文中所有相关文件已经上传到资源包中，直接下载
home下创建whisper文件夹
下载
https://github.com/ggerganov/whisper.cpp
上传到whisper目录
解压
下载ffmpeg
wget https://johnvansickle.com/ffmpeg/releases/ffmpeg-release-amd64-static.tar.xz -O ffmpeg.tar.xz创建需要的依赖文件夹
mkdir-p offline-rpms&&cd offline-rpms下载依赖的脚本download_rpms.sh
#!/bin/bash set-e echo"[INFO] 正在下载离线 RPM 包（CentOS 7.9 + EPEL 3.17.5）..."BASE_URL="https://mirrors.aliyun.com/centos-vault/7.9.2009/os/x86_64/Packages"EPEL="https://mirrors.aliyun.com/epel/7/x86_64/Packages"# 基础编译工具 wget--no-check-certificate-O gcc.rpm"$BASE_URL/gcc-4.8.5-44.el7.x86_64.rpm"wget--no-check-certificate-O gcc-c++.rpm"$BASE_URL/gcc-c++-4.8.5-44.el7.x86_64.rpm"wget--no-check-certificate-O make.rpm"$BASE_URL/make-3.82-24.el7.x86_64.rpm"wget--no-check-certificate-O git.rpm"$BASE_URL/git-1.8.3.1-23.el7_8.x86_64.rpm"wget--no-check-certificate-O perl-Error.rpm"$BASE_URL/perl-Error-0.17020-2.el7.noarch.rpm"#cmake33.17.5wget--no-check-certificate-O cmake3.rpm"$EPEL/c/cmake3-3.17.5-1.el7.x86_64.rpm"wget--no-check-certificate-O cmake3-data.rpm"$EPEL/c/cmake3-data-3.17.5-1.el7.noarch.rpm"# 依赖（全部使用阿里云 EPEL 中真实存在的版本） wget--no-check-certificate-O libuv.rpm"$EPEL/l/libuv-1.44.2-1.el7.x86_64.rpm"wget--no-check-certificate-O jsoncpp.rpm"$EPEL/j/jsoncpp-0.10.5-2.el7.x86_64.rpm"# ← 关键修正 wget--no-check-certificate-O rhash.rpm"$EPEL/r/rhash-1.3.4-2.el7.x86_64.rpm"echo"[SUCCESS] 所有 RPM 包下载完成！"echo"下一步："echo" yum localinstall -y *.rpm"echo" ln -sf /usr/bin/cmake3 /usr/bin/cmake"执行下载脚本下载依赖./download_rpms.sh
构建
yum localinstall-y*.rpm 或者 rpm-ivh--nodeps*.rpmcmake链接
ln-sf/usr/bin/cmake3/usr/bin/cmake cmake--version回到上级
cd..创建安装脚本install_whisper.sh
#!/bin/bash set-e RED='\033[0;31m';GREEN='\033[0;32m';YELLOW='\033[1;33m';NC='\033[0m'log(){echo-e"${GREEN}[INFO]${NC} $1";}warn(){echo-e"${YELLOW}[WARN]${NC} $1";}error(){echo-e"${RED}[ERROR]${NC} $1";exit1;}cd"$(dirname "$0")"# 检查 root["$EUID"-ne0]&&error"请用 root 运行"if[-f/opt/rh/devtoolset-9/enable];then source/opt/rh/devtoolset-9/enable fi #===============1.部署 FFmpeg===============log"部署 FFmpeg..."if[!-f ffmpeg.tar.xz];then error"缺少 ffmpeg.tar.xz"fi mkdir-p/opt/whisper/{bin,models}tar-xJf ffmpeg.tar.xz-C/tmp FF_BIN=$(find/tmp-name ffmpeg-type f|head-n1)if[-z"$FF_BIN"];then error"ffmpeg.tar.xz 中未找到 ffmpeg 可执行文件"fi cp"$FF_BIN"/opt/whisper/bin/ffmpeg chmod+x/opt/whisper/bin/ffmpeg ln-sf/opt/whisper/bin/ffmpeg/usr/local/bin/ffmpeg log"FFmpeg 部署完成"#===============2.编译 whisper.cpp===============log"编译 whisper.cpp..."SRC_DIR=""if[-d"whisper.cpp-master"];then SRC_DIR="whisper.cpp-master"elif[-f"whisper.cpp-master.zip"];then log"解压 whisper.cpp-master.zip..."unzip-qo whisper.cpp-master.zip SRC_DIR="whisper.cpp-master"elseerror"缺少 whisper.cpp 源码目录或 ZIP 包"fi cd"$SRC_DIR"# 清理旧构建（可选） rm-rf build # 使用 CMake 构建 cmake-B build-DCMAKE_BUILD_TYPE=Release cmake--build build--config Release-j$(nproc)# 关键：可执行文件在 build/bin/下 BIN_DIR="build/bin"if[!