Audio

Rerun plays back audio files logged with the AssetAudio archetype. See its reference page for the supported formats, how playback lines up with time, and an example.

Raw PCM samples

There is no archetype for raw PCM samples yet. Instead, encode the samples as a WAV file in memory and log that. For live audio, do this for short segments, and log each segment at the time it starts:

import io
import wave

import numpy as np
import rerun as rr


def wav_bytes(samples: np.ndarray, sample_rate: int) -> bytes:
    """Encode mono int16 samples as a WAV file."""
    buf = io.BytesIO()
    with wave.open(buf, "wb") as wav:
        wav.setnchannels(1)
        wav.setsampwidth(2)  # 16-bit
        wav.setframerate(sample_rate)
        wav.writeframes(samples.astype(np.int16).tobytes())
    return buf.getvalue()


rr.init("rerun_example_pcm_audio", spawn=True)

sample_rate = 16_000
t = np.arange(sample_rate) / sample_rate
samples = (np.sin(2 * np.pi * 440 * t) * 10_000).astype(np.int16)  # One second of 440 Hz

rr.set_time("time", duration=0.0)
rr.log("audio", rr.AssetAudio(contents=wav_bytes(samples, sample_rate), media_type="audio/wav"))

To plot the raw signal instead of playing it, log the samples as Scalars, for example with send_columns.