micro_speech — Micro Speech Audio Module Example¶
micro_speech module runs Google’s TensorFlow Lite for Microcontrollers Micro Speech framework
for voice recognition.
Please see this guide for training a new model.
- class micro_speech.MicroSpeech¶
Creates a MicroSpeech voice recognition class.
Pass this method to
audio.start_streaming()to fill the
MicroSpeechclass with audio samples.
MicroSpeechwill compute the FFT of the audio samples and keep a sliding window internally of the FFT the last 100ms or so of audio samples received as features for voice recognition.
- listen(tf_model[, threshold=0.9[, timeout=1000[, filter=None]]])¶
Executes the tensor flow lite model
tf_model, which should be a path to a tensor flow lite model on disk, on the audio stream.
This method will continue to execute the model until it classifies a result that has a confidence ratio above
thresholdand that’s within the range specified by
For example, if the model is designed to classify sounds into the four labels [‘Silence’, ‘Unknown’, ‘Yes’, ‘No’], then a
thresholdof 0.7 mean that listen() only returns when the confidence score for one of those classes goes above 0.7.
filtercan then be
[2, 3]to specify that we only care about ‘Yes’ or ‘No’ going above 0.7.
timeoutis the amount of time to run the model on audio data. If zero then listen will run forever until a result passes the threshold and filter criteria.
Returns the index of the the label with the highest confidence score. E.g. for the example above 0, 1, 2, or 3 for [‘Silence’, ‘Unknown’, ‘Yes’, ‘No’] respectively.