From 801dbe1ec2a7fd5d4c0df1b38b80e08ea1be78b7 Mon Sep 17 00:00:00 2001 From: impressionyang Date: Thu, 11 Jun 2026 13:50:37 +0800 Subject: [PATCH] =?UTF-8?q?fix:=20=E9=99=8D=E4=BD=8E=20VAD=20=E8=83=BD?= =?UTF-8?q?=E9=87=8F=E9=98=88=E5=80=BC=E5=B9=B6=E6=B7=BB=E5=8A=A0=20STT=20?= =?UTF-8?q?=E6=8E=A8=E7=90=86=E8=AF=8A=E6=96=AD=E6=97=A5=E5=BF=97?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit - VAD 能量阈值从 0.015 降至 0.003，适配低增益麦克风 - transcribeChunk 添加 WAV 文件大小、样本数、RMS 诊断日志 - onChunkCompleted 增加空文件路径检查 Co-Authored-By: Claude Opus 4.7 --- src/audio/streaming_audio_writer.cpp | 6 +++--- src/ui/stt_test_page.cpp | 16 ++++++++++++++++ 2 files changed, 19 insertions(+), 3 deletions(-) diff --git a/src/audio/streaming_audio_writer.cpp b/src/audio/streaming_audio_writer.cpp index 8d389f5..01647a8 100644 --- a/src/audio/streaming_audio_writer.cpp +++ b/src/audio/streaming_audio_writer.cpp @@ -54,8 +54,8 @@ bool StreamingAudioWriter::start(int sampleRate, bool debugEnabled, const QStrin wasSpeaking_ = false; silenceFramesAfterSpeech_ = 0; - // 初始化 VAD（30ms 帧，低能量阈值） - vad_ = std::make_unique(sampleRate_, 30, 0.015f, 3); + // 初始化 VAD（30ms 帧，降低能量阈值以适配低增益麦克风） + vad_ = std::make_unique(sampleRate_, 30, 0.003f, 3); // VAD 帧大小 vadFrameSize_ = sampleRate_ * 30 / 1000; @@ -128,7 +128,7 @@ void StreamingAudioWriter::writeSamples(const std::vector& samples) { samplesWritten_ = 0; silenceFramesAfterSpeech_ = 0; wasSpeaking_ = false; - vad_ = std::make_unique(sampleRate_, 30, 0.015f, 3); + vad_ = std::make_unique(sampleRate_, 30, 0.003f, 3); if (!openNewFile()) { LOG_ERROR(kTag, "无法打开新的 WAV 文件，停止录制"); diff --git a/src/ui/stt_test_page.cpp b/src/ui/stt_test_page.cpp index 0990604..5c18dab 100644 --- a/src/ui/stt_test_page.cpp +++ b/src/ui/stt_test_page.cpp @@ -196,6 +196,12 @@ void STTTestPage::onChunkCompleted(const QString& filePath, int durationMs) { LOG_INFO(kTag, QString("WAV 片段 #%1 已完成: %2 (%3ms)") .arg(completedCount_).arg(filePath).arg(durationMs)); + if (filePath.isEmpty()) { + LOG_ERROR(kTag, "chunkCompleted 信号中文件路径为空，跳过识别"); + statusLabel_->setText("错误：文件路径为空"); + return; + } + statusLabel_->setText(QString("正在识别 #%1 (%2ms)...").arg(completedCount_).arg(durationMs)); // 在后台线程对 WAV 文件进行识别 @@ -221,6 +227,9 @@ void STTTestPage::transcribeChunk(const QString& filePath, int /* durationMs */) if (!file.open(QIODevice::ReadOnly)) { errorMsg = QString("无法打开 WAV 文件: %1").arg(filePath); } else { + // 诊断：文件大小 + qint64 fileSize = file.size(); + // 跳过 44 字节 WAV 头 file.seek(44); QByteArray raw = file.readAll(); @@ -229,13 +238,20 @@ void STTTestPage::transcribeChunk(const QString& filePath, int /* durationMs */) // int16 -> float int numSamples = raw.size() / 2; std::vector samples(numSamples); + double rms = 0.0; for (int i = 0; i < numSamples; i++) { int16_t val = *reinterpret_cast(raw.data() + i * 2); samples[i] = static_cast(val) / 32767.0f; + rms += static_cast(samples[i]) * samples[i]; } + rms = std::sqrt(rms / std::max(1, numSamples)); + + LOG_DEBUG(kTag, QString("WAV 诊断: %1 (文件大小: %2 字节, 样本: %3, RMS: %4)") + .arg(filePath).arg(fileSize).arg(numSamples).arg(rms, 0, 'f', 6)); if (!sttEngine_->isLoaded()) { text = "[错误] 模型未加载"; + LOG_ERROR(kTag, "模型未加载，无法推理"); } else { QString language = configManager_->get("stt.language").toString(); auto result = sttEngine_->infer(samples, audioSampleRate_, language);