Upgrade whisper.cpp from v1.7.2 fork to upstream v1.8.4

.github/actions/build-whisper/action.yaml

@@ -4,11 +4,11 @@ inputs:
   whisper-cpp-repo:
     description: 'The whisper.cpp repository to use'
     required: true
-    default: 'dharmab/whisper.cpp'
+    default: 'ggml-org/whisper.cpp'
   whisper-cpp-version:
     description: 'The version of whisper.cpp to use'
     required: true
-    default: 'v1.7.2-windows-fix'
+    default: 'v1.8.4'
   os:
     description: 'The operating system to build for'
     required: true

.github/workflows/skyeye.yaml

@@ -137,6 +137,7 @@ jobs:
           install: |
             base-devel
             git
+            mingw-w64-ucrt-x86_64-cmake
             mingw-w64-ucrt-x86_64-gcc
             mingw-w64-ucrt-x86_64-toolchain
             mingw-w64-ucrt-x86_64-opus

Dockerfile

@@ -3,6 +3,7 @@ FROM golang:1.26 AS builder
 RUN apt-get update && apt-get install -y \
     git \
     make \
+    cmake \
     lsb-release \
     gcc \
     libopus-dev \

Makefile

@@ -24,18 +24,21 @@ SKYEYE_BIN = skyeye
 SKYEYE_SCALER_BIN = skyeye-scaler
 
 WHISPER_CPP_PATH = third_party/whisper.cpp
-LIBWHISPER_PATH = $(WHISPER_CPP_PATH)/libwhisper.a
+WHISPER_CPP_BUILD_DIR = $(WHISPER_CPP_PATH)/build_go
+LIBWHISPER_PATH = $(WHISPER_CPP_BUILD_DIR)/src/libwhisper.a
 WHISPER_H_PATH = $(WHISPER_CPP_PATH)/include/whisper.h
-WHISPER_CPP_REPO = https://github.com/dharmab/whisper.cpp.git
-WHISPER_CPP_VERSION = v1.7.2-windows-fix
-WHISPER_CPP_BUILD_ENV =
+WHISPER_CPP_REPO = https://github.com/ggml-org/whisper.cpp.git
+WHISPER_CPP_VERSION = v1.8.4
+WHISPER_CPP_CMAKE_ARGS =
 
 # Compiler variables and flags
 GOBUILDVARS = GOARCH=$(GOARCH)
 ABS_WHISPER_CPP_PATH = $(abspath $(WHISPER_CPP_PATH))
+ABS_WHISPER_CPP_BUILD_DIR = $(abspath $(WHISPER_CPP_BUILD_DIR))
+LIBRARY_PATHS = $(ABS_WHISPER_CPP_BUILD_DIR)/src:$(ABS_WHISPER_CPP_BUILD_DIR)/ggml/src
 BUILD_VARS = CGO_ENABLED=1 \
   C_INCLUDE_PATH="$(ABS_WHISPER_CPP_PATH)/ggml/include:$(ABS_WHISPER_CPP_PATH)/include" \
-  LIBRARY_PATH="$(ABS_WHISPER_CPP_PATH)"
+  LIBRARY_PATH="$(LIBRARY_PATHS)"
 BUILD_FLAGS = -tags nolibopusfile
 
 # Populate --version from Git tag
@@ -50,8 +53,11 @@ ifeq ($(OS_DISTRIBUTION),macOS)
 CC=$(shell brew --prefix llvm)/bin/clang
 CXX=$(shell brew --prefix llvm)/bin/clang++
 BUILD_VARS += CC=$(CC) CXX=$(CXX)
-# Enable GPU acceleration
-WHISPER_CPP_BUILD_ENV = GGML_METAL=1
+LIBRARY_PATHS := $(LIBRARY_PATHS):$(ABS_WHISPER_CPP_BUILD_DIR)/ggml/src/ggml-metal:$(ABS_WHISPER_CPP_BUILD_DIR)/ggml/src/ggml-blas
+# Link OpenMP runtime for ggml-cpu on macOS (Go bindings only specify -fopenmp on Linux)
+BUILD_VARS += CGO_LDFLAGS=-fopenmp
+WHISPER_CPP_CMAKE_ARGS = -DCMAKE_C_COMPILER=$(CC) -DCMAKE_CXX_COMPILER=$(CXX) \
+  -DCMAKE_C_FLAGS=-Wno-elaborated-enum-base -DCMAKE_CXX_FLAGS=-Wno-elaborated-enum-base
 endif
 
