SpeechConfiguration

A comma-separated list of wakeword keywords

The name of the window function to apply to each audio frame before calculating the STFT.

The desired linear Root Mean Squared (RMS) signal energy, which is used for signal normalization and should be tuned to the RMS target used during wakeword model training.

The Exponentially Weighted Moving Average (EWMA) update rate for the current Root Mean Squared (RMS) signal energy (0 for no RMS normalization).

The size of the signal window used to calculate the STFT, in number of samples - should be a power of 2 for maximum efficiency.

The length of time to skip each time the overlapping STFT is calculated, in milliseconds.

The length of a frame in the mel spectrogram used as an input to the wakeword recognizer encoder, in milliseconds.

The number of filterbank components in each mel spectrogram frame sent to the wakeword recognizer.

The size of the wakeword recognizer’s encoder state output.

The size of the wakeword recognizer’s encoder window output.

The length of the sliding window of encoder output used as an input to the wakeword recognizer classifier, in milliseconds.

The threshold of the wakeword recognizer classifier’s posterior output, above which the wakeword recognizer activates the pipeline, in the range [0, 1].

The minimum length of an activation, in milliseconds. Used to ignore a Voice Activity Detector (VAD) deactivation after the wakeword.

The maximum length of an activation, in milliseconds. Used to time out the speech pipeline activation.

Indicate to the VAD the level of permissiveness to non-speech activation.

Delay between a VAD deactivation and the delivery of the recognition results.

Audio sampling rate, in Hz.

Audio frame width, in milliseconds.

Length of time to allow an Apple ASR request to run, in milliseconds.

The pre-emphasis filter weight to apply to the normalized audio signal, in a range of [0, 1].

The filename of the machine learning model used for the filtering step.

The filename of the machine learning model used for the encoding step.

The filename of the machine learning model used for the detect step.

The filesystem path to the machine learning model for the filtering step.

The filesystem path to the machine learning model for the encoding step.

The filesystem path to the machine learning model for the detect step.

Text To Speech API client identifier key.

Text To Speech API client secret key.

The filesystem path to the vocabulary used for tokenizer encoding.

The index in the vocabulary of the terminator token. Determined by the NLU vocabulary.

The index in the vocabulary of the terminator token. Determined by the NLU vocabulary.

The filesystem path to the machine learning model for Natural Language Understanding processing.

The filesystem path to the model metadata for Natural Language Understanding processing.

The maximum utterance length the NLU can process. Determined by the NLU model.

Debugging trace levels, for simple filtering.

Delegate events will be sent using the specified dispatch queue.

Automatically run Spokestack’s NLU classification on ASR transcripts for clients that use the Spokestack facade.

The filename of the machine learning model used for the filtering step of the keyword recognizer.

The filename of the machine learning model used for the encoding step of the keyword recognizer.

The filename of the machine learning model used for the detect step of the keyword recognizer.

The filename of the model metadata for keyword recognition

The filesystem path to the machine learning model for the filtering step of the keyword recognizer.

The filesystem path to the machine learning model for the encoding step of the keyword recognizer.

The filesystem path to the machine learning model for the detect step of the keyword recognizer.

The threshold of the keyword recognizer’s posterior output, above which the keyword recognizer emits a recognition event for the most probable keyword.

The filesystem path to the model metadata for keyword recognition

A comma-separated list of keywords to recognize.

The name of the window function to apply to each audio frame before calculating the STFT.

The size of the signal window used to calculate the STFT, in number of samples - should be a power of 2 for maximum efficiency.

The length of time to skip each time the overlapping STFT is calculated, in milliseconds.

The length of a frame in the mel spectrogram used as an input to the wakeword recognizer encoder, in milliseconds.

The number of filterbank components in each mel spectrogram frame sent to the wakeword recognizer.

The size of the wakeword recognizer’s encoder window output.

The length of the sliding window of encoder output used as an input to the wakeword recognizer classifier, in milliseconds.

Timeout in seconds used for semaphore waits in the speech pipeline