Compare commits

..

No commits in common. "baec3482a770d72044d30a1507c257691e27b4d1" and "85b67780b1b39aa089f134a91fd9a0e0f9206b94" have entirely different histories.

17 changed files with 7 additions and 1015 deletions

View File

@ -28,7 +28,7 @@ set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake")
# Qt 6
find_package(Qt6 REQUIRED COMPONENTS Core Widgets Concurrent Network DBus)
find_package(Qt6 REQUIRED COMPONENTS Core Widgets Concurrent Network)
qt_standard_project_setup()
# ONNX Runtime
@ -53,9 +53,6 @@ set(SOURCES
src/core/whisper_tokenizer.cpp
src/core/audio_processor.cpp
src/core/vad.cpp
src/core/caps_lock_voice_hotkey.cpp
src/core/wayland_text_injector.cpp
src/core/voice_input_service.cpp
# Audio
src/audio/audio_capture.cpp
@ -90,9 +87,6 @@ set(HEADERS
src/core/whisper_tokenizer.h
src/core/audio_processor.h
src/core/vad.h
src/core/caps_lock_voice_hotkey.h
src/core/wayland_text_injector.h
src/core/voice_input_service.h
src/audio/audio_capture.h
src/audio/audio_decoder.h
@ -130,7 +124,6 @@ target_link_libraries(${PROJECT_NAME} PRIVATE
Qt6::Widgets
Qt6::Concurrent
Qt6::Network
Qt6::DBus
${ONNXRUNTIME_LIBRARIES}
${PORTAUDIO_LIBRARIES}
pthread

View File

@ -74,9 +74,7 @@ void ConfigManager::loadDefaults() {
{"language", "zh"},
{"streaming", true},
{"beam_size", 5},
{"temperature", 0.0},
{"debug_save_audio", false},
{"capslock_voice_enabled", false}
{"temperature", 0.0}
}},
{"audio", QVariantMap{
{"input_device", -1},

View File

@ -7,7 +7,7 @@
#define DR_WAV_IMPLEMENTATION
#define DR_MP3_IMPLEMENTATION
#define DR_FLAC_IMPLEMENTATION
#include "dr_wav.h"
#include <dr_wav.h>
#include <dr_mp3.h>
#include <dr_flac.h>
#endif

View File

@ -1,222 +0,0 @@
#include "caps_lock_voice_hotkey.h"
#include "utils/logger.h"
#include <QDBusConnection>
#include <QDBusInterface>
#include <QDBusMessage>
#include <QDBusObjectPath>
#include <QCoreApplication>
#include <QUuid>
static const char* const kTag = "CapsLockVoiceHotkey";
// Portal 常量
static const char* const kPortalService = "org.freedesktop.portal.Desktop";
static const char* const kPortalObjectPath = "/org/freedesktop/portal/desktop";
static const char* const kGlobalShortcutsIface = "org.freedesktop.portal.GlobalShortcuts";
static const char* const kRequestIface = "org.freedesktop.portal.Request";
namespace impress {
struct CapsLockVoiceHotkey::Impl {
QString sessionPath;
QString pendingRequestPath;
enum State { Idle, WaitingSession, WaitingBind, Active };
State state = Idle;
/** 生成唯一 token */
static QString makeToken(const QString& prefix) {
return prefix + "_" + QUuid::createUuid().toString().mid(1, 8);
}
/** 构造 session path从 sender 名和 token */
static QString makeSessionPath(const QString& sender, const QString& token) {
QString safeSender = sender;
safeSender.remove(0, 1); // 去掉前导 ':'
safeSender.replace('.', '_');
return QString("/org/freedesktop/portal/desktop/session/%1/%2")
.arg(safeSender, token);
}
/** 获取 session bus */
static QDBusConnection bus() {
return QDBusConnection::sessionBus();
}
};
CapsLockVoiceHotkey::CapsLockVoiceHotkey(QObject* parent)
: QObject(parent)
, impl_(std::make_unique<Impl>())
{}
CapsLockVoiceHotkey::~CapsLockVoiceHotkey() {
stop();
}
bool CapsLockVoiceHotkey::start() {
if (active_) return true;
QDBusConnection bus = Impl::bus();
if (!bus.isConnected()) {
emit error("无法连接到 D-Bus session bus");
return false;
}
// 连接信号
bus.connect(kPortalService, kPortalObjectPath,
kGlobalShortcutsIface, "Activated",
this, SLOT(handleActivated(QString)));
bus.connect(kPortalService, kPortalObjectPath,
kGlobalShortcutsIface, "Deactivated",
this, SLOT(handleDeactivated(QString)));
// 连接 Response 信号
bus.connect(kPortalService, QString(),
kRequestIface, "Response",
this, SLOT(onPortalResponse(uint, QVariantMap)));
// 发送 CreateSession
QDBusInterface portal(kPortalService, kPortalObjectPath,
kGlobalShortcutsIface, bus);
QString sessionToken = Impl::makeToken("io_impress_sess");
QString requestToken = Impl::makeToken("io_impress_req");
QVariantMap options;
options["handle_token"] = requestToken;
options["session_handle_token"] = sessionToken;
QDBusMessage reply = portal.call("CreateSession", options);
if (reply.type() == QDBusMessage::ErrorMessage) {
emit error(QString("CreateSession 失败: %1").arg(reply.errorMessage()));
LOG_ERROR(kTag, reply.errorMessage());
return false;
}
// 保存预期 session path
QString sender = bus.baseService();
impl_->sessionPath = Impl::makeSessionPath(sender, sessionToken);
impl_->state = Impl::WaitingSession;
LOG_INFO(kTag, "CreateSession 已发送,等待用户授权...");
LOG_DEBUG(kTag, QString("Session path: %1").arg(impl_->sessionPath));
return true;
}
void CapsLockVoiceHotkey::stop() {
if (!active_ && impl_->state == Impl::Idle) return;
QDBusConnection bus = Impl::bus();
bus.disconnect(kPortalService, kPortalObjectPath,
kGlobalShortcutsIface, "Activated",
this, SLOT(handleActivated(QString)));
bus.disconnect(kPortalService, kPortalObjectPath,
kGlobalShortcutsIface, "Deactivated",
this, SLOT(handleDeactivated(QString)));
bus.disconnect(kPortalService, QString(),
kRequestIface, "Response",
this, SLOT(onPortalResponse(uint, QVariantMap)));
active_ = false;
recording_ = false;
impl_->state = Impl::Idle;
impl_->sessionPath.clear();
LOG_INFO(kTag, "CapsLock 语音快捷键已停止");
}
void CapsLockVoiceHotkey::onPortalResponse(uint response, const QVariantMap& results) {
if (impl_->state == Impl::WaitingSession) {
handleSessionResponse(response, results);
} else if (impl_->state == Impl::WaitingBind) {
handleBindResponse(response, results);
}
}
void CapsLockVoiceHotkey::handleSessionResponse(uint response, const QVariantMap& results) {
if (impl_->state != Impl::WaitingSession) return;
if (response != 0) {
emit error(QString("Session 被拒绝 (response=%1)").arg(response));
LOG_ERROR(kTag, QString("Session 被拒绝: %1").arg(response));
impl_->state = Impl::Idle;
return;
}
QString actualPath = results.value("session_handle").toString();
if (!actualPath.isEmpty()) {
impl_->sessionPath = actualPath;
}
LOG_INFO(kTag, QString("Session 已授权: %1").arg(impl_->sessionPath));
// 发送 BindShortcuts
impl_->state = Impl::WaitingBind;
QDBusInterface portal(kPortalService, kPortalObjectPath,
kGlobalShortcutsIface, Impl::bus());
QString bindToken = Impl::makeToken("io_impress_bind");
QVariantMap shortcutProps;
shortcutProps["description"] = "语音输入CapsLock";
QList<QVariant> shortcuts;
QVariantMap shortcutEntry;
shortcutEntry["id"] = "voice_input";
shortcutEntry["properties"] = shortcutProps;
shortcuts.append(shortcutEntry);
QVariantMap bindOptions;
bindOptions["handle_token"] = bindToken;
QDBusMessage reply = portal.call("BindShortcuts",
QDBusObjectPath(impl_->sessionPath),
shortcuts,
QString(), // parent_window (空 = Wayland 模式)
bindOptions);
if (reply.type() == QDBusMessage::ErrorMessage) {
emit error(QString("BindShortcuts 失败: %1").arg(reply.errorMessage()));
LOG_ERROR(kTag, reply.errorMessage());
impl_->state = Impl::Idle;
return;
}
impl_->pendingRequestPath = reply.arguments().isEmpty() ?
QString() : reply.arguments()[0].toString();
LOG_INFO(kTag, "BindShortcuts 已发送,等待用户设置快捷键...");
}
void CapsLockVoiceHotkey::handleBindResponse(uint response, const QVariantMap&) {
if (impl_->state != Impl::WaitingBind) return;
if (response != 0) {
emit error(QString("快捷键绑定被拒绝 (response=%1)").arg(response));
LOG_ERROR(kTag, QString("Bind 被拒绝: %1").arg(response));
impl_->state = Impl::Idle;
return;
}
// 快捷键绑定成功
active_ = true;
impl_->state = Impl::Active;
emit ready();
LOG_INFO(kTag, "快捷键已注册CapsLock 语音输入已就绪");
}
void CapsLockVoiceHotkey::handleActivated(const QString& shortcutId) {
if (!active_) return;
LOG_DEBUG(kTag, QString("快捷键按下: %1").arg(shortcutId));
recording_ = true;
emit recordingStarted();
}
void CapsLockVoiceHotkey::handleDeactivated(const QString& shortcutId) {
if (!active_) return;
LOG_DEBUG(kTag, QString("快捷键松开: %1").arg(shortcutId));
recording_ = false;
emit recordingStopped();
}
} // namespace impress

View File

@ -1,66 +0,0 @@
#pragma once
#include <QObject>
#include <QString>
#include <QTimer>
#include <memory>
namespace impress {
/**
* @brief CapsLock
*
* 使 freedesktop GlobalShortcuts D-Bus Portal Wayland
*
* 1. CapsLock 1
* 2.
* 3. CapsLock
* 4. < 1s CapsLock
*
* GNOME
*/
class CapsLockVoiceHotkey : public QObject {
Q_OBJECT
public:
explicit CapsLockVoiceHotkey(QObject* parent = nullptr);
~CapsLockVoiceHotkey() override;
/** @brief 初始化并注册快捷键(首次需要用户授权) */
bool start();
/** @brief 停止并注销快捷键 */
void stop();
/** @brief 是否已激活 */
bool isActive() const { return active_; }
/** @brief 当前是否正在录音CapsLock 长按超过 1s 后) */
bool isRecording() const { return recording_; }
signals:
/** @brief 开始录音(长按超过 1 秒后) */
void recordingStarted();
/** @brief 停止录音(松开快捷键后) */
void recordingStopped();
/** @brief 快捷键已注册(用户授权后) */
void ready();
/** @brief 初始化失败 */
void error(const QString& message);
private:
struct Impl;
std::unique_ptr<Impl> impl_;
bool active_ = false;
bool recording_ = false;
void handleSessionResponse(uint response, const QVariantMap& results);
void handleBindResponse(uint response, const QVariantMap& results);
void handleActivated(const QString& shortcutId);
void handleDeactivated(const QString& shortcutId);
void onPortalResponse(uint response, const QVariantMap& results);
};
} // namespace impress

View File

@ -12,10 +12,6 @@
#include <QMutex>
#include <QMutexLocker>
#include <QFileInfo>
#include <QDir>
#include <QDateTime>
#include <QFile>
#include <QDataStream>
#include <algorithm>
#include <cmath>
@ -26,54 +22,6 @@
static const char* const kTag = "SenseVoiceEngine";
/**
* @brief WAV dr_wav
*/
static bool saveWav16(const QString& path, const std::vector<float>& samples, int sampleRate) {
QFile file(path);
if (!file.open(QIODevice::WriteOnly)) return false;
int numSamples = static_cast<int>(samples.size());
int dataSize = numSamples * 2; // 16-bit mono
int totalSize = 36 + dataSize; // RIFF header size
quint16 audioFormat = 1; // PCM
quint16 numChannels = 1;
quint32 byteRate = sampleRate * 2; // sampleRate * numChannels * bitsPerSample/8
quint16 blockAlign = 2; // numChannels * bitsPerSample/8
quint16 bitsPerSample = 16;
QDataStream out(&file);
out.setByteOrder(QDataStream::LittleEndian);
// RIFF header
out.writeRawData("RIFF", 4);
out << static_cast<quint32>(totalSize);
out.writeRawData("WAVE", 4);
// fmt chunk
out.writeRawData("fmt ", 4);
out << static_cast<quint32>(16); // chunk size
out << audioFormat;
out << numChannels;
out << static_cast<quint32>(sampleRate);
out << byteRate;
out << blockAlign;
out << bitsPerSample;
// data chunk
out.writeRawData("data", 4);
out << static_cast<quint32>(dataSize);
// PCM data (float → int16)
for (float s : samples) {
s = std::max(-1.0f, std::min(1.0f, s)); // clip
qint16 val = static_cast<qint16>(s * 32767.0f);
out << val;
}
return file.error() == QFile::FileError::NoError;
}
namespace impress {
/** 语言代码映射 */
@ -279,11 +227,6 @@ bool SenseVoiceEngine::isLoaded() const {
return loaded_;
}
void SenseVoiceEngine::setDebugSaveAudio(bool enable) {
debugSaveAudio_ = enable;
LOG_INFO(kTag, QString("调试录音保存: %1").arg(enable ? "开启" : "关闭"));
}
/** CTC 贪婪解码:去重 + 去除空白 */
static std::vector<int> ctcGreedyDecode(const std::vector<int>& tokens, int blankToken) {
std::vector<int> result;
@ -343,53 +286,9 @@ RecognitionResult SenseVoiceEngine::infer(const std::vector<float>& samples,
try {
// 1. 重采样到 16kHz
Timer preprocessTimer;
// 调试模式:保存原始音频到 WAV 文件
if (debugSaveAudio_ && !samples.empty()) {
QString debugDir = "/tmp/impress_audio_debug";
QDir dir;
if (!dir.exists(debugDir)) {
dir.mkpath(debugDir);
}
QString timestamp = QDateTime::currentDateTime().toString("yyyyMMdd_HHmmss_zzz");
QString wavPath = QString("%1/audio_%2_%3Hz.wav")
.arg(debugDir).arg(timestamp).arg(sampleRate);
if (saveWav16(wavPath, samples, sampleRate)) {
LOG_DEBUG(kTag, QString("调试音频已保存: %1 (%2 样本, %3Hz)")
.arg(wavPath).arg(samples.size()).arg(sampleRate));
} else {
LOG_WARNING(kTag, QString("无法创建调试音频文件: %1").arg(wavPath));
}
}
std::vector<float> processedSamples = samples;
int currentSampleRate = sampleRate;
// 计算输入音频 RMS 电平用于诊断
double rms = 0.0;
bool hasNaN = false;
for (float s : samples) {
if (std::isnan(s) || std::isinf(s)) { hasNaN = true; break; }
rms += s * s;
}
rms = std::sqrt(rms / samples.size());
if (hasNaN) {
result.text = "[错误] 输入音频包含 NaN/Inf 值,请检查麦克风设备";
result.latency_ms = timer.elapsedMs();
LOG_ERROR(kTag, QString("输入音频包含无效值 (NaN/Inf), 样本数: %1").arg(samples.size()));
return result;
}
if (rms < 1e-6) {
result.text = "";
result.latency_ms = timer.elapsedMs();
LOG_DEBUG(kTag, QString("静音段 (RMS: %1), 跳过推理").arg(rms, 0, 'f', 6));
return result;
}
LOG_DEBUG(kTag, QString("输入音频 RMS: %1 (样本数: %2)").arg(rms, 0, 'f', 6).arg(samples.size()));
if (sampleRate != 16000) {
AudioProcessor processor(16000);
processedSamples = processor.resample(samples, sampleRate);
@ -471,14 +370,13 @@ RecognitionResult SenseVoiceEngine::infer(const std::vector<float>& samples,
for (int t = 0; t < seqLen; t++) {
int offset = t * vocabSize;
int bestAbsIdx = argmax(logitsData, offset, offset + vocabSize);
int bestToken = bestAbsIdx - offset; // 绝对索引 → token ID
int bestToken = argmax(logitsData, offset, offset + vocabSize);
if (bestToken != SenseVoiceTokenizer::kTokenBlank) {
rawTokens.push_back(bestToken);
// 计算置信度
float maxLogit = logitsData[bestAbsIdx];
float maxLogit = logitsData[offset + bestToken];
// 近似置信度: 使用 softmax 的最大值位置
totalConf += maxLogit;
confCount++;
@ -500,8 +398,8 @@ RecognitionResult SenseVoiceEngine::infer(const std::vector<float>& samples,
result.text = "";
} else if (impl_->tokenizer.isLoaded()) {
result.text = impl_->tokenizer.decode(decodedTokens);
LOG_DEBUG(kTag, QString("解码文本: %1 个 token → %2 字符: %3")
.arg(decodedTokens.size()).arg(result.text.length()).arg(result.text));
LOG_DEBUG(kTag, QString("解码文本: %1 个 token → %2 字符")
.arg(decodedTokens.size()).arg(result.text.length()));
} else {
// 降级:输出 token ID
QString decodedText;

View File

@ -48,9 +48,6 @@ public:
int sampleRate,
const QString& language = QString());
/** @brief 设置调试模式:开启后每次推理保存音频到 WAV */
void setDebugSaveAudio(bool enable);
signals:
void modelLoaded(const QString& modelPath);
void modelLoadError(const QString& modelPath, const QString& error);
@ -61,7 +58,6 @@ private:
struct Impl;
std::unique_ptr<Impl> impl_;
bool loaded_ = false;
bool debugSaveAudio_ = false;
};
} // namespace impress

View File

@ -1,236 +0,0 @@
#include "voice_input_service.h"
#include "audio/audio_capture.h"
#include "sense_voice_engine.h"
#include "caps_lock_voice_hotkey.h"
#include "wayland_text_injector.h"
#include "app/config_manager.h"
#include "utils/logger.h"
#include <QThread>
#include <QTimer>
#include <QtConcurrent>
static const char* const kTag = "VoiceInputService";
namespace impress {
struct VoiceInputService::Impl {
AudioCapture* audioCapture = nullptr;
SenseVoiceEngine* sttEngine = nullptr;
CapsLockVoiceHotkey* hotkey = nullptr;
WaylandTextInjector* injector = nullptr;
};
VoiceInputService::VoiceInputService(ConfigManager* configManager, QObject* parent)
: QObject(parent)
, configManager_(configManager)
, impl_(std::make_unique<Impl>())
{
longPressTimer_ = new QTimer(this);
longPressTimer_->setSingleShot(true);
connect(longPressTimer_, &QTimer::timeout, this, [this]() {
// 长按超时仍未松开 → 确认为长按录音
if (!longPressDetected_) {
longPressDetected_ = true;
emit statusChanged("正在录音...");
}
});
}
VoiceInputService::~VoiceInputService() {
stop();
}
bool VoiceInputService::start() {
if (running_) return true;
// 1. 初始化音频采集
impl_->audioCapture = new AudioCapture(this);
connect(impl_->audioCapture, &AudioCapture::audioDataReady,
this, &VoiceInputService::onAudioData);
// 2. 初始化 STT 引擎并加载模型
impl_->sttEngine = new SenseVoiceEngine(this);
// 从配置加载模型
QString modelPath = configManager_->get("stt.model_path").toString();
QString tokensPath = configManager_->get("stt.tokens_path").toString();
QString device = configManager_->get("stt.device").toString();
int numThreads = configManager_->get("stt.num_threads").toInt();
if (!modelPath.isEmpty()) {
LOG_INFO(kTag, QString("正在加载 STT 模型: %1").arg(modelPath));
bool modelLoaded = impl_->sttEngine->loadModelSync(modelPath, tokensPath, device, numThreads);
if (!modelLoaded) {
emit error(QString("STT 模型加载失败: %1").arg(modelPath));
LOG_ERROR(kTag, "STT 模型加载失败");
} else {
LOG_INFO(kTag, "STT 模型加载成功");
// 同步调试音频设置
bool debugSave = configManager_->get("stt.debug_save_audio").toBool();
impl_->sttEngine->setDebugSaveAudio(debugSave);
}
} else {
LOG_WARNING(kTag, "模型路径为空,请先在配置中设置模型路径");
}
// 3. 初始化全局快捷键
impl_->hotkey = new CapsLockVoiceHotkey(this);
connect(impl_->hotkey, &CapsLockVoiceHotkey::recordingStarted,
this, &VoiceInputService::onHotkeyActivated);
connect(impl_->hotkey, &CapsLockVoiceHotkey::recordingStopped,
this, &VoiceInputService::onHotkeyDeactivated);
connect(impl_->hotkey, &CapsLockVoiceHotkey::ready,
this, [this]() {
emit statusChanged("语音输入就绪(快捷键已注册)");
});
connect(impl_->hotkey, &CapsLockVoiceHotkey::error,
this, &VoiceInputService::error);
// 4. 初始化文本注入器
impl_->injector = new WaylandTextInjector(this);
if (!impl_->injector->initialize()) {
emit error("文本注入器初始化失败,无法注入识别结果");
LOG_ERROR(kTag, "文本注入器初始化失败");
}
// 启动快捷键(首次会弹出授权对话框)
if (!impl_->hotkey->start()) {
emit error("全局快捷键启动失败");
return false;
}
running_ = true;
emit statusChanged("语音输入已启动(等待授权...");
LOG_INFO(kTag, "语音输入服务已启动");
return true;
}
void VoiceInputService::stop() {
if (!running_) return;
longPressTimer_->stop();
if (impl_->audioCapture) {
impl_->audioCapture->stop();
}
if (impl_->sttEngine) {
impl_->sttEngine->unloadModel();
}
if (impl_->hotkey) {
impl_->hotkey->stop();
}
running_ = false;
recording_ = false;
longPressDetected_ = false;
audioBuffer_.clear();
LOG_INFO(kTag, "语音输入服务已停止");
}
void VoiceInputService::onHotkeyActivated() {
LOG_DEBUG(kTag, "快捷键激活(按下)");
recording_ = true;
longPressDetected_ = false;
audioBuffer_.clear();
// 启动长按定时器
longPressTimer_->start(longPressThreshold_);
// 开始音频采集(后台预采集)
int deviceIndex = -1; // 默认设备
impl_->audioCapture->start(deviceIndex, 16000, 20);
emit statusChanged("等待长按确认...");
}
void VoiceInputService::onHotkeyDeactivated() {
LOG_DEBUG(kTag, "快捷键停用(松开)");
recording_ = false;
longPressTimer_->stop();
// 停止音频采集
if (impl_->audioCapture && impl_->audioCapture->isRunning()) {
impl_->audioCapture->stop();
}
if (!longPressDetected_) {
// 短按 → 模拟 CapsLock 按键
LOG_DEBUG(kTag, "短按,模拟 CapsLock");
simulateCapsLock();
emit statusChanged("短按:切换 CapsLock");
} else {
// 长按 → 停止录音并转写
stopRecordingAndTranscribe();
}
longPressDetected_ = false;
}
void VoiceInputService::onAudioData(const std::vector<float>& samples, int sampleRate) {
if (!recording_) return;
audioSampleRate_ = sampleRate;
audioBuffer_.insert(audioBuffer_.end(), samples.begin(), samples.end());
}
void VoiceInputService::stopRecordingAndTranscribe() {
if (audioBuffer_.empty()) {
emit statusChanged("未检测到音频输入");
return;
}
emit statusChanged("正在识别...");
QString language = configManager_->get("stt.language").toString();
(void)QtConcurrent::run([this, buffer = audioBuffer_, lang = language]() {
QString text;
if (!impl_->sttEngine->isLoaded()) {
LOG_WARNING(kTag, "模型未加载,跳过推理");
text = "[错误] 模型未加载,请先在配置中设置模型路径";
} else {
auto result = impl_->sttEngine->infer(buffer, audioSampleRate_, lang);
text = result.text;
}
QMetaObject::invokeMethod(this, [this, text]() {
onRecognitionComplete(text);
}, Qt::QueuedConnection);
});
audioBuffer_.clear();
}
void VoiceInputService::onRecognitionComplete(const QString& text) {
if (text.isEmpty()) {
emit statusChanged("识别结果:无语音输入");
return;
}
emit recognitionResult(text);
emit statusChanged(QString("识别结果: %1").arg(text));
// 注入文本到光标位置
if (impl_->injector && impl_->injector->isInitialized()) {
impl_->injector->injectText(text);
LOG_INFO(kTag, QString("文本已注入: %1").arg(text));
} else {
LOG_WARNING(kTag, "文本注入器未就绪,无法注入");
}
}
void VoiceInputService::simulateCapsLock() {
if (impl_->injector && impl_->injector->isInitialized()) {
// CapsLock keysym = 0xffe5
unsigned int capslockKeysym = 0xffe5;
impl_->injector->simulateKeycode(capslockKeysym);
LOG_DEBUG(kTag, "模拟 CapsLock 按键已注入");
} else {
LOG_WARNING(kTag, "文本注入器未初始化,无法模拟 CapsLock");
}
}
} // namespace impress

View File

@ -1,80 +0,0 @@
#pragma once
#include <QObject>
#include <QString>
#include <QTimer>
#include <vector>
#include <memory>
namespace impress {
class AudioCapture;
class SenseVoiceEngine;
class CapsLockVoiceHotkey;
class WaylandTextInjector;
class ConfigManager;
/**
* @brief CapsLock
*
* STT
*
* 1.
* 2. 1
* 3.
* 4. < 1 CapsLock
*/
class VoiceInputService : public QObject {
Q_OBJECT
public:
explicit VoiceInputService(ConfigManager* configManager, QObject* parent = nullptr);
~VoiceInputService() override;
/** @brief 启动服务(初始化所有组件) */
bool start();
/** @brief 停止服务 */
void stop();
/** @brief 是否已启动 */
bool isRunning() const { return running_; }
/** @brief 是否正在录音 */
bool isRecording() const { return recording_; }
/** @brief 长按阈值(毫秒),默认 1000ms */
void setLongPressThreshold(int ms) { longPressThreshold_ = ms; }
int longPressThreshold() const { return longPressThreshold_; }
signals:
void statusChanged(const QString& status);
void recognitionResult(const QString& text);
void error(const QString& message);
private slots:
void onHotkeyActivated();
void onHotkeyDeactivated();
void onAudioData(const std::vector<float>& samples, int sampleRate);
void onRecognitionComplete(const QString& text);
private:
struct Impl;
ConfigManager* configManager_ = nullptr;
std::unique_ptr<Impl> impl_;
bool running_ = false;
bool recording_ = false;
bool longPressDetected_ = false;
int longPressThreshold_ = 1000;
std::vector<float> audioBuffer_;
int audioSampleRate_ = 16000;
QTimer* longPressTimer_ = nullptr;
void startRecording();
void stopRecordingAndTranscribe();
void simulateCapsLock();
};
} // namespace impress

