Custom Wake Word

When I try to record a custom wake word it says there is too much noise (even though it is completely silent), I get a bunch of errors, and it goes back to prompt asking to hit enter to record a wake word. It seems to just continue to loop thorough this. I have recorded at least 2 dozen times but there is only 1 wave file in recording folder.

Do record your custom wakework on a good mic, on a good computer. Use audacity to trim it the neareast possible to the sound start and stop and save them as 0/1/2.wav 12khz 16bit.

This is the config file, save as config.json

{
    "num_mel_bins": 13, 
    "window_type": "povey", 
    "kind": "personal", 
    "window_size": 10, 
    "cepstral_lifter": 22.0, 
    "use_energy": false, 
    "energy_floor": 0.0, 
    "from_mfcc": 1, 
    "dtw_ref": 0.22, 
    "raw_energy": true, 
    "frame_length_ms": 25.0, 
    "frame_shift_ms": 10.0, 
    "preemphasis_coefficient": 0.97, 
    "to_mfcc": 13, 
    "sample_rate": 16000, 
    "dither": 0.0, 
    "mel_low_freq": 20, 
    "shift": 10, 
    "num_mfcc": 13, 
    "band_radius": 10, 
    "hotword_key": "HOTWORD_NAME"
}

The config says the sample rate is 16kHz, but you say 12kHz. Is that right?

My bad, typed too fast, sry…

16000mhz, 16bits signed wav

1 Like

I have the same problem. and I don’t have any json file.and when I run sudo systemctl stop snips-audio-server; python script_recording.py buttercup is never end.according to site it must be 3 times but for me it doesn’t have any limitation.and at the end after 20 times test it will gives me wav files. I already uninstall python-numpy python-pyaudio python-soundfile and reinstall it.