1) Open the AI Voice Inspector Interface. 2) Navigate to the "Select Samples" Tab: This tab offers preset options to quickly try out the tool. Choose from the provided samples of real and synthetic speakers. Click "Submit." 3) View the Output: The output will appear in the "Classification" tab on the right side. The selected speaker waveforms will also be displayed. You can play and listen to the waveforms. 4) Clear Selections: Before making a new choice, click "Clear." 5) Navigate to the "Upload or Record Audio" Tab: Use this tab if you wish to try the tool with your own voice or upload a speech file. 6) Record Your Own Speech: Click "Record" to start recording your speech. Click "Stop" when done. Submit the recording. 7) Upload a Speech File: Drop a file into the drop section. Click "Submit." 8) Clear Selections: Make sure to click "Clear" before the next input. 9) Test and Leave Feedback: Feel free to test the tool and provide your feedback.
About
The AI Voice Inspector POC build leverages deep learning techniques to differentiate between real and synthetically generated audio speech samples. Key aspects: Feature Engineering: Extracts spectral and temporal features from audio waveforms using advanced signal processing techniques. Cybersecurity Application: Reliably detecting synthetic and deep fake audio content, providing a robust defence mechanism against such emerging audio-based attacks. Core deep learning model: Keras Bidirectional GRU and FN architecture engineered to deliver robust accuracy and generalisation. Diverse Dataset: Synthetic audio from cutting-edge open-source TTS/vocoder models, and real speaker recordings from a recognised speech corpus. Continuous development: Further training on additional speech corpora for robustness against evolving audio synthesis techniques. Advanced Model Variation: In development. This cutting-edge build has the potential for reliable synthetic audio detection, maintaining privacy and security against AI-driven threats.
