Tried it on a Pi Zero2 and works well with minimal load.
I couldn’t get the example to run and swapped to a callback with sounddevice code hack here.
# Copyright 2022 David Scripka. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# Imports
import sounddevice as sd
import numpy as np
from openwakeword.model import Model
import argparse
import threading
# Parse input arguments
parser=argparse.ArgumentParser()
parser.add_argument(
"--chunk_size",
help="How much audio (in number of samples) to predict on at once",
type=int,
default=1280,
required=False
)
parser.add_argument(
"--model_path",
help="The path of a specific model to load",
type=str,
default="",
required=False
)
parser.add_argument(
"--inference_framework",
help="The inference framework to use (either 'onnx' or 'tflite'",
type=str,
default='tflite',
required=False
)
args=parser.parse_args()
CHANNELS = 1
RATE = 16000
CHUNK = args.chunk_size
sd.default.samplerate = RATE
sd.default.channels = CHANNELS
sd.default.dtype= ('int16', 'int16')
def sd_callback(rec, frames, time, status):
audio = np.frombuffer(rec, dtype=np.int16)
prediction = owwModel.predict(audio)
# Column titles
n_spaces = 16
output_string_header = """
Model Name | Score | Wakeword Status
--------------------------------------
"""
for mdl in owwModel.prediction_buffer.keys():
# Add scores in formatted table
scores = list(owwModel.prediction_buffer[mdl])
curr_score = format(scores[-1], '.20f').replace("-", "")
output_string_header += f"""{mdl}{" "*(n_spaces - len(mdl))} | {curr_score[0:5]} | {"--"+" "*20 if scores[-1] <= 0.5 else "Wakeword Detected!"}
"""
# Print results table
print("\033[F"*(4*n_models+1))
print(output_string_header, " ", end='\r')
# Load pre-trained openwakeword models
if args.model_path != "":
owwModel = Model(wakeword_models=[args.model_path], inference_framework=args.inference_framework)
else:
owwModel = Model(inference_framework=args.inference_framework)
n_models = len(owwModel.models.keys())
# Start streaming from microphone
with sd.InputStream(channels=CHANNELS,
samplerate=RATE,
blocksize=int(CHUNK),
callback=sd_callback):
threading.Event().wait()
Just create a /etc/asound.conf
pcm.!default {
type asym
playback.pcm "plughw:0"
capture.pcm "plughw:0"
}
As that will create a plug device which basically autoconverts to the format you want.
hw:0 is the card and really always create a plughw:device than direct hw:device as it makes things so much easier.
The onnx model also runs but with a bit more load so the default is likely the way to go.