View File

@ -1,181 +0,0 @@
#include "wayland_text_injector.h"
#include "utils/logger.h"
#include <QLibrary>
#include <QThread>
static const char* const kTag = "WaylandTextInjector";
namespace impress {
// XTest 函数指针类型
typedef int (*XTestFakeKeyEventFn)(void* display, unsigned int keycode,
int is_press, unsigned long delay);
typedef void* (*XOpenDisplayFn)(const char* display_name);
typedef int (*XCloseDisplayFn)(void* display);
typedef unsigned int (*XKeysymToKeycodeFn)(void* display, unsigned long keysym);
typedef unsigned long (*XStringToKeysymFn)(const char* str);
typedef int (*XSyncFn)(void* display, int discard);
struct WaylandTextInjector::Impl {
QLibrary x11Lib;
QLibrary xtstLib;
XOpenDisplayFn XOpenDisplay = nullptr;
XCloseDisplayFn XCloseDisplay = nullptr;
XKeysymToKeycodeFn XKeysymToKeycode = nullptr;
XStringToKeysymFn XStringToKeysym = nullptr;
XSyncFn XSyncFnPtr = nullptr;
XTestFakeKeyEventFn XTestFakeKeyEvent = nullptr;
void* display = nullptr;
bool loadLibraries() {
// 加载 libX11
x11Lib.setFileName("libX11.so.6");
if (!x11Lib.load()) {
LOG_ERROR(kTag, QString("无法加载 libX11: %1").arg(x11Lib.errorString()));
return false;
}
// 加载 libXtst
xtstLib.setFileName("libXtst.so.6");
if (!xtstLib.load()) {
LOG_ERROR(kTag, QString("无法加载 libXtst: %1").arg(xtstLib.errorString()));
return false;
}
// 解析 X11 符号
XOpenDisplay = reinterpret_cast<XOpenDisplayFn>(x11Lib.resolve("XOpenDisplay"));
XCloseDisplay = reinterpret_cast<XCloseDisplayFn>(x11Lib.resolve("XCloseDisplay"));
XKeysymToKeycode = reinterpret_cast<XKeysymToKeycodeFn>(x11Lib.resolve("XKeysymToKeycode"));
XStringToKeysym = reinterpret_cast<XStringToKeysymFn>(x11Lib.resolve("XStringToKeysym"));
XSyncFnPtr = reinterpret_cast<XSyncFn>(x11Lib.resolve("XSync"));
// 解析 XTest 符号
XTestFakeKeyEvent = reinterpret_cast<XTestFakeKeyEventFn>(
xtstLib.resolve("XTestFakeKeyEvent"));
if (!XOpenDisplay || !XCloseDisplay || !XKeysymToKeycode ||
!XSyncFnPtr || !XTestFakeKeyEvent) {
LOG_ERROR(kTag, "无法解析 X11/XTest 符号");
return false;
}
// 打开 X11 显示(通过 XWayland
display = XOpenDisplay(nullptr);
if (!display) {
display = XOpenDisplay(":0");
}
if (!display) {
LOG_ERROR(kTag, "无法连接 X11 显示XWayland");
return false;
}
LOG_INFO(kTag, "XTest 文本注入器已初始化");
return true;
}
};
WaylandTextInjector::WaylandTextInjector(QObject* parent)
: QObject(parent)
, impl_(std::make_unique<Impl>())
{}
WaylandTextInjector::~WaylandTextInjector() {
if (impl_->display && impl_->XCloseDisplay) {
impl_->XCloseDisplay(impl_->display);
}
}
bool WaylandTextInjector::initialize() {
if (initialized_) return true;
initialized_ = impl_->loadLibraries();
return initialized_;
}
bool WaylandTextInjector::injectText(const QString& text) {
if (!initialized_) {
LOG_ERROR(kTag, "文本注入器未初始化");
return false;
}
if (text.isEmpty()) return true;
LOG_DEBUG(kTag, QString("注入文本 (%1 字符): %2").arg(text.length()).arg(text));
for (int i = 0; i < text.length(); i++) {
if (!injectChar(text[i])) {
LOG_WARNING(kTag, QString("字符注入失败: '%1'").arg(text[i]));
}
// 字符间短暂延迟
QThread::usleep(10000); // 10ms
}
LOG_DEBUG(kTag, "文本注入完成");
return true;
}
bool WaylandTextInjector::injectChar(QChar ch) {
if (!impl_->display) return false;
// 处理常见字符映射
unsigned long keysym;
if (ch.isLetterOrNumber() || ch.isPunct() || ch.isSymbol()) {
// ASCII 字符直接使用 keysym
keysym = ch.unicode();
} else if (ch == '\n' || ch == '\r') {
keysym = 0xff0d; // XK_Return
} else if (ch == '\t') {
keysym = 0xff09; // XK_Tab
} else if (ch == ' ') {
keysym = 0x020; // XK_space
} else {
// 尝试通过 XStringToKeysym 解析
QByteArray ba = QString(ch).toUtf8();
keysym = impl_->XStringToKeysym(ba.constData());
if (keysym == 0) {
return false; // 不支持的字符
}
}
unsigned int keycode = impl_->XKeysymToKeycode(impl_->display, keysym);
if (keycode == 0) return false;
// Shift 处理(大写字母需要按住 Shift
bool needShift = ch.isUpper() && ch.isLetter();
if (needShift) {
unsigned int shiftCode = impl_->XKeysymToKeycode(impl_->display, 0xffe1); // XK_Shift_L
if (shiftCode) {
impl_->XTestFakeKeyEvent(impl_->display, shiftCode, 1, 0);
}
}
// 按键按下 + 释放
impl_->XTestFakeKeyEvent(impl_->display, keycode, 1, 0);
impl_->XTestFakeKeyEvent(impl_->display, keycode, 0, 0);
if (needShift) {
unsigned int shiftCode = impl_->XKeysymToKeycode(impl_->display, 0xffe1);
if (shiftCode) {
impl_->XTestFakeKeyEvent(impl_->display, shiftCode, 0, 0);
}
}
impl_->XSyncFnPtr(impl_->display, 0);
return true;
}
bool WaylandTextInjector::simulateKeycode(unsigned int keycode) {
if (!impl_->display || !impl_->XTestFakeKeyEvent) return false;
LOG_DEBUG(kTag, QString("模拟 keycode: 0x%1").arg(keycode, 0, 16));
// 按下 + 释放
impl_->XTestFakeKeyEvent(impl_->display, keycode, 1, 0);
impl_->XTestFakeKeyEvent(impl_->display, keycode, 0, 0);
impl_->XSyncFnPtr(impl_->display, 0);
return true;
}
} // namespace impress

View File

@ -1,44 +0,0 @@
#pragma once
#include <QObject>
#include <QString>
#include <memory>
namespace impress {
/**
* @brief
*
* XTest (XWayland) RemoteDesktop Portal
* 使 dlopen libXtst XTest
*/
class WaylandTextInjector : public QObject {
Q_OBJECT
public:
explicit WaylandTextInjector(QObject* parent = nullptr);
~WaylandTextInjector() override;
/** @brief 初始化(加载 XTest 库) */
bool initialize();
/** @brief 将文本注入到当前光标位置 */
bool injectText(const QString& text);
/** @brief 是否已初始化 */
bool isInitialized() const { return initialized_; }
/** @brief 模拟 X11 keycode 按下+释放(用于 CapsLock 等系统按键) */
bool simulateKeycode(unsigned int keycode);
signals:
void error(const QString& message);
private:
struct Impl;
std::unique_ptr<Impl> impl_;
bool initialized_ = false;
bool injectChar(QChar ch);
};
} // namespace impress

View File

@ -174,10 +174,6 @@ void FileTranscribePage::onStartTranscribe() {
return;
}
// 从配置同步调试开关到引擎
sttEngine_->setDebugSaveAudio(
configManager_->get("stt.debug_save_audio").toBool());
isTranscribing_ = true;
currentTaskIndex_ = 0;
progressBar_->setVisible(true);

