Use this URL in any client supporting remote Streamable HTTP MCP. Calls return structured JSON; inputs are processed without persistence.
Call inspect_wav
Accepts plain base64 of mono/stereo uncompressed integer PCM WAV (8/16/24/32-bit), up to 524288 decoded bytes and 15 seconds. Channels are zero-based. Normalize samples by 2^(bits-1); 8-bit WAV uses unsigned samples centered at 128. Null dBFS represents negative infinity for zero amplitude. Digital silence counts exactly zero-valued samples, not quiet speech or background noise. Full-scale sample counts are evidence of rail values, not a clipping diagnosis. No LUFS, true-peak or psychoacoustic loudness claim is made. Bytes are not stored or modified.
POST https://app.getsupers.com/sites/agent-wav-inspect/call
Content-Type: application/json
{
"audio_base64": "UklGRsQAAABXQVZFZm10IBAAAAABAAEAQB8AAIA+AAACABAAZGF0YaAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA"
}
The HTTP and MCP interfaces execute the same implementation. Validation errors are returned explicitly. See the tool notes above for its supported inputs and behavior.
PCM sample statistics distinguish exact silence from loudness
Read the explanation
The saved base64 example is mono sixteen bit PCM at eight thousand frames per second. It contains eighty frames, all exact zeros, for one hundredth of a second. At four pixels per count total, zero and nonzero samples measure three hundred twenty, three hundred twenty and zero. Peak, root mean square and DC offset are therefore zero, and the digital silence fraction is one. Zero amplitude dBFS is represented by null for negative infinity, not zero decibels. This follows from locally decoded fixture bytes; it does not prove a server invocation. Signed samples normalize by two to the power of bits minus one. A hypothetical sixteen bit sequence zero, sixteen thousand three hundred eighty four, zero and negative sixteen thousand three hundred eighty four becomes zero, one half, zero and negative one half. Peak magnitude is one half; mean DC is zero. Two of four samples are exact zero, a silence fraction of one half. At eighty pixels per count total, zeros and nonzeros measure three hundred twenty, one hundred sixty and one hundred sixty. Quiet nonzero audio is not digital silence, and the channel counts remain separate for stereo. For that example sum of squared normalized samples is one half, divided by four gives one eighth. Taking the square root gives approximately zero point three five three five five three root mean square, below the one half sample peak. At six hundred pixels per amplitude peak, RMS and absolute DC have widths three hundred, about two hundred twelve point one three two and zero. Rail-valued sample counts are evidence of full scale samples, not a clipping diagnosis. This inspector does not calculate LUFS, true peak or perceived loudness and changes no audio. Offline native argument editing is preserved, but the absent shared application client means no HTTP or MCP result is verified.
PCM Signal Metrics and Integer WAV Analysis
How does this inspector calculate peak, RMS, DC offset, and exact digital silence from raw integer PCM WAV audio?
The tool reads uncompressed RIFF/WAV headers and parses little-endian integer PCM frames for mono or stereo streams at 8, 16, 24, or 32 bits per sample. Samples are normalized against their theoretical full scale of 2^(bits - 1), mapping unsigned 8-bit values centered around 128 to zero and evaluating signed integers for 16, 24, and 32 bits. The inspector computes per-channel peak magnitude, root-mean-square energy, arithmetic mean DC offset, full-scale rail hit occurrences, and the fraction of frames containing digital zero.
Digital silence measures exact zero-valued integer samples (value 0 for signed PCM, or 128 for 8-bit unsigned PCM), not ambient background noise or quiet speech. Full-scale sample counts tally boundary rail values (-2^(bits-1) or 2^(bits-1)-1) as indicators of potential saturation, but do not diagnose inter-sample peaks or analog waveform clipping. This tool does not compute ITU-R BS.1770 LUFS integrated loudness, psychoacoustic contours, or oversampled true peaks.
Try a worked example
Leaving the default 16-bit 8 kHz mono silent audio fixture in the JSON arguments textarea and clicking 'Run tool' executes inspect_wav. Because all 160 sample bytes represent digital zeroes, the resulting JSON output reports sample_peak of 0.0 with sample_peak_dbfs as null (negative infinity), rms of 0.0 with rms_dbfs as null, dc_offset of 0.0, digital_silence_fraction of 1.0, and 0 full-scale rail hits across 80 frames (0.01 seconds).
Container Parsing and Sample Normalization
Uncompressed integer PCM files store audio as raw sample frames following a RIFF header. According to Python's standard wave library specifications, valid uncompressed PCM formats indicate compression type 'NONE' and encode sample widths of 1, 2, 3, or 4 bytes per channel frame. Python Standard Library: wave — Read and write WAV files
The inspect_wav implementation centers 8-bit unsigned bytes by subtracting 128, while 16-bit, 24-bit, and 32-bit samples are parsed as little-endian signed integers. Each sample value is scaled relative to full_scale = 1 << (8 * width - 1) before computing squared sums and absolute maximums.
Signal Statistics and Silence Detection
Root-mean-square (RMS) energy is calculated as sqrt(sum(sample^2) / frames) / full_scale. When peak or RMS energy is greater than zero, the decibels relative to full scale (dBFS) is computed using 20 * log10(normalized_value); if amplitude is zero, dBFS returns null to signify negative infinity.
Digital silence fraction explicitly tracks the proportion of frames where the integer value equals zero exactly. It does not gate analog noise floors or speech pauses, providing an objective check for synthetic digital zeros or unwritten buffers.