impressionyang/impress_voice_input
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 2 Wiki Activity

fix: 降低 VAD 能量阈值并添加 STT 推理诊断日志

Browse Source 
- 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 <noreply@anthropic.com>
This commit is contained in:
Alvin Young 2026-06-11 13:50:37 +08:00
parent 271fccb39b
commit 801dbe1ec2
2 changed files with 19 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
src/audio/streaming_audio_writer.cpp
View File

@ -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<VoiceActivityDetector>(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<VoiceActivityDetector>(sampleRate_, 30, 0.003f, 3);
// VAD 帧大小
vadFrameSize_ = sampleRate_ * 30 / 1000;
@ -128,7 +128,7 @@ void StreamingAudioWriter::writeSamples(const std::vector<float>& samples) {
samplesWritten_ = 0;
silenceFramesAfterSpeech_ = 0;
wasSpeaking_ = false;
vad_ = std::make_unique<VoiceActivityDetector>(sampleRate_, 30, 0.015f, 3);
vad_ = std::make_unique<VoiceActivityDetector>(sampleRate_, 30, 0.003f, 3);
if (!openNewFile()) {
LOG_ERROR(kTag, "无法打开新的 WAV 文件，停止录制");

16
src/ui/stt_test_page.cpp
View File

@ -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<float> samples(numSamples);
double rms = 0.0;
for (int i = 0; i < numSamples; i++) {
int16_t val = *reinterpret_cast<const int16_t*>(raw.data() + i * 2);
samples[i] = static_cast<float>(val) / 32767.0f;
rms += static_cast<double>(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);