on_boarding.md

graph LR
    CLI_Argument_Parser["CLI Argument Parser"]
    Configuration_Manager["Configuration Manager"]
    Audio_Input_Handler["Audio Input Handler"]
    Mel_Spectrogram_Processor["Mel Spectrogram Processor"]
    Speech_Embedding_Engine["Speech Embedding Engine"]
    Wake_Word_Inference_Engine["Wake Word Inference Engine"]
    Output_Handler["Output Handler"]
    CLI_Argument_Parser -- "configures" --> Configuration_Manager
    Configuration_Manager -- "provides settings to" --> Mel_Spectrogram_Processor
    Configuration_Manager -- "provides settings to" --> Speech_Embedding_Engine
    Configuration_Manager -- "provides settings to" --> Wake_Word_Inference_Engine
    Audio_Input_Handler -- "streams raw audio to" --> Mel_Spectrogram_Processor
    Mel_Spectrogram_Processor -- "outputs mel-spectrograms to" --> Speech_Embedding_Engine
    Speech_Embedding_Engine -- "outputs speech embeddings to" --> Wake_Word_Inference_Engine
    Wake_Word_Inference_Engine -- "sends detection events to" --> Output_Handler
    click Speech_Embedding_Engine href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/openWakeWord-cpp/Speech_Embedding_Engine.md" "Details"
    click Wake_Word_Inference_Engine href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/openWakeWord-cpp/Wake_Word_Inference_Engine.md" "Details"

Loading

Details

The system operates as a real-time audio processing pipeline for wake word detection. It begins with a CLI Argument Parser that initializes the Configuration Manager with operational parameters and model paths. The Audio Input Handler continuously streams raw audio, which is then transformed into mel-spectrograms by the Mel Spectrogram Processor using a dedicated ONNX model. These mel-spectrograms are subsequently processed by the Speech Embedding Engine to generate speech embeddings, also via an ONNX model. The Wake Word Inference Engine then utilizes these embeddings with multiple wake word ONNX models to identify wake word occurrences. Finally, the Output Handler is responsible for presenting the detection results to the user.

CLI Argument Parser

Parses command-line inputs to configure the application's operational parameters, including model paths, audio device settings, and detection thresholds.

Related Classes/Methods:

• ensureArg
• printUsage

Configuration Manager

Centralizes and provides access to all application settings, ensuring consistent configuration across different processing stages.

Related Classes/Methods:

• Settings

Audio Input Handler

Manages the acquisition of raw audio data from the system's microphone or a specified audio source, streaming it for further processing.

Related Classes/Methods: None

Mel Spectrogram Processor

Transforms raw audio samples into mel-spectrogram features, utilizing an ONNX model (melspectrogram.onnx) for efficient signal processing.

Related Classes/Methods:

• audioToMels
• Ort::Session (usage of)
• melspectrogram.onnx

Speech Embedding Engine [Expand]

Generates high-dimensional speech embeddings from mel-spectrograms using a dedicated ONNX model (embedding_model.onnx), preparing data for wake word classification.

Related Classes/Methods:

• melsToFeatures
• Ort::Session (usage of)
• embedding_model.onnx

Wake Word Inference Engine [Expand]

Executes multiple wake word ONNX models on the speech embeddings to determine the probability of a wake word detection, acting as the core classifier.

Related Classes/Methods:

• featuresToOutput
• Ort::Session (usage of)
• *.onnx (various wake word models)

Output Handler

Processes and presents the results of wake word detection, such as printing messages to the console or triggering further actions upon detection.

Related Classes/Methods:

• featuresToOutput

FAQ