Audio forensics for recordings that may be edited or synthesised. We examine spectral continuity, background noise and encoding traces, and mark the exact time ranