Files
live-caption/apps/gui/main.cpp
T
blumia 6c04071bb5 feat(gui): expose the streaming and debug options in the advanced page
These were command-line only. Add them to the settings dialog so the whole
feature set is reachable without arguments: stream mode, window limit,
segment limit, prefix-rollback token count, cold-start chunk count, max
tokens per step, thread count, verbose logging and the model directory. The
streaming group is grouped and greyed out for streaming-ONNX models, which
only use the commit granularity. The backend fingerprint now includes them,
so changing one restarts the recogniser; appearance changes still apply live.

Also add a test-only --settings-tab so a specific page can be screenshotted.

设置「高级」页补上原来只有命令行才有的选项:流式策略/滑窗上限/单段最长/前缀回滚
token 数/冷启动 chunk 数/单段最多生成 token/线程数/详细日志/模型目录。流式那一组
单独成组,选了流式 ONNX 模型时置灰(它们只用「更新粒度」)。这些值纳入「后端指纹」,
改动会重启识别后端;外观类改动仍然即时生效。另加测试用 --settings-tab 以便给单个
页签截图。
2026-10-12 17:36:43 +08:00

377 lines
16 KiB
C++
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
// 图形界面版:悬浮字幕窗 + 托盘 + 设置对话框。
// 完全通过 GUI 使用;命令行参数保留,仅用于自动化测试(显式给出时覆盖设置)。
#include <cstdio>
#include <memory>
#include <QAction>
#include <QApplication>
#include <QCommandLineOption>
#include <QCommandLineParser>
#include <QIcon>
#include <QMenu>
#include <QMessageBox>
#include <QPainter>
#include <QPixmap>
#include <QPushButton>
#include <QStringList>
#include <QSystemTrayIcon>
#include <QTabWidget>
#include <QTimer>
#include "appcli.h"
#include "appsettings.h"
#include "captionwindow.h"
#include "iasr.h"
#include "livetranscriber.h"
#include "modelstore.h"
#include "qwen3asr.h"
#include "settingsdialog.h"
#include "streamingasr.h"
#include "transcript.h"
namespace {
// 托盘图标:直接画一个,免得依赖图标资源文件
QIcon makeTrayIcon()
{
QPixmap pm(64, 64);
pm.fill(Qt::transparent);
QPainter p(&pm);
p.setRenderHint(QPainter::Antialiasing);
p.setBrush(QColor(20, 20, 20, 235));
p.setPen(Qt::NoPen);
p.drawRoundedRect(2, 12, 60, 40, 10, 10);
p.setPen(QColor(240, 240, 240));
QFont f = p.font();
f.setPixelSize(26);
f.setBold(true);
p.setFont(f);
p.drawText(pm.rect(), Qt::AlignCenter, QStringLiteral("字"));
return QIcon(pm);
}
// 把设置应用到字幕窗(字号/不透明度/行数/位置)
void applyAppearance(lc::CaptionWindow & window, const lc::AppSettings & cfg)
{
window.setPanelOpacity(cfg.opacityPercent);
window.setTextFontSize(cfg.fontSize);
window.setLineCounts(cfg.collapsedLines, cfg.expandedLines);
if (cfg.hasWindowPos) window.move(cfg.windowPos);
}
bool hasUsableModel()
{
using namespace lc::models;
const lc::AppSettings & cfg = lc::AppSettings::instance();
if (cfg.modelId.isEmpty()) return false;
const ModelInfo * m = ModelStore::find(cfg.modelId);
return m && ModelStore::isInstalled(*m);
}
// 后端相关设置的「指纹」:这些值变了就必须重建后端(外观类改动不需要)。
QString backendKeyOf(const lc::AppSettings & c)
{
return QStringList{ c.modelId,
c.language,
c.captureDevice,
c.modelDir,
c.streamMode,
QString::number(c.chunkSec),
QString::number(c.streamWindowSec),
QString::number(c.streamSegmentSec),
QString::number(c.unfixedTokens),
QString::number(c.unfixedChunks),
QString::number(c.maxTokens),
