sense-sdk-python-tutorials

• sense_file.py - file-mode: load a local audio file and run inference to completion.
• sense_stream.py - stream-mode: capture the microphone and infer in real time.

1. Virtual environment + dependencies

# Create and activate an isolated environment
python -m venv .venv
. .venv/bin/activate

# Install the Sense wheel for your interpreter/arch (built by the release pipeline)
pip install sense/sense-<ver>-cp3XY-cp3XY-linux_<arch>.whl

# Extra deps for the tutorials (microphone capture for sense_stream.py)
pip install -r sense-stream/requirements.txt

sense_stream.py uses sounddevice, which needs the system PortAudio library:

sudo apt-get install -y libportaudio2      # Debian/Ubuntu

The Sense wheel itself has no dependencies.

2. Run

Set PROJECT_KEY at the top of each script. Run from the repository root so the config.json there is used if present (missing is non-fatal).

python sense-file/sense_file.py audio-files/siren.wav   # file mode
python sense-stream/sense_stream.py                     # mic mode (Ctrl-C to stop)

3. API

import sense

with sense.session(project_key, config_json):         # init() / terminate()
    p = sense.create_file_processor(path, listener)   # or create_stream_processor(listener)
    p.sensitivity = "HIGH"                            # property (get/set)
    p.result_summary_enabled = True                   # write-only config properties
    with p:                                           # start() on enter, stop() on exit
        p.push(pcm, channels, sense.SampleFormat.INT16, rate)   # stream mode

• Constants: sense.SampleFormat.{FLOAT32,INT16,INT32,FLOAT64} (an IntEnum; pass a member straight to push).
• Results: subclass sense.ResultListener and override on_result(frame). frame has tags (.name, .probability), start_time, end_time, prediction_time_ms, summaries, error, and is_ok().
• The underlying snake_case methods (get_sensitivity(), stop(), ...) remain available; the properties/context-managers are idiomatic aliases.