-d"$BIN_DIR"];then error"构建失败：未找到 build/bin 目录"fi # 安装主程序（main->whisper-cpp） mkdir-p/opt/whisper/bin/opt/whisper/models cp"$BIN_DIR/whisper-cli"/opt/whisper/bin/whisper-cpp chmod+x/opt/whisper/bin/whisper-cpp ln-sf/opt/whisper/bin/whisper-cpp/usr/local/bin/whisper-cpp # 复制模型（如果存在）ifls models/*.bin >/dev/null 2>&1; then cp models/*.bin /opt/whisper/models/ log "模型已复制: $(ls /opt/whisper/models/)" else warn "未在 $SRC_DIR/models/ 中发现 .bin 模型文件，请手动放入" fi cd .. # =============== 3. 创建便捷转写命令 =============== cat > /usr/local/bin/whisper-transcribe << 'EOF' #!/bin/bash if [ $# -lt 2 ]; then echo "用法: whisper-transcribe <模型名> <音频文件> [输出文本文件]" echo "示例: whisper-transcribe base input.mp3 result.txt" exit 1 fi MODEL="$1" AUDIO="$2" OUT="${3:-/dev/stdout}" BIN="/opt/whisper/bin/whisper-cpp" MODEL_PATH="/opt/whisper/models/ggml-${MODEL}.bin" [ ! -f "$MODEL_PATH" ] && { echo "❌ 模型不存在: $MODEL_PATH"; exit 1; } [ ! -f "$AUDIO" ] && { echo "❌ 音频文件不存在: $AUDIO"; exit 1; } # 自动转换为 16kHz 单声道 WAV if [[ "$AUDIO" != *.wav ]]; then TMP_WAV="/tmp/whisper_$(basename "$AUDIO" | sha256sum | cut -d' ' -f1).wav" /opt/whisper/bin/ffmpeg -y -i "$AUDIO" -ar 16000 -ac 1 -c:a pcm_s16le "$TMP_WAV" >/dev/null 2>&1 AUDIO="$TMP_WAV" else # 检查采样率和声道（简单处理） SR=$(ffprobe -v quiet -select_streams a:0 -show_entries stream=sample_rate -of csv=p=0 "$AUDIO" 2>/dev/null) CH=$(ffprobe -v quiet -select_streams a:0 -show_entries stream=channels -of csv=p=0 "$AUDIO" 2>/dev/null) if [ "$SR" != "16000" ] || [ "$CH" != "1" ]; then TMP_WAV="/tmp/whisper_$(basename "$AUDIO" .wav)_conv.wav" /opt/whisper/bin/ffmpeg -y -i "$AUDIO" -ar 16000 -ac 1 -c:a pcm_s16le "$TMP_WAV" >/dev/null 2>&1 AUDIO="$TMP_WAV" fi fi # 执行推理并输出文本 "$BIN" -m "$MODEL_PATH" -f "$AUDIO" --output-txt TXT_FILE="${AUDIO%.wav}.wav.txt" if [ -f "$TXT_FILE" ]; then cat "$TXT_FILE" > "$OUT" rm -f "$TMP_WAV" "$TXT_FILE" 2>/dev/null echo "✅ 转写完成 → $OUT" else "$BIN" -m "$MODEL_PATH" -f "$AUDIO" fi EOF chmod +x /usr/local/bin/whisper-transcribe log "✅ Whisper (C++版) 部署成功！" echo echo "📌 使用方法：" echo " whisper-transcribe base your_audio.mp3 output.txt" echo echo "💡 提示：" echo " - 模型需放在 /opt/whisper/models/，如 ggml-base.bin" echo " - 中文推荐模型：base、small（不要用 .en 结尾的）" echo " - 支持 mp3/wav/m4a 等格式，自动转换"执行安装脚本安装./install_whisper.sh
报错c++版本问题
增加一个修复脚本进行修复fix_centos7_repo.sh
