Well there are some news. The returned minimum buffer size is always constant 640 Bytes! One of PJSIPs endpoint is the android jni device (according to https://trac.pjsip.org/repos/wiki/media-flow ). This device holds a record stream. To open the stream it's necessary to get the minimum buffer size from androids jni layer. This is done by ... inputBuffSizeRec = (*jni_env)->CallStaticIntMethod(jni_env, stream->record_class, bufsize_method, param->clock_rate, channelInCfg, sampleFormat); ... BUT android returns always 640 bytes. This will be a big problem for the functionality of the jni dev since the AndroidRecorderCallback expects a exactly amount of read bytes. Those amount of bytes must match the requested amount of bytes. The requested amount of bytes will be calculated in the android_create_stream(...) function by the following code sniped: ... buffSize = stream->param.samples_per_frame*stream->param.bits_per_sample/8; stream->rec_buf_size = stream->play_buf_size = buffSize; ... In my case this size is 1764 bytes. The read back amount of bytes will be 640, because the stream->record is initialized with a 640 bytes buffer. So if the AndroidRecorderCallback will be called it never copies any bytes! ====================================================================== while (!stream->quit_flag) { pjmedia_frame frame; pj_status_t status; int bytesRead; if (!stream->running) { (*jni_env)->CallVoidMethod(jni_env, stream->record, stop_method); pj_sem_wait(stream->rec_sem); if (stream->quit_flag) break; (*jni_env)->CallVoidMethod(jni_env, stream->record, record_method); } bytesRead = (*jni_env)->CallIntMethod(jni_env, stream->record, read_method, inputBuffer, 0, size); if (bytesRead <= 0 || bytesRead != size) { <--------------========== There is the problem PJ_LOG (4, (THIS_FILE, "Record thread : error %d reading data Bytes read: (%d), size(rec_buf_size): (%d)",bytesRead, size)); continue; } buf = (*jni_env)->GetByteArrayElements(jni_env, inputBuffer, 0); frame.type = PJMEDIA_FRAME_TYPE_AUDIO; frame.size = size; frame.bit_info = 0; frame.buf = (void *)buf; frame.timestamp.u64 = stream->rec_timestamp.u64; status = (*stream->rec_cb)(stream->user_data, &frame); (*jni_env)->ReleaseByteArrayElements(jni_env, inputBuffer, buf, JNI_ABORT); stream->rec_timestamp.u64 += stream->param.samples_per_frame / stream->param.channel_count; } ====================================================================== Now I figured out that the audio hal will statically return an integer value of 320. This value will be doubled by the framework and leads to the 640 Bytes. What can I do to fix this Problem? Ralf Zeuka Callom GmbH Neue Stra?e 97-99 89073 Ulm Tel.: +49 731 140015 0 Fax: +49 731 140015 15 eMail:? ralf.zeuka at callom.com www.callom.com Gesch?ftsf?hrer: Marcus Brachert, Timo Weith?ner - Registergericht Ulm HRB 72968 > -----Urspr?ngliche Nachricht----- > Von: pjsip [mailto:pjsip-bounces at lists.pjsip.org] Im Auftrag von Ralf Zeuka | > Callom GmbH > Gesendet: Freitag, 25. September 2015 17:25 > An: pjsip list <pjsip at lists.pjsip.org> > Betreff: [pjsip] Sound shutters - buffer mismatch in android jni dev - armv7 > cortex m5 > > Hi there, > > I'am using PJSIP on an arm-cortex M5 architecture. (C1 from Hardkernel) But > I've some trouble with the sound. > > System: Android 4.4.2 on ARMv7 Cortex-M5 > PJSIP: Version 2.3 > > I'd like to use the android_jni_dev and the libpjsua.so . > When I start an outgoing call I get frequently errors via logcat out from the > android_jni_dev.c . > > "D/PJSUA Native( 1559): 17:55:02.479 android_jni_de Record thread : error > 640 reading data" > > According to this error I expanded the corresponding log message. > > ======================================================= > ... > bytesRead = (*jni_env)->CallIntMethod(jni_env, stream->record, > read_method, inputBuffer, > 0, size); > > if (bytesRead <= 0 || bytesRead != size) { > PJ_LOG (4, (THIS_FILE, "Record thread : error %d reading > data Bytes read: (%d), size(rec_buf_size): (%d)", > bytesRead, bytesRead, (int) size)); > continue; > } > ... > ======================================================= > Code: android_jni_dev::AndroidRecorderCallback(...) > > So I could see a buffer mismatch: > > D/PJSUA Native( 1559): 17:55:02.525 android_jni_de Record thread : > error 640 reading data Bytes read: (640), size(rec_buf_size): (1764) > > A fast but bad fix for this would be to set the 'inputBuffSizeRec' to 1764 > within the following piece of code: > ============================================== > ... > stream->record = (*jni_env)->NewObject(jni_env, > stream->record_class, > constructor_method, > mic_source, > param->clock_rate, > channelInCfg, > sampleFormat, > inputBuffSizeRec); // ==> 1764 > ... > ============================================== > Code: android_jni_dev::android_create_stream(...) > > The right size should be calculated by: > ========================================================== > ===================== > ... > inputBuffSizeRec = (*jni_env)->CallStaticIntMethod(jni_env, > stream->record_class, > bufsize_method, > param->clock_rate, > channelInCfg, > sampleFormat); > > ... > ========================================================== > ===================== > Code: android_jni_dev::android_create_stream(...) > > The 'size' var is also calculated near this code by the following: > ========================================================== > ======== > ... > buffSize = stream->param.samples_per_frame*stream- > >param.bits_per_sample/8; > stream->rec_buf_size = stream->play_buf_size = buffSize; ... > ========================================================== > ======== > Code: android_jni_dev::android_create_stream(...) > > > After setting the 'inputBuffSizeRec' to the fix value 1764 the error message > doesn't appear anymore. > > But now the sound shutters' from time to time. > AudioFlinger has some overflows and PJSUA has some underflows. > Furthermore messages with 'buffer adjustments' will be printed out. > > I measured the output signal with an oscilloscope and printed the pcm > stream within the 'conference.c' to a file. > > Both show me the same: > > During the shutter phase the same signal will be repeated. So I think if this > happens, pjsua will send the same frame again and again. > > > I'am wondering why the buffer size will not be calculated correctly? > Now I need some help to figure out what's going wrong. > > Regards, > > > Ralf Zeuka > > Callom GmbH > Neue Stra?e 97-99 > 89073 Ulm > > Tel.: +49 731 140015 0 > Fax: +49 731 140015 15 > > eMail:? ralf.zeuka at callom.com > www.callom.com > Gesch?ftsf?hrer: > Marcus Brachert, Timo Weith?ner - Registergericht Ulm HRB 72968 > > _______________________________________________ > Visit our blog: http://blog.pjsip.org > > pjsip mailing list > pjsip at lists.pjsip.org > http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
