A model trained to detect the presence of the phrase "hey mycroft" in an audio recording of speech.
Other similar phrases such as just "mycroft" or may also work, but likely with higher false-reject rates. Similarly, a short pause after the speaking the wakeword is recommended, but the model may also detect the presence of the wakeword is a continuous stream of speech in certain cases.
The model is a simple 3-layer full-connected network, that takes the flattened input features from the frozen audio embedding mode. ReLU activations and layer norms are inserted between the layers. A representative (but not exact) example of this structure is shown below.
==========================================================================================
Layer (type:depth-idx) Output Shape Param #
==========================================================================================
Sequential [1, 1] --
├─Flatten: 1-1 [1, 1536] --
├─Linear: 1-2 [1, 64] 98,368
├─LayerNorm: 1-3 [1, 64] 128
├─ReLU: 1-4 [1, 64] --
├─Linear: 1-5 [1, 64] 4,160
├─LayerNorm: 1-6 [1, 64] 128
├─ReLU: 1-7 [1, 64] --
├─Linear: 1-8 [1, 1] 65
├─Sigmoid: 1-9 [1, 1] --
==========================================================================================
Total params: 102,849
Trainable params: 102,849
Non-trainable params: 0
Total mult-adds (M): 0.10
==========================================================================================
Input size (MB): 0.01
Forward/backward pass size (MB): 0.00
Params size (MB): 0.41
Estimated Total Size (MB): 0.42
==========================================================================================
The model was trained on approximately ~100,000 synthetically generated clips of the "hey mycroft" wake phrase using two text-to-speech (TTS) models:
- NVIDIA WAVEGLOW with the LibriTTS multi-speaker model
- VITS with the VCTK multi-speaker model
Clips were generated both with the trained speaker embeddings, and also mixtures of individual speaker embeddings to produce novel voices. See the Synthetic Data Generation documentation page for more details.
The following phrases were included in the training data:
- "hey mycroft"
- "hey mycroft
<random words>
"
After generating the synthetic positive wakewords, they are augmented in two ways:
- Mixing with clips from the ACAV100M dataset referenced below at ratios of 0 to 30 dB
- Reverberated with simulated room impulse response functions from the BIRD Impulse Response Dataset
The model was trained on approximately ~31,000 hours of negative data, with the approximate composition shown below:
- ~10,000 hours of noise, music, and speech from the ACAV100M dataset
- ~10,000 hours from the Common Voice 11 dataset, representing multiple languages
- ~10,000 hours of podcasts downloaded from the Podcastindex database
- ~1,000 hours of music from the Free Music Archive dataset
In addition to the above, the total negative dataset also includes reverberated versions of the ACAV100M dataset (also using the simulated room impulse responses from the BIRD Impulse Response Dataset dataset). Currently, adversarial STT generations were not added to the training data for this model.
The positive test examples of the "hey mycroft" wakeword were collected manually in a realistic home environment from both near and far-field microphones, at distances ranging from ~3 to ~30 feet. The (male) speaker has a relatively neutral American english accent, and the recordings were captured with normal background noise included fans/air conditioning and a running dishwasher in a kitchen. A total of 51 clips were recorded in this manner.
The false-accept/false-reject curve for the model on the test data is shown below. Decreasing the threshold
parameter when using the model will increase the false-accept rate and decrease the false-reject rate.
While the model was trained to be robust to background noise and reverberation, it will still perform the best when the audio is relatively clean and free of overly loud background noise. In particular, the presence of audio playback of music/speech from the same device that is capturing the microphone stream may result in significantly higher false-reject rates unless acoustic echo cancellation (AEC) is performed via hardware or software.