Read an audio analysis in its context

Start from a project or a single file#

A recording project keeps the capture plan, equipment, microphone geometry and instrument organization together. OrgRec 0.3.2 also provides Analyze Audio for a standalone file, without requiring a project catalogue or microphone setup first. This is useful for inspecting a received recording or testing an excerpt before committing to a larger capture workflow. An unknown instrument or recording context should remain unknown rather than being filled with plausible defaults.

For a long recording, choose an excerpt appropriate to the question. An attack, sustained region and release describe different behavior. Inspect the waveform for boundaries and interruptions, use the spectrogram to examine changes over time, and compare pitch and partial information within a clearly identified channel. A stable sustained region may be suitable for a pitch estimate while excluding the onset behavior relevant to articulation.

Pitch needs a reference and a scope#

A frequency estimate and a musical pitch label answer different questions. The frequency describes the analyzed signal under the selected method and region; the label interprets it against a reference, temperament and note identity. Noise, missing fundamentals, unstable attacks and ambiguous octave estimates can affect that interpretation. OrgRec’s provisional treatment of unknown pitch should be carried into a research note until the identification is reviewed.

cents = 1200 × log2(measured_frequency / reference_frequency)

A +88.01-cent offset relative to 440 Hz corresponds to:
440 × 2^(88.01 / 1200) ≈ 462.95 Hz

This calculation translates a stated offset; it does not establish why the offset exists. In the Cuntz analysis, the thesis reports a median over multiple tones. That aggregate can include temperament, and missing session temperature prevents a temperature-corrected comparison with the historical reference. It should not be presented as a direct measurement of one reference pipe under documented thermal conditions.

Count observations at the correct level#

Two channels from one take are two signal observations of a shared event. Differences between them can reflect position, room response and the recording chain. They are not automatically independent repetitions of the instrument measurement. Similarly, a difference between left and right spectral features is not a source-versus-resynthesis error unless the experimental design actually compares those two entities.

The thesis uses a multivariate, median-centered and MAD-scaled distance to flag review candidates. Its threshold belongs to a particular feature set and reference group. A flag suggests that a recording deserves inspection; it does not establish a defective pipe. Listen to the identified region, examine its acquisition conditions, compare neighboring recordings and retain the reviewer’s interpretation separately from the machine-generated indicator.

Keep original capture and analysis products distinct#

• Retain the original take and identify the analyzed excerpt by stable boundaries. Preserve the relationship between sample position and time.
• Record the selected channel, analysis method, parameters and tool version with each exported result.
• Treat segmentation, loop selection, level changes and other edits as derived products. Preserve the input that made a later revision possible.
• Review sustained-loop boundaries by listening as well as inspecting them. A smooth loop is a playback decision, not proof that the original sound was stationary.
• Keep analyst comments, uncertain note assignments and excluded material with the result rather than presenting only the accepted values.

OrgRec 0.3.2 supports analysis exports including PDF, JSON, CSV, PNG and SVG, plus complete export folders with excerpt and provenance checksums. Choose a form suited to reuse: a figure communicates a view; tabular or structured output supports further analysis; the identified audio excerpt lets another researcher inspect the signal. A checksum makes the exported evidence identifiable, while the recording context explains what that evidence means.