 # Windows-specific settings
@@ -62,12 +68,14 @@ SKYEYE_SCALER_BIN = skyeye-scaler.exe
 # Override Windows Go environment with MSYS2 UCRT64 Go environment
 GO = /ucrt64/bin/go
 GOBUILDVARS += GOROOT="/ucrt64/lib/go" GOPATH="/ucrt64"
+# Use MSYS Makefiles generator for cmake on MSYS2
+WHISPER_CPP_CMAKE_ARGS = -G "MSYS Makefiles"
 # Static linking on Windows to avoid MSYS2 dependency at runtime
 LIBRARIES = opus soxr
 CFLAGS = $(shell pkg-config $(LIBRARIES) --cflags --static)
 BUILD_VARS += CFLAGS='$(CFLAGS)'
 EXTLDFLAGS = $(shell pkg-config $(LIBRARIES) --libs --static)
-LDFLAGS += -linkmode external -extldflags "$(EXTLDFLAGS) -static"
+LDFLAGS += -linkmode external -extldflags "$(EXTLDFLAGS) -lgomp -static"
 endif
 
 BUILD_VARS += LDFLAGS='$(LDFLAGS)'
@@ -82,6 +90,7 @@ install-msys2-dependencies:
 	pacman -Syu --needed \
 	  git \
 	  base-devel \
+	  $(MINGW_PACKAGE_PREFIX)-cmake \
 	  $(MINGW_PACKAGE_PREFIX)-toolchain \
 	  $(MINGW_PACKAGE_PREFIX)-go \
 	  $(MINGW_PACKAGE_PREFIX)-opus \
@@ -92,6 +101,7 @@ install-arch-linux-dependencies:
 	sudo pacman -Syu \
 	  git \
 	  base-devel \
+	  cmake \
 	  go \
 	  opus \
 	  libsoxr
@@ -102,6 +112,7 @@ install-debian-dependencies:
 	sudo apt-get install -y \
 	  git \
 	  build-essential \
+	  cmake \
 	  golang-go \
 	  libopus-dev \
 	  libopus0 \
@@ -114,6 +125,7 @@ install-fedora-dependencies:
 	  git \
 	  development-tools \
 	  c-development \
+	  cmake \
 	  golang \
 	  opus-devel \
 	  opus \
@@ -125,6 +137,7 @@ install-macos-dependencies:
 	xcode-select --install || true
 	brew install \
 	  git \
+	  cmake \
 	  llvm \
 	  pkg-config \
 	  go \
@@ -136,8 +149,18 @@ download-whisper-%:
 	curl -L -o $*.bin https://huggingface.co/ggerganov/whisper.cpp/resolve/main/$*.bin
 
 $(LIBWHISPER_PATH) $(WHISPER_H_PATH):
-	if [ ! -f $(LIBWHISPER_PATH) -o ! -f $(WHISPER_H_PATH) ]; then git -C "$(WHISPER_CPP_PATH)" checkout --quiet $(WHISPER_CPP_VERSION) || git clone --depth 1 --branch $(WHISPER_CPP_VERSION) -c advice.detachedHead=false "$(WHISPER_CPP_REPO)" "$(WHISPER_CPP_PATH)" && $(WHISPER_CPP_BUILD_ENV) make -C $(WHISPER_CPP_PATH)/bindings/go whisper; fi
-	if [ -f third_party/whisper.cpp/whisper.a ] && [ ! -f $(LIBWHISPER_PATH) ]; then cp third_party/whisper.cpp/whisper.a $(LIBWHISPER_PATH); fi
+	if [ ! -f $(LIBWHISPER_PATH) -o ! -f $(WHISPER_H_PATH) ]; then \
+		git -C "$(WHISPER_CPP_PATH)" checkout --quiet $(WHISPER_CPP_VERSION) || \
+		git clone --depth 1 --branch $(WHISPER_CPP_VERSION) -c advice.detachedHead=false "$(WHISPER_CPP_REPO)" "$(WHISPER_CPP_PATH)" && \
+		cmake -S "$(WHISPER_CPP_PATH)" -B "$(WHISPER_CPP_BUILD_DIR)" \
+			-DCMAKE_BUILD_TYPE=Release \
+			-DBUILD_SHARED_LIBS=OFF \
+			$(WHISPER_CPP_CMAKE_ARGS) && \
+		cmake --build "$(WHISPER_CPP_BUILD_DIR)" --target whisper && \
+		for f in $$(find "$(WHISPER_CPP_BUILD_DIR)" -name '*.a' ! -name 'lib*'); do \
+			mv "$$f" "$$(dirname $$f)/lib$$(basename $$f)"; \
+		done; \
+	fi
 
