[MM-59980] Upgrade to whisper.cpp v1.7.1 (#33)

* Avoid AV512 CPU extensions

* Upgrade to whisper.cpp v1.7.1

* Go is not really needed in CI

* Fix condition

* Update opus

* Fix expression

.github/workflows/ci.yml

@@ -2,49 +2,33 @@ on: [push]
 name: CI
 jobs:
   lint:
-    env:
-      GOPATH: ${{ github.workspace }}
-
     defaults:
       run:
-        working-directory: ${{ env.GOPATH }}/src/github.com/${{ github.repository }}
+        working-directory: ${{ github.workspace }}/src/github.com/${{ github.repository }}
 
     runs-on: ubuntu-latest
 
     steps:
     - name: Checkout Code
       uses: actions/checkout@v4
       with:
-        path: ${{ env.GOPATH }}/src/github.com/${{ github.repository }}
-    - name: Install Go
-      uses: actions/setup-go@v5
-      with:
-        go-version-file: ${{ env.GOPATH }}/src/github.com/${{ github.repository }}/go.mod
-        cache-dependency-path: ${{ env.GOPATH }}/src/github.com/${{ github.repository }}/go.sum
+        path: ${{ github.workspace }}/src/github.com/${{ github.repository }}
     - name: Lint code
       run: |
         make go-lint
 
   test:
-    env:
-      GOPATH: ${{ github.workspace }}
-
     defaults:
       run:
-        working-directory: ${{ env.GOPATH }}/src/github.com/${{ github.repository }}
+        working-directory: ${{ github.workspace }}/src/github.com/${{ github.repository }}
 
     runs-on: ubuntu-latest
 
     steps:
     - name: Checkout Code
       uses: actions/checkout@v4
       with:
-        path: ${{ env.GOPATH }}/src/github.com/${{ github.repository }}
-    - name: Install Go
-      uses: actions/setup-go@v5
-      with:
-        go-version-file: ${{ env.GOPATH }}/src/github.com/${{ github.repository }}/go.mod
-        cache-dependency-path: ${{ env.GOPATH }}/src/github.com/${{ github.repository }}/go.sum
+        path: ${{ github.workspace }}/src/github.com/${{ github.repository }}
     - name: Execute Tests
       run: |
         go mod download

Makefile

@@ -45,12 +45,12 @@ endif
 
 ## CGO dependencies
 # Whisper.cpp
-WHISPER_VERSION ?= "1.6.2"
-WHISPER_SHA ?= "da7988072022acc3cfa61b370b3c51baad017f1900c3dc4e68cb276499f66894"
+WHISPER_VERSION ?= "1.7.1"
+WHISPER_SHA ?= "97f19a32212f2f215e538ee37a16ff547aaebc54817bd8072034e02466ce6d55"
 WHISPER_MODELS ?= "tiny base small"
 # Opus
-OPUS_VERSION ?= "1.4"
-OPUS_SHA ?= "c9b32b4253be5ae63d1ff16eea06b94b5f0f2951b7a02aceef58e3a3ce49c51f"
+OPUS_VERSION ?= "1.5.2"
+OPUS_SHA ?= "65c1d2f78b9f2fb20082c38cbe47c951ad5839345876e46941612ee87f9a7ce1"
 # ONNX Runtime
 ONNX_VERSION ?= "1.18.1"
 # Azure Speech SDK

build/Dockerfile

@@ -2,23 +2,28 @@
 # A multi stage build, with golang used as a builder
 # and debian:sid-slim as runner
 ARG GO_VERSION
-FROM --platform=$TARGETPLATFORM debian:bookworm-20240612-slim as base
+FROM --platform=$TARGETPLATFORM debian:bookworm-20240612-slim AS base
 
 # Setup system dependencies
 WORKDIR /workdir
 
-# Install TLS certificates
+# Install TLS certificates and libgomp
 # hadolint ignore=DL3008
 RUN apt-get update && \
-    apt-get install --no-install-recommends -y ca-certificates && \
+    apt-get install --no-install-recommends -y ca-certificates libgomp1 && \
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
 # Create unprivileged user to run the recorder process
 RUN groupadd -r calls && useradd -mr -g calls calls
 