#!/bin/bash set-e RED='\033[0;31m';GREEN='\033[0;32m';YELLOW='\033[1;33m';NC='\033[0m'log(){echo-e"${GREEN}[INFO]${NC} $1";}warn(){echo-e"${YELLOW}[WARN]${NC} $1";}error(){echo-e"${RED}[ERROR]${NC} $1";exit1;}["$EUID"-ne0]&&error"请以 root 身份运行"log"检测是否为 CentOS 7..."if!grep-q"CentOS Linux 7"/etc/os-release2>/dev/null;then warn"警告：未检测到 CentOS 7，脚本可能不适用。"read-p"是否继续？(y/N): "-n1-r echo[[!$REPLY=~^[Yy]$]]&&exit1fi log"备份原有 repo 文件..."mkdir-p/etc/yum.repos.d/backup mv/etc/yum.repos.d/*.repo /etc/yum.repos.d/backup/ 2>/dev/null || true log "创建 CentOS 7 Vault Base 源（阿里云镜像）..." cat > /etc/yum.repos.d/CentOS-Base.repo << 'EOF' [base] name=CentOS-7 - Base baseurl=https://mirrors.aliyun.com/centos-vault/7.9.2009/os/$basearch/ gpgcheck=1 gpgkey=https://mirrors.aliyun.com/centos-vault/RPM-GPG-KEY-CentOS-7 enabled=1 [updates] name=CentOS-7 - Updates baseurl=https://mirrors.aliyun.com/centos-vault/7.9.2009/updates/$basearch/ gpgcheck=1 gpgkey=https://mirrors.aliyun.com/centos-vault/RPM-GPG-KEY-CentOS-7 enabled=1 [extras] name=CentOS-7 - Extras baseurl=https://mirrors.aliyun.com/centos-vault/7.9.2009/extras/$basearch/ gpgcheck=1 gpgkey=https://mirrors.aliyun.com/centos-vault/RPM-GPG-KEY-CentOS-7 enabled=1 EOF log "创建 SCL 源（devtoolset），禁用 GPG 检查..." cat > /etc/yum.repos.d/CentOS-SCL.repo << 'EOF' [centos-sclo-rh] name=CentOS-7 - SCLo rh baseurl=https://mirrors.aliyun.com/centos-vault/7.9.2009/sclo/$basearch/rh/ gpgcheck=0 enabled=1 [centos-sclo-sclo] name=CentOS-7 - SCLo sclo baseurl=https://mirrors.aliyun.com/centos-vault/7.9.2009/sclo/$basearch/sclo/ gpgcheck=0 enabled=0 EOF log "清理并重建 YUM 缓存..." yum clean all >/dev/null 2>&1 rm -rf /var/cache/yum yum makecache >/dev/null 2>&1 log "安装 devtoolset-9（GCC 9.x）..." yum install -y devtoolset-9-gcc devtoolset-9-gcc-c++ log "验证 GCC 版本..." source /opt/rh/devtoolset-9/enable gcc_version=$(gcc --version | head -n1) log "GCC 版本: $gcc_version" if [[ $gcc_version != *" 9."* ]]; then error "GCC 9 未正确安装" fi # ============================== # 🔧 自动生成 whisper 构建脚本 # ============================== WHISPER_DIR="$(pwd)" BUILD_SCRIPT="$WHISPER_DIR/build-whisper.sh" log "生成自动构建脚本: $BUILD_SCRIPT" cat > "$BUILD_SCRIPT" << 'EOF' #!/bin/bash # 自动启用 devtoolset-9 并构建 whisper.cpp set -e SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)" cd "$SCRIPT_DIR" if [ -f /opt/rh/devtoolset-9/enable ]; then source /opt/rh/devtoolset-9/enable echo "[INFO] 已启用 devtoolset-9 (GCC $(gcc --version | head -n1 | cut -d' ' -f4))" else echo "[WARN] devtoolset-9 未找到，使用系统默认编译器" fi # 执行原始安装脚本 exec ./install_whisper.sh EOF chmod +x "$BUILD_SCRIPT" log "✅ 修复完成！请使用以下命令构建 whisper：" echo echo " cd $WHISPER_DIR" echo " ./install_whisper.sh" echo echo "💡 此脚本会自动启用 GCC 9，无需手动 source"执行修复脚本./fix_centos7_repo.sh
再次安装./install_whisper.sh
安装成功后
下载模型（科学上网）
https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.bin
将模型文件放到指定路径/opt/whisper/models下
测试：指定音频解析文本
whisper-transcribe base/home/whisper/0d3b1eab-a6dd-4ef5-9560-4d7583124585.mp3 output.txt开放API 访问，响应字符串
脚本：install_whisper_api.sh
