added non-blocking version of python demo #32
Conversation
|
This is amazing. I think the community would love this as it is extremely useful for smaller devices such as RPi zero. I have few minor comments and only one major one. All can be addressed easily :) Let read through my comments and address if necessary. Happy to discuss :) |
|
Happy to address your comments, but ... I can't find them 😅 Am I missing something? |
| input=True, | ||
| frames_per_buffer=frame_length, | ||
| input_device_index=self._input_device_index, | ||
| stream_callback=_audio_callback) |
There was a problem hiding this comment.
Could you please replace the tabs with space?
| """ | ||
|
|
||
| num_keywords = len(self._keyword_file_paths) | ||
|
|
| print('[%s] detected keyword #%d' % (str(datetime.now()), result)) | ||
| # or add it here if you use multiple keywords | ||
|
|
||
| return (None, pyaudio.paContinue) |
There was a problem hiding this comment.
We can safely remove the parentheses here :)
| result = porcupine.process(pcm) | ||
| if num_keywords == 1 and result: | ||
| print('[%s] detected keyword' % str(datetime.now())) | ||
| # add you own code execution here ... it will not block the recognition |
| sample_rate = porcupine.sample_rate | ||
| num_channels = 1 | ||
| audio_format = pyaudio.paInt16 | ||
| if not self.frame_length: |
There was a problem hiding this comment.
if checking for None explicitly maybe we can do "if self.frame_length is None:"?
| self._model_file_path = model_file_path | ||
| self._keyword_file_paths = keyword_file_paths | ||
| self._sensitivity = float(sensitivity) | ||
| self.frame_length = frame_length |
There was a problem hiding this comment.
Let's make it a protected member (self.frame_length -> self._frame_length) as it will not be used outside of class.
| sensitivities=[self._sensitivity] * num_keywords) | ||
|
|
||
| pa = pyaudio.PyAudio() | ||
| sample_rate = porcupine.sample_rate |
There was a problem hiding this comment.
Let's move this above try as you are using it in finally statement as well.
|
|
||
| num_keywords = len(self._keyword_file_paths) | ||
|
|
||
| def _audio_callback(in_data, frame_count, time_info, status): |
There was a problem hiding this comment.
This could be that I don't understand the unblocking API well ...
But the frame_count here needs to be set exactly to porcupine.frame_length otherwise the interface won't work. If the frame_count is larger then we only read the first part of it and then throw the rest. Therefore we won't be able to detect keywords. I actually tried with 4096 and I can't make it work. Could you please confirm? Also if it is smaller then we never process audio.
I think for non_blocking we need to remove the option to set frame_length and just set it to porcupine.frame_length internally in the class.
Let me know what you think :)
|
Sorry, my mistake. Forgot to click on submit. |
|
I've addressed all the comments and removed the frame_length parameter. I think you are right, it will not work anyway if it is not the same as inside porcupine. I put it in for testing, but as far as I remember changing the value resulted in no keyword detection all the time and then I just left it at the default value 🙈 I wasn't entirely sure what happens inside porcupine though and thought it might be useful at some point. Guess not ^^. |
|
Thanks a lot :) |
see:
#27
Starting from the input-buffer overflow issue I've encountered on my raspberry Pi zero I've changed the demo to run pyaudio in non-blocking mode (basically it does a callback to the execution function instead of waiting for processing to finish). It also gives you the chance to add some custom actions to the callback which always crashed the program in the blocking version.