QString::number(c.threads),
QString::number(c.verbose ? 1 : 0) }
.join(QLatin1Char('|'));
}
} // namespace
int main(int argc, char ** argv)
{
QGuiApplication::setHighDpiScaleFactorRoundingPolicy(
Qt::HighDpiScaleFactorRoundingPolicy::PassThrough);
QApplication app(argc, argv);
// 关键:本程序常驻托盘,"最后一个窗口关闭"不等于退出。
// 否则首次启动时设置对话框一关(那时它是唯一窗口),Qt 就会把主事件循环也结束掉,
// 表现为"点了确定程序就自己退出"。
QApplication::setQuitOnLastWindowClosed(false);
QCoreApplication::setApplicationName(QStringLiteral("live-caption"));
QCoreApplication::setOrganizationName(QStringLiteral("live-caption"));
lc::AppSettings & cfg = lc::AppSettings::instance();
cfg.load();
// ---- 命令行(仅用于自动化测试,显式给出时覆盖设置)----
QCommandLineParser parser;
parser.setApplicationDescription(QStringLiteral("实时字幕(悬浮窗)"));
parser.addHelpOption();
lc::app::addModelOptions(parser);
lc::app::addStreamOptions(parser);
lc::app::addBackendOptions(parser);
parser.addOption(QCommandLineOption("opacity", "背景不透明度(20-100)", "n"));
parser.addOption(QCommandLineOption("screenshot", "启动后自动截图并退出", "path"));
parser.addOption(QCommandLineOption("screenshot-delay", "截图前等待秒数", "sec", "6"));
parser.addOption(QCommandLineOption("no-tray", "不创建托盘图标(测试用)"));
parser.addOption(QCommandLineOption("screenshot-settings", "打开设置对话框截图后退出", "path"));
parser.addOption(QCommandLineOption("settings-tab", "截图设置对话框时先切到某个页签(0-2 或 模型/显示/高级)", "tab"));
parser.addOption(QCommandLineOption("settings-autoclose", "复现用:自动操作设置对话框(reject|close|accept|ok|apply)", "how"));
parser.addOption(QCommandLineOption("quit-after", "运行若干秒后优雅退出(测试退出路径用)", "sec"));
parser.process(app);
if (parser.isSet("asr-model")) cfg.modelId = parser.value("asr-model");
if (parser.isSet("language")) cfg.language = parser.value("language");
if (parser.isSet("opacity")) cfg.opacityPercent = parser.value("opacity").toInt();
if (parser.isSet("model-dir")) cfg.modelDir = parser.value("model-dir");
if (!cfg.modelDir.isEmpty()) lc::models::ModelStore::setRootOverride(cfg.modelDir);
// 仅用于验证渲染:打开设置对话框截图后退出
if (parser.isSet("screenshot-settings")) {
const QString path = parser.value("screenshot-settings");
lc::SettingsDialog dlg(!hasUsableModel());
// 可选:先切页签,便于给某个页单独截图
const QString tab = parser.value("settings-tab");
if (!tab.isEmpty()) {
if (auto * tw = dlg.findChild<QTabWidget *>()) {
bool numeric = false;
const int n = tab.toInt(&numeric);
if (numeric) {
tw->setCurrentIndex(qBound(0, n, tw->count() - 1));
} else {
for (int i = 0; i < tw->count(); ++i) {
if (tw->tabText(i) == tab) { tw->setCurrentIndex(i); break; }
}
}
}
}
dlg.show();
QTimer::singleShot(1200, &dlg, [&dlg, path] {
dlg.grab().save(path);
std::fprintf(stderr, "设置对话框截图: %s\n", qUtf8Printable(path));
dlg.accept();
});
dlg.exec();
return 0;
}
// 复现用:把"点 ✕ / 点确定"自动化,用来定位崩溃
const QString autoclose = parser.value("settings-autoclose");
const auto scheduleAutoclose = [&autoclose](QDialog & dlg) {
if (autoclose.isEmpty()) return;
QTimer::singleShot(700, &dlg, [&dlg, &autoclose] {
if (autoclose == QLatin1String("reject")) dlg.reject();
else if (autoclose == QLatin1String("close")) dlg.close(); // 标题栏 ✕
