Your First Audio Scene
The following tutorial page will guide you through the necessary steps to create and render an audio scene. "Audio scene" is how we refer to the process of convolving impulse responses (often from a CAD based device) with speech and noise in the Treble SDK. With the audio scene generator, we have created helper methods that makes it straightforward to define relative levels (signal-to-noise ratios or SNRs) between talkers and noise, create conversational rules governing source turn-taking, simulate measurement noise, and more. The scene generator outputs audio files with associated metadata, which can be used for downstream tasks, for instance training or evaluating neural networks.
While you may run the following notebook directly by calling jupyter notebook from the command line, we recommend using Visual Studio Code and its Python extension for ease-of-use.
The notebook assumes that you have installed the SDK and have a valid credential file.
The notebook uses a premade dataset, available to all users in the SDK, so it is not necessary to have run any simulation before starting creating audio scenes.
After opening the notebook in VSCode, press the "Select Kernel" button and choose "Python Environments". Choose the python environment where the SDK is installed.