Audio Sample Rate, Channel, and Clipping Review: Separate Container Facts from Signal Findings

The source dates shown in this guide identify documentation checks on 22 October 2026, not original publication dates. Follow each citation for the documentation edition being discussed.

Record the artifact reference, digest or controlled identifier, container, codec, duration, sample rate, channel count, channel layout, sample format, and whether the object is an original or a derivative. A channel count is not the same as a channel layout: two channels may be left and right, dual mono, or an application-specific pair. Treat an unknown layout as unknown instead of guessing from the count.

The python-soundfile, 22 Oct 2026 (opens in a new tab) documentation describes a SoundFile as having a sample rate, data format, and number of channels, and its frame index represents one time-step containing one sample per channel. PyAV exposes analogous AudioFrame properties such as sample_rate, layout, samples, and format PyAV, 22 Oct 2026 (opens in a new tab). These are observations about a decoded representation, not evidence of recording intent or listening quality.

Metadata is a declaration to compare, not an independent proof. For a bounded review, collect a small summary per decoded audio frame: frame index, timestamp if present, sample rate, layout, sample count per channel, and a stable issue class. FFmpeg’s ashowinfo reports per-frame timestamps, sample format, channel layout, rate, and samples per channel FFmpeg, 22 Oct 2026 (opens in a new tab).

Selected stream fields

ffprobe -v error -select_streams a:0 \
  -show_entries stream=index,codec_name,sample_rate,channels,channel_layout,sample_fmt,duration \
  -of json MEDIA_REFERENCE

The FFmpeg, 22 Oct 2026 (opens in a new tab) documentation supports selecting audio streams, stream fields, and machine-readable output. Keep MEDIA_REFERENCE as a controlled identifier in a report. Do not include a private path, a recording title, tags, or raw packet data in a public diagnostic.

Synthetic decoded summary

{
  "declared": {"sample_rate_hz": 48000, "channels": 2, "layout": "stereo", "sample_format": "fltp"},
  "decoded_frames": [
    {"index": 0, "sample_rate_hz": 48000, "channels": 2, "samples_per_channel": 1024},
    {"index": 1, "sample_rate_hz": 48000, "channels": 2, "samples_per_channel": 1024}
  ]
}

The values are invented for the worksheet. A mismatch such as two declared channels and one decoded channel is a stream-contract finding. It does not explain why the change occurred or whether a listener would prefer one representation. Record the first and last affected frame, not the raw samples.

Clipping is easiest to misuse when it is turned into a universal threshold. First document the decoded sample representation and the policy. For a synthetic normalized float fixture, one local policy might count a sample at or beyond full scale as a potential clipped sample.

Bounded synthetic sample check

from collections.abc import Iterable

def clipping_summary(samples: Iterable[float], limit: float = 1.0, max_samples: int = 100_000):
    inspected = 0
    clipped = 0
    peak = 0.0
    limited = False
    for value in samples:
        if inspected >= max_samples:
            limited = True
            break
        magnitude = abs(value)
        inspected += 1
        clipped += magnitude >= limit
        peak = max(peak, magnitude)
    return {"sample_count": inspected, "potential_clipped_samples": clipped, "absolute_peak": peak, "sample_limit_reached": limited}

synthetic_channel = [0.10, -0.35, 0.99, 1.0, -1.02]
print(clipping_summary(synthetic_channel))

This function is deliberately bounded by max_samples and does not log values. The synthetic result would report two potential clipped samples under this policy. On real data, define whether the threshold is inclusive or exclusive, whether the check is per channel, and how floating-point rounding is handled. An encoded integer stream and a normalized float stream need an explicit representation contract before their thresholds are compared.

FFmpeg’s volumedetect can report mean volume, maximum volume per sample, and a histogram. Its documentation also states that the filter works on 16-bit signed integer samples and converts other input as needed FFmpeg, 22 Oct 2026 (opens in a new tab). Use that result as a diagnostic in the filter’s representation, not as a complete clipping count or a listening judgment. If per-channel counts are needed, decode into a bounded analysis buffer and report only counts, positions, and policy identifiers.

Record one row per check: artifact reference, stream index, declared rate, observed rate set, declared channel layout, observed layout set, sample format, inspected frame range, peak policy, potential clipping count, issue class, evidence reference, owner decision, and unresolved questions.

Keep “not inspected,” “not available,” and “not observed under this policy” distinct. A short window can miss a transient event, and a decoding-software summary cannot answer whether content is intelligible or pleasant to hear.

The public report can contain a controlled artifact identifier, stream index, frame range, stable issue class, numeric counts, and tool versions. It should not contain raw samples, private paths, recording names, voice labels, or unbounded decoding errors. If a source must be revisited, retain the evidence reference and owner-approved handling boundary rather than copying the audio into a worksheet.

Pair this check with the data provenance log and export guidance. Those resources own source, rights, processing, and broader export questions; this guide owns sample rate, channel, and bounded signal observations.

After the owner has a supported non-health scope, authority to describe the artifact, and a metadata or synthetic review packet, the qualified appraisal request path may be a next preparation step. It does not validate sample rate, accept audio, produce a quote, match a buyer, issue a legal opinion, or grant a license. Keep raw recordings outside this guide and within the owner-approved handling boundary.

Takeaways

• Record sample rate, channels, layout, and sample format as separate contract fields.
• Compare declarations with bounded decoded-frame observations instead of treating metadata as proof.
• Define clipping representation, threshold, window, and channel policy before counting potential crossings.
• Use stable issue classes and counts; keep raw samples and private labels out of shared diagnostics.
• Treat a clean technical check as one input to later provenance, privacy, rights, and accuracy review.