else if (autoclose == QLatin1String("accept")) dlg.accept();
else if (autoclose == QLatin1String("ok")) { // 真的点「确定」
for (auto * b : dlg.findChildren<QPushButton *>()) {
if (b->text() == QObject::tr("确定")) { b->click(); break; }
}
}
else if (autoclose == QLatin1String("apply")) { // 点「应用」,应保持打开
for (auto * b : dlg.findChildren<QPushButton *>()) {
if (b->text() == QObject::tr("应用")) { b->click(); break; }
}
QTimer::singleShot(400, &dlg, [&dlg] {
std::fprintf(stderr, "点「应用」后对话框仍可见: %s\n",
dlg.isVisible() ? "是 ✅" : "否 ❌");
dlg.reject();
});
}
});
};
// ---- 启动:没有可用模型就先开设置 ----
if (!hasUsableModel()) {
lc::SettingsDialog dlg(true);
scheduleAutoclose(dlg);
if (dlg.exec() != QDialog::Accepted || !hasUsableModel()) return 0;
}
lc::CaptionWindow window;
applyAppearance(window, cfg);
// 语言 pill:流式模型不提供语言识别,就显示设置里选的语言
if (!cfg.language.isEmpty()) {
window.setLanguage(QString::fromStdString(lc::languageDisplayName(cfg.language.toStdString())));
}
// 注意声明顺序:asr 必须在 live 之前声明,
// 这样析构时 live(会 join 工作线程)先销毁、asr 后销毁,避免 use-after-free。
std::unique_ptr<lc::IAsr> asr;
lc::LiveTranscriber live; // 后端在 startTranscriber() 里注入
const auto startTranscriber = [&]() -> bool {
live.stop();
asr.reset();
live.setCaptureDevice(cfg.captureDevice);
const lc::models::ModelInfo * m = lc::models::ModelStore::find(cfg.modelId);
if (!m) {
qCritical("找不到模型 %s", qUtf8Printable(cfg.modelId));
return false;
}
std::string err;
asr = lc::app::makeAsrForModel(*m, cfg.language, cfg.threads, cfg.maxTokens,
cfg.verbose || parser.isSet("verbose"), &err);
if (!asr) {
qCritical("加载模型失败: %s", err.c_str());
return false;
}
// 流式参数:只有 llama 后端(Qwen3-ASR)用得到大部分参数,
// 流式 ONNX 只关心「更新粒度」。
lc::StreamOptions sop;
sop.windowSec = cfg.streamWindowSec;
sop.maxUtteranceSec = cfg.streamSegmentSec;
sop.unfixedTokenNum = cfg.unfixedTokens;
sop.unfixedChunkNum = cfg.unfixedChunks;
sop.maxTokensPerStep = cfg.maxTokens;
sop.incremental = (cfg.streamMode != QLatin1String("append")
&& cfg.streamMode != QLatin1String("window"));
sop.appendOnly = (cfg.streamMode == QLatin1String("append"));
if (asr->kind() == lc::AsrKind::NativeStreaming) {
sop.chunkSec = (cfg.chunkSec > 0.0) ? cfg.chunkSec : 0.32;
} else if (cfg.chunkSec > 0.0) {
sop.chunkSec = cfg.chunkSec;
}
QString qerr;
if (!live.start(*asr, sop, &qerr)) {
qCritical("启动采集失败: %s", qUtf8Printable(qerr));
return false;
}
return true;
};
QObject::connect(&live, &lc::LiveTranscriber::caption, &window,
[&window](const QString & committed, const QString & tail,
const QString & lang) {
if (!lang.isEmpty()) {
window.setLanguage(
QString::fromStdString(lc::languageDisplayName(lang.toStdString())));
}
window.setCaption(committed, tail);
});
QObject::connect(&live, &lc::LiveTranscriber::failed, &window, [](const QString & m) {
qWarning("采集失败: %s", qUtf8Printable(m));
});
// 后端相关项(模型/语言/采集设备/流式参数)变化才需要重启;外观类改动即时生效
QString activeKey = backendKeyOf(cfg);
const auto openSettings = [&] {
lc::SettingsDialog dlg(false);
QObject::connect(&dlg, &lc::SettingsDialog::settingsApplied, &window, [&] {
lc::models::ModelStore::setRootOverride(cfg.modelDir);
