lifeiteng / codingmath Goto Github PK

Do you know how to deal with 44.1k samplerate?

there's no check for apm->ProcessReverseStream(echo_frame); actually,it will return
kBadSampleRateError = -7.
just treat the echo_frame like frame,initialize it and the prog will run happily. the aec module will work.
webrtc::AudioFrame *echo_frame = NULL;
if (is_echo_cancel) {
echo_frame = new webrtc::AudioFrame();
echo_frame->num_channels_ = 1;
echo_frame->sample_rate_hz_ = 16000;
echo_frame->samples_per_channel_ = frame->sample_rate_hz_ * frame_step / 1000.0;
}

你好，我在数字编程上看到你这篇博文《构建 CTC 语音识别解码网络》，在构建好解码图后，如何得到更加准确的输出呢？最后的利用解码图（fst)解码的具体过程是什么样的？
“解码时从解码网络获得TransitionId转为对应的PdfId做为 RNN 输出层索引获得” ，我最想理解你这句话，发现看不懂
期待你的回复，谢谢