 .PHONY: whisper
 whisper: $(LIBWHISPER_PATH) $(WHISPER_H_PATH)

go.mod

@@ -11,7 +11,7 @@ require (
 	github.com/dharmab/goacmi v1.0.3
 	github.com/dharmab/numwords v1.0.1
 	github.com/gammazero/deque v0.2.1
-	github.com/ggerganov/whisper.cpp/bindings/go v0.0.0-20241121150429-8c6a9b8bb6a0
+	github.com/ggerganov/whisper.cpp/bindings/go v0.0.0-20260319084013-9386f2394010
 	github.com/go-audio/aiff v1.1.0
 	github.com/gofrs/flock v0.13.0
 	github.com/gopxl/beep/v2 v2.0.3

go.sum

@@ -210,8 +210,8 @@ github.com/fzipp/gocyclo v0.6.0 h1:lsblElZG7d3ALtGMx9fmxeTKZaLLpU8mET09yN4BBLo=
 github.com/fzipp/gocyclo v0.6.0/go.mod h1:rXPyn8fnlpa0R2csP/31uerbiVBugk5whMdlyaLkLoA=
 github.com/gammazero/deque v0.2.1 h1:qSdsbG6pgp6nL7A0+K/B7s12mcCY/5l5SIUpMOl+dC0=
 github.com/gammazero/deque v0.2.1/go.mod h1:LFroj8x4cMYCukHJDbxFCkT+r9AndaJnFMuZDV34tuU=
-github.com/ggerganov/whisper.cpp/bindings/go v0.0.0-20241121150429-8c6a9b8bb6a0 h1:vSv0WhsCkhau0BdC1H7Q2lHR8pzz0skaQkXZdEPYp/4=
-github.com/ggerganov/whisper.cpp/bindings/go v0.0.0-20241121150429-8c6a9b8bb6a0/go.mod h1:qyHjS/50ORo01H0NsuEEGsQR9VCtOcEye0gUl2sx1s8=
+github.com/ggerganov/whisper.cpp/bindings/go v0.0.0-20260319084013-9386f2394010 h1:+m6yO3VYSAUUzpG9lz4ktBx0cCba/Nvww8OKDgnapQQ=
+github.com/ggerganov/whisper.cpp/bindings/go v0.0.0-20260319084013-9386f2394010/go.mod h1:qyHjS/50ORo01H0NsuEEGsQR9VCtOcEye0gUl2sx1s8=
 github.com/ghostiam/protogetter v0.3.20 h1:oW7OPFit2FxZOpmMRPP9FffU4uUpfeE/rEdE1f+MzD0=
 github.com/ghostiam/protogetter v0.3.20/go.mod h1:FjIu5Yfs6FT391m+Fjp3fbAYJ6rkL/J6ySpZBfnODuI=
 github.com/go-audio/aiff v1.1.0 h1:m2LYgu/2BarpF2yZnFPWtY3Tp41k0A4y51gDRZZsEuU=

pkg/recognizer/whisper.go

@@ -48,6 +48,7 @@ func (r *whisperRecognizer) Recognize(ctx context.Context, sample []float32, ena
 
 	err = wCtx.Process(
 		sample,
+		nil,
 		func(segment whisper.Segment) {
 			event := log.Debug()
 			if enableTranscriptionLogging {