applyAppearance(window, cfg); // 透明度/字号等立刻可见
if (backendKeyOf(cfg) != activeKey) {
if (startTranscriber()) activeKey = backendKeyOf(cfg);
}
});
if (dlg.exec() == QDialog::Accepted) {
lc::models::ModelStore::setRootOverride(cfg.modelDir);
applyAppearance(window, cfg);
if (backendKeyOf(cfg) != activeKey) {
if (!startTranscriber()) {
QMessageBox::warning(&window, QObject::tr("启动失败"),
QObject::tr("模型加载失败,请检查设置。"));
} else {
activeKey = backendKeyOf(cfg);
}
}
}
};
// 语言 pill → 设置
QObject::connect(&window, &lc::CaptionWindow::languageClicked, &window, openSettings);
// ---- 托盘 ----
QSystemTrayIcon * tray = nullptr;
QMenu * trayMenu = nullptr;
if (!parser.isSet("no-tray")) {
tray = new QSystemTrayIcon(makeTrayIcon(), &app);
tray->setToolTip(QObject::tr("实时字幕"));
trayMenu = new QMenu(&window); // 必须有归属,否则退出时是悬空顶层窗口
auto * showAct = trayMenu->addAction(QObject::tr("显示/隐藏字幕"));
auto * setAct = trayMenu->addAction(QObject::tr("设置…"));
trayMenu->addSeparator();
auto * quitAct = trayMenu->addAction(QObject::tr("退出"));
tray->setContextMenu(trayMenu);
QObject::connect(showAct, &QAction::triggered, &window,
[&window] { window.setVisible(!window.isVisible()); });
QObject::connect(setAct, &QAction::triggered, &window, openSettings);
QObject::connect(quitAct, &QAction::triggered, &app, &QApplication::quit);
QObject::connect(tray, &QSystemTrayIcon::activated, &window,
[&window](QSystemTrayIcon::ActivationReason r) {
if (r == QSystemTrayIcon::Trigger) {
window.setVisible(!window.isVisible());
}
});
tray->show();
}
// 关闭按钮 = 隐藏到托盘(否则关掉就没法再打开了)
QObject::connect(&window, &lc::CaptionWindow::closed, &window, [&] {
if (tray) {
window.hide();
} else {
qApp->quit();
}
});
// 记住拖动后的位置
QObject::connect(&window, &lc::CaptionWindow::moved, &window, [&](const QPoint & p) {
cfg.windowPos = p;
cfg.hasWindowPos = true;
cfg.save();
});
if (!startTranscriber()) {
QMessageBox::critical(nullptr, QObject::tr("启动失败"),
QObject::tr("无法加载模型,请检查设置。"));
return 1;
}
window.show();
// 复现用:程序起来后再开一次设置并自动关闭(模拟从语言标签/托盘进设置)
if (!autoclose.isEmpty() && hasUsableModel()) {
QTimer::singleShot(2500, &window, [&] {
lc::SettingsDialog dlg(false);
scheduleAutoclose(dlg);
if (dlg.exec() == QDialog::Accepted) {
applyAppearance(window, cfg);
if (!startTranscriber()) qWarning("重启转写失败");
}
QTimer::singleShot(500, qApp, &QCoreApplication::quit);
});
}
// 自截图(验证渲染用)
if (parser.isSet("screenshot")) {
const QString path = parser.value("screenshot");
QTimer::singleShot(parser.value("screenshot-delay").toInt() * 1000, &window,
[&window, path] {
const bool ok1 = window.grab().save(path);
window.setCollapsed(false);
QString expanded = path;
expanded.replace(QStringLiteral(".png"), QStringLiteral("-expanded.png"));
const bool ok2 = window.grab().save(expanded);
std::fprintf(stderr, "截图: 折叠态 %s / 展开态 %s\n",
ok1 ? "成功" : "失败", ok2 ? "成功" : "失败");
QCoreApplication::quit();
});
}
if (parser.isSet("quit-after")) {
QTimer::singleShot(parser.value("quit-after").toInt() * 1000, qApp,
&QCoreApplication::quit);
}
const int rc = QCoreApplication::exec();
// 显式收尾,确保工作线程停下来之后再释放后端
live.stop();
asr.reset();
return rc;
}