Issues encountered

Audio not properly captured

When running the application one time, no audio was being saved, despite clear and articulate speaking into the microphone. Looking at the output for the detection node, we discovered that Google Speech was not detecting any speech whatsoever. To solve this problem, we first looked at the audio files Google Speech was listening to. The audio files were just static, with no vocals at all.

The issue was that the default audio capture channel that audio_capture was listening to was the "pulse" channel. Despite numerous attempts to change the default audio device to the USB microphone, audio capture would still listen to "pulse".

This issue was fixed by adding the roslaunch argument device:plughw:3,0 to the command roslaunch audio_capture capture.launch, so that the whole command was then

              `roslaunch audio_capture capture.launch device:=plughw:3,0`

The parameter plughw:3,0 was obtained by running the command arecord -l and finding the information about the microphone being used. For us, the information looked like this:

card 3: CameraB409241 [USB Camera-B4.09.24.1], device 0: USB Audio [USB Audio] Subdevices: 1/1 Subdevice #0: subdevice #0

The numbers 3,0 were obtained by looking at the card number on the first line, and the Subdevice number on the third line.

If the microphone is built into the computer, you might need to replace plughw with just hw.