 ARG GO_VERSION
-FROM --platform=$TARGETPLATFORM golang:${GO_VERSION}-bookworm as builder
+FROM --platform=$TARGETPLATFORM golang:${GO_VERSION}-bookworm AS builder
+# Install patch
+# hadolint ignore=DL3008
+RUN apt-get update && \
+    apt-get install --no-install-recommends -y patch
+
 ARG TARGETOS
 ARG TARGETARCH
 ARG GO_BUILD_PLATFORMS=${TARGETOS}-${TARGETARCH}
@@ -37,7 +42,7 @@ ARG AZURE_SDK_SHA
 # We first copy just the build directory so that we can properly cache dependencies which can take a while to download.
 COPY ./build /src/build
 WORKDIR /src
-RUN /bin/bash ./build/prepare_deps.sh ${OPUS_VERSION} ${OPUS_SHA} ${WHISPER_VERSION} ${WHISPER_SHA} "${WHISPER_MODELS}" ${ONNX_VERSION} ${TARGETARCH} ${AZURE_SDK_VERSION} ${AZURE_SDK_SHA}
+RUN /bin/bash ./build/prepare_deps.sh ${OPUS_VERSION} ${OPUS_SHA} ${WHISPER_VERSION} ${WHISPER_SHA} "${WHISPER_MODELS}" ${ONNX_VERSION} ${TARGETARCH} ${AZURE_SDK_VERSION} ${AZURE_SDK_SHA} "true"
 
 COPY . /src
 RUN /bin/bash ./build/build.sh ${OPUS_VERSION} ${OPUS_SHA} ${WHISPER_VERSION} ${WHISPER_SHA} "${WHISPER_MODELS}" ${ONNX_VERSION} ${TARGETARCH} ${AZURE_SDK_VERSION} ${AZURE_SDK_SHA}

build/build.sh

@@ -12,9 +12,9 @@ AZURE_SDK_VERSION=$8
 AZURE_SDK_SHA=$9
 
 OPUS_INCLUDE_PATH="/tmp/opus-${OPUS_VERSION}/include"
-WHISPER_INCLUDE_PATH="/tmp/whisper.cpp-${WHISPER_VERSION}"
+WHISPER_INCLUDE_PATH="/tmp/whisper.cpp-${WHISPER_VERSION}/include:/tmp/whisper.cpp-${WHISPER_VERSION}/ggml/include"
 OPUS_LIBRARY_PATH="/tmp/opus-${OPUS_VERSION}/.libs"
-WHISPER_LIBRARY_PATH=${WHISPER_INCLUDE_PATH}
+WHISPER_LIBRARY_PATH="/tmp/whisper.cpp-${WHISPER_VERSION}"
 ONNX_INCLUDE_PATH="/tmp/onnxruntime-linux-${ONNX_VERSION}/include"
 ONNX_LIBRARY_PATH="/tmp/onnxruntime-linux-${ONNX_VERSION}/lib"
 AZURE_SDK_INCLUDE_PATH="/tmp/SpeechSDK-Linux-${AZURE_SDK_VERSION}/include/c_api"

build/lint.sh

@@ -11,9 +11,9 @@ AZURE_SDK_VERSION=$7
 AZURE_SDK_SHA=$8
 
 OPUS_INCLUDE_PATH="/tmp/opus-${OPUS_VERSION}/include"
-WHISPER_INCLUDE_PATH="/tmp/whisper.cpp-${WHISPER_VERSION}"
+WHISPER_INCLUDE_PATH="/tmp/whisper.cpp-${WHISPER_VERSION}/include:/tmp/whisper.cpp-${WHISPER_VERSION}/ggml/include"
 OPUS_LIBRARY_PATH="/tmp/opus-${OPUS_VERSION}/.libs"
-WHISPER_LIBRARY_PATH=${WHISPER_INCLUDE_PATH}
+WHISPER_LIBRARY_PATH="/tmp/whisper.cpp-${WHISPER_VERSION}"
 ONNX_INCLUDE_PATH="/tmp/onnxruntime-linux-${ONNX_VERSION}/include"
 ONNX_LIBRARY_PATH="/tmp/onnxruntime-linux-${ONNX_VERSION}/lib"
 AZURE_SDK_INCLUDE_PATH="/tmp/SpeechSDK-Linux-${AZURE_SDK_VERSION}/include/c_api"

