Configuration

Configuration options for lexigram-multimedia-beat.

Overview

Loaded from the multimedia_beat: section of application.yaml when installed standalone, or from multimedia.beat: when nested under the lexigram-multimedia umbrella. Environment variable prefix: LEX_MULTIMEDIA_BEAT__.

BeatAnalysisConfig extends lexigram.config.BaseConfig and declares config_section = "multimedia_beat", so framework config loading maps the YAML/env tree onto it and BeatAnalysisGenerationProvider consumes the instance directly.

multimedia_beat:
  backend: "librosa"            # librosa (in-process) | madmom (reference server)
  librosa_sample_rate: 22050
  madmom_base_url: "http://localhost:5600"
  timeout: 30.0

Basic Example

from lexigram.multimedia.beat import BeatAnalysisModule
from lexigram.multimedia.beat.config import BeatAnalysisConfig

app.add_module(
    BeatAnalysisModule.configure(
        config=BeatAnalysisConfig(backend="madmom", timeout=60.0)
    )
)

BeatAnalysisModule.configure() with no arguments uses BeatAnalysisConfig() — the in-process librosa backend. stub() also forces librosa regardless of YAML, which keeps tests server-free.

Options

Option	Type	Default	Description
`backend`	`Literal["librosa", "madmom"]`	`"librosa"`	Backend selection. `librosa` runs in-process (needs the `[librosa]` extra); `madmom` calls a reference server (needs `[madmom-server]`)
`librosa_sample_rate`	`int`	`22050`	Sample rate passed to `librosa.load()` by `LibrosaBeatAnalysisProvider`
`madmom_base_url`	`str`	`"http://localhost:5600"`	Base URL of the madmom server; `POST {base}/analyze`, `GET {base}/health`
`timeout`	`float`	`30.0`	HTTP timeout (seconds) for madmom requests (`aiohttp.ClientTimeout(total=...)`)

An unknown backend value raises ProviderNotInstalledError from contracts at register() time — startup fails fast with an actionable message rather than at first request.

Environment Variables

Variable	Description
`LEX_MULTIMEDIA_BEAT__BACKEND`	Beat backend: `librosa` or `madmom`
`LEX_MULTIMEDIA_BEAT__LIBROSA_SAMPLE_RATE`	Sample rate for the librosa backend
`LEX_MULTIMEDIA_BEAT__MADMOM_BASE_URL`	Madmom reference-server URL
`LEX_MULTIMEDIA_BEAT__TIMEOUT`	Request timeout in seconds

LEX_MULTIMEDIA_BEAT__BACKEND=madmom \
LEX_MULTIMEDIA_BEAT__MADMOM_BASE_URL=http://10.0.0.5:5600 \
LEX_MULTIMEDIA_BEAT__TIMEOUT=60.0 \
  python -m my_app

Under the umbrella, the nested key uses multimedia:: LEX_MULTIMEDIA__BEAT__BACKEND=madmom — see the umbrella’s Configuration doc.

Advanced Configuration

Two profiles, one deployment

multimedia_beat:
  backend: "madmom"
  madmom_base_url: "http://beat-worker:5600"
  timeout: 60.0

# off in local dev — include this small override file
multimedia_beat:
  backend: "librosa"

Full programmatic setup (madmom + resilience)

from lexigram.multimedia.beat import BeatAnalysisModule
from lexigram.multimedia.beat.config import BeatAnalysisConfig
from lexigram.resilience import ResilienceModule

app.add_module(ResilienceModule.configure())
app.add_module(
    BeatAnalysisModule.configure(
        config=BeatAnalysisConfig(
            backend="madmom",
            madmom_base_url="http://localhost:5600",
            timeout=30.0,
        )
    )
)

BeatAnalysisGenerationProvider.register() resolves RetryPolicyProtocol and CircuitBreakerProtocol optionally — configured via ResilienceConfig — and passes them to MadmomBeatAnalysisProvider so server calls are protected end to end.

Best Practices

Start with librosa; move to madmom only when syncopation/accuracy demands it.
Keep timeout comfortably above the madmom model’s per-request latency (30 s default is conservative for most content).
Never version a madmom_base_url pointing at a local machine into production — use env vars or a service name.
Pin librosa_sample_rate to your source material’s native rate to avoid resampling artifacts.
In tests, prefer BeatAnalysisModule.stub() so no server and no extra install is needed.