View File

@ -2,7 +2,6 @@
#include "stt_test_page.h"
#include "file_transcribe_page.h"
#include "settings_page.h"
#include "core/voice_input_service.h"
#include "app/config_manager.h"
#include "utils/logger.h"
@ -27,28 +26,6 @@ MainWindow::MainWindow(ConfigManager* configManager, QWidget* parent)
setupMenuBar();
loadStyleSheet();
// 初始化语音输入服务
voiceInputService_ = new VoiceInputService(configManager_, this);
connect(voiceInputService_, &VoiceInputService::statusChanged,
this, [this](const QString& status) {
LOG_DEBUG(kTag, QString("语音输入状态: %1").arg(status));
});
connect(voiceInputService_, &VoiceInputService::error,
this, [this](const QString& err) {
LOG_ERROR(kTag, err);
});
connect(voiceInputService_, &VoiceInputService::recognitionResult,
this, [this](const QString& text) {
LOG_INFO(kTag, QString("语音识别结果: %1").arg(text));
});
// 监听配置变化,动态启停语音输入服务
connect(configManager_, &ConfigManager::configChanged,
this, &MainWindow::onVoiceInputConfigChanged);
// 启动时检查配置
onVoiceInputConfigChanged();
LOG_INFO(kTag, "主窗口已创建");
}
@ -102,24 +79,8 @@ void MainWindow::loadStyleSheet() {
}
void MainWindow::closeEvent(QCloseEvent* event) {
if (voiceInputService_) {
voiceInputService_->stop();
}
LOG_INFO(kTag, "主窗口关闭");
QMainWindow::closeEvent(event);
}
void MainWindow::onVoiceInputConfigChanged() {
if (!voiceInputService_) return;
bool enabled = configManager_->get("stt.capslock_voice_enabled").toBool();
if (enabled && !voiceInputService_->isRunning()) {
voiceInputService_->start();
LOG_INFO(kTag, "CapsLock 语音输入已启用");
} else if (!enabled && voiceInputService_->isRunning()) {
voiceInputService_->stop();
LOG_INFO(kTag, "CapsLock 语音输入已关闭");
}
}
} // namespace impress

View File

@ -10,7 +10,6 @@ class ConfigManager;
class STTTestPage;
class FileTranscribePage;
class SettingsPage;
class VoiceInputService;
/**
* @brief
@ -30,10 +29,8 @@ private:
void setupUI();
void setupMenuBar();
void loadStyleSheet();
void onVoiceInputConfigChanged();
ConfigManager* configManager_;
VoiceInputService* voiceInputService_;
STTTestPage* sttPage_;
FileTranscribePage* transcribePage_;
SettingsPage* settingsPage_;

View File

@ -83,14 +83,6 @@ void SettingsPage::setupUI() {
streamingCheck_->setChecked(true);
sttLayout->addRow("流式识别:", streamingCheck_);
debugSaveAudioCheck_ = new QCheckBox("保存调试音频到 /tmp/impress_audio_debug/", this);
debugSaveAudioCheck_->setToolTip("开启后,每次识别会将原始音频保存为 WAV 文件,用于调试音频质量问题");
sttLayout->addRow("调试录音:", debugSaveAudioCheck_);
capslockVoiceCheck_ = new QCheckBox("启用 CapsLock 长按语音输入", this);
capslockVoiceCheck_->setToolTip("长按 CapsLock 键 1 秒后触发录音,松开后自动转写并输入到光标位置");
sttLayout->addRow("快捷语音:", capslockVoiceCheck_);
beamSizeSpin_ = new QSpinBox(this);
beamSizeSpin_->setRange(1, 20);
beamSizeSpin_->setValue(5);
@ -182,8 +174,6 @@ void SettingsPage::loadFromConfig() {
sampleRateSpin_->setValue(configManager_->get("stt.sample_rate").toInt());
languageCombo_->setCurrentText(configManager_->get("stt.language").toString());
streamingCheck_->setChecked(configManager_->get("stt.streaming").toBool());
debugSaveAudioCheck_->setChecked(configManager_->get("stt.debug_save_audio").toBool());
capslockVoiceCheck_->setChecked(configManager_->get("stt.capslock_voice_enabled").toBool());
beamSizeSpin_->setValue(configManager_->get("stt.beam_size").toInt());
temperatureSpin_->setValue(configManager_->get("stt.temperature").toDouble());
@ -206,8 +196,6 @@ void SettingsPage::saveToConfig() {
configManager_->set("stt.sample_rate", sampleRateSpin_->value());
configManager_->set("stt.language", languageCombo_->currentText());
configManager_->set("stt.streaming", streamingCheck_->isChecked());
configManager_->set("stt.debug_save_audio", debugSaveAudioCheck_->isChecked());
configManager_->set("stt.capslock_voice_enabled", capslockVoiceCheck_->isChecked());
configManager_->set("stt.beam_size", beamSizeSpin_->value());
configManager_->set("stt.temperature", temperatureSpin_->value());

View File

@ -51,8 +51,6 @@ private:
QSpinBox* sampleRateSpin_;
QComboBox* languageCombo_;
QCheckBox* streamingCheck_;
QCheckBox* debugSaveAudioCheck_;
QCheckBox* capslockVoiceCheck_;
QSpinBox* beamSizeSpin_;
QDoubleSpinBox* temperatureSpin_;

View File

@ -135,10 +135,6 @@ void STTTestPage::onToggleRecording() {
return;
}
// 从配置同步调试开关到引擎
sttEngine_->setDebugSaveAudio(
configManager_->get("stt.debug_save_audio").toBool());
// 异步加载模型
if (!sttEngine_->isLoaded() ||
currentModelPath_ != modelPath) {