build/prepare_deps.sh

@@ -10,6 +10,7 @@ ONNX_VERSION=$6
 TARGET_ARCH=$7
 AZURE_SDK_VERSION=$8
 AZURE_SDK_SHA=$9
+IS_BUILD=${10}
 ONNX_ARCH=x64
 ONNX_SHA=a0994512ec1e1debc00c18bfc7a5f16249f6ebd6a6128ff2034464cc380ea211
 if [ "$TARGET_ARCH" == "arm64" ]; then
@@ -35,6 +36,7 @@ wget https://github.com/ggerganov/whisper.cpp/archive/refs/tags/v${WHISPER_VERSI
 echo "${WHISPER_SHA} v${WHISPER_VERSION}.tar.gz" | sha256sum --check && \
 tar xf v${WHISPER_VERSION}.tar.gz && \
 cd whisper.cpp-${WHISPER_VERSION} && \
+([[ "$TARGET_ARCH" == "amd64" ]] && [[ "$IS_BUILD" == "true" ]] && echo "Patching Whisper.CPP Makefile" && patch -p1 Makefile /src/build/whisper.patch) || true && \
 for model in ${MODELS}; do ./models/download-ggml-model.sh "${model}"; done && \
 make -j4 libwhisper.a UNAME_M=${UNAME_M} && \
 cd /tmp && \

build/run_tests.sh

@@ -12,9 +12,9 @@ AZURE_SDK_VERSION=$8
 AZURE_SDK_SHA=$9
 
 OPUS_INCLUDE_PATH="/tmp/opus-${OPUS_VERSION}/include"
-WHISPER_INCLUDE_PATH="/tmp/whisper.cpp-${WHISPER_VERSION}"
+WHISPER_INCLUDE_PATH="/tmp/whisper.cpp-${WHISPER_VERSION}/include:/tmp/whisper.cpp-${WHISPER_VERSION}/ggml/include"
 OPUS_LIBRARY_PATH="/tmp/opus-${OPUS_VERSION}/.libs"
-WHISPER_LIBRARY_PATH=${WHISPER_INCLUDE_PATH}
+WHISPER_LIBRARY_PATH="/tmp/whisper.cpp-${WHISPER_VERSION}"
 ONNX_INCLUDE_PATH="/tmp/onnxruntime-linux-${ONNX_VERSION}/include"
 ONNX_LIBRARY_PATH="/tmp/onnxruntime-linux-${ONNX_VERSION}/lib"
 AZURE_SDK_INCLUDE_PATH="/tmp/SpeechSDK-Linux-${AZURE_SDK_VERSION}/include/c_api"

build/whisper.patch

@@ -0,0 +1,21 @@
+diff --git a/Makefile b/Makefile
+index 32b7cbb..9ac28a4 100644
+--- a/Makefile
++++ b/Makefile
+@@ -361,12 +361,12 @@ ifndef RISCV
+
+ ifeq ($(UNAME_M),$(filter $(UNAME_M),x86_64 i686 amd64))
+ 	# Use all CPU extensions that are available:
+-	MK_CFLAGS     += -march=native -mtune=native
+-	HOST_CXXFLAGS += -march=native -mtune=native
++	#MK_CFLAGS     += -march=native -mtune=native
++	#HOST_CXXFLAGS += -march=native -mtune=native
+
+ 	# Usage AVX-only
+-	#MK_CFLAGS   += -mfma -mf16c -mavx
+-	#MK_CXXFLAGS += -mfma -mf16c -mavx
++	MK_CFLAGS   += -mfma -mf16c -mavx
++	MK_CXXFLAGS += -mfma -mf16c -mavx
+
+ 	# Usage SSSE3-only (Not is SSE3!)
+ 	#MK_CFLAGS   += -mssse3

cmd/transcriber/apis/whisper.cpp/context.go

@@ -1,6 +1,6 @@
 package whisper
 
-// #cgo linux LDFLAGS: -l:libwhisper.a -lm -lstdc++
+// #cgo linux LDFLAGS: -l:libwhisper.a -lm -lstdc++ -fopenmp
 // #cgo darwin LDFLAGS: -lwhisper -lstdc++ -framework Accelerate
 // #include <whisper.h>
 // #include <stdlib.h>