] --> ARB["wov_arbiter.104.1"]
+        ARB --> HC11["host-copier.11\n(ALSA hw:0,11)"]
+    end
+
+    subgraph Host_Path["Pipeline 107 — ECNS Host Capture Path"]
+        MIX105_2 --> MO107["mixout 107.1"]
+        MO107 --> HC10["host-copier.10\n(ALSA hw:0,10)"]
+    end
+
+    style ECNS fill:#1b4f72,stroke:#555,color:#fff
+    style KPB fill:#1c4966,stroke:#555,color:#fff
+    style ARB fill:#4a235a,stroke:#555,color:#fff
+    style HC10 fill:#1e8449,stroke:#555,color:#fff
+    style HC11 fill:#2d5a27,stroke:#555,color:#fff
+```
+
+---
+
+## Specifications
+
+| Parameter | Specification |
+|---|---|
+| Module UUID | `214f8a6c-493a-4b1e-8b1e0d3b6f8a2c15` |
+| Processing Domain | DP (Data Processing) thread |
+| Execution Period | 20 ms (320 samples at 16 kHz) |
+| Input Channels | 4 (Ch 0,1: Microphones; Ch 2,3: Echo Reference) |
+| Input Format | 16 kHz · 16-bit signed little-endian (`S16_LE`) |
+| Output Pin 0 | 1 channel (Mono clean mic) · 16 kHz · `S16_LE` · OBS = 640 bytes (20ms) |
+| Output Pin 1 | 2 channels (Stereo clean mic) · 16 kHz · `S16_LE` · OBS = 1280 bytes (20ms) |
+| Sinks Acceptance | Independent sink buffer limit checks (`kpb_frames` vs `host_frames`) |
+| Dynamic Pipeline | Compatible (`dynamic_pipeline 1`) |
+
+---
+
+## Verification & Usage
+
+### 1. Recording Clean Stereo Audio (Pin 1 / PCM 10)
+```bash
+arecord -D hw:0,10 -c 2 -r 16000 -f S16_LE -d 3 /tmp/ecns_stereo.wav
+```
+
+### 2. Recording Keyword Detection Pre-Roll Audio (Pin 0 / PCM 11)
+```bash
+arecord -D hw:0,11 -c 1 -r 16000 -f S16_LE -d 3 /tmp/wov_mono.wav
+```