#!/bin/bash set-e RED='\033[0;31m';GREEN='\033[0;32m';YELLOW='\033[1;33m';NC='\033[0m'log(){echo-e"${GREEN}[INFO]${NC} $1";}error(){echo-e"${RED}[ERROR]${NC} $1";exit1;}["$EUID"-ne0]&&error"请用 root 运行"WHISPER_SRC="/home/whisper/whisper.cpp-master"API_BIN="/opt/whisper/bin/whisper-api"# 自定义名字，没问题 PORT=6688[!-d"$WHISPER_SRC"]&&error"未找到 whisper.cpp 源码目录: $WHISPER_SRC"cd"$WHISPER_SRC"log"修改 server.cpp 为纯文本输出..."# 注意：源文件名仍是 server.cpp，但编译后叫 whisper-server sed-i's/res\.set_content(result\.dump(), "application\/json");/res.set_content(text, "text\/plain; charset=utf-8");/'examples/server/server.cpp log"清理旧构建..."rm-rf build log"配置 CMake（启用 SERVER + 内置依赖）..."cmake-B build \-DCMAKE_BUILD_TYPE=Release \-DBUILD_SERVER=ON \-DWHISPER_USE_LOCAL_LIBS=ON log"编译 whisper-server（可能需要 1~3 分钟）..."cmake--build build--config Release-j$(nproc)# ✅ 关键修正：检查 whisper-server 而不是 serverif[!-f"build/bin/whisper-server"];then error"编译失败：未生成 build/bin/whisper-server"fi log"安装 API 服务到 /opt/whisper/bin/whisper-api..."mkdir-p/opt/whisper/bin cp build/bin/whisper-server"$API_BIN"chmod+x"$API_BIN"log"创建 systemd 服务..."cat>/etc/systemd/system/whisper-api.service<<EOF[Unit]Description=Whisper.cpp Plain-Text API Server After=network.target[Service]Type=simple User=root WorkingDirectory=/opt/whisper ExecStart=$API_BIN-m/opt/whisper/models/ggml-base.bin--port $PORT Restart=always RestartSec=5[Install]WantedBy=multi-user.targetEOFsystemctl daemon-reload log"✅ 部署成功！"echo echo"📌 启动服务："echo" systemctl start whisper-api"echo" systemctl enable whisper-api"echo echo"📌 手动测试："echo" $API_BIN -m /opt/whisper/models/ggml-base.bin --port 6688"echo echo"📌 API 调用示例："echo" curl -X POST http://localhost:6688/inference -F \"file=@your_audio.mp3\""
执行脚本./install_whisper_api.sh
启动服务
systemctl start whisper-api测试：
curl-X POST http://localhost:6688/inference -F "file=@/home/whisper/0d3b1eab-a6dd-4ef5-9560-4d7583124585.mp3"
我发现这个翻译的不准
换个ggml-small.bin模型
https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-small.bin
下载好之后放到：/opt/whisper/models/
修改 systemd 服务，使用新模型
vi/etc/systemd/system/whisper-api.service将-m参数模型改为：
ExecStart=/opt/whisper/bin/whisper-api-m/opt/whisper/models/ggml-small.bin--port6688保存后重载并重启：
systemctl daemon-reloadsystemctl restart whisper-api我尝试强制指定语言为中文进行解析
curl-X POST"http://localhost:6688/inference?language=zh"-F"file=@/home/whisper/0d3b1eab-a6dd-4ef5-9560-4d7583124585.mp3"我发现怎么返回的是英文呢
[root@localhost whisper]# curl-X POST"http://localhost:6688/inference?language=zh"-F"file=@/home/whisper/0d3b1eab-a6dd-4ef5-9560-4d7583124585.mp3"{"text":" and the image quality is high.\n"}[root@localhost whisper]#更改配置指定中文
vi/etc/systemd/system/whisper-api.serviceExecStart=/opt/whisper/bin/whisper-api-m/opt/whisper/models/ggml-small.bin--language zh--port6688
再次保存重启测试
systemctl daemon-reloadsystemctl restart whisper-apicurl-X POST"http://localhost:6688/inference"-F"file=@/home/whisper/0d3b1eab-a6dd-4ef5-9560-4d7583124585.mp3"虽然有偏差，但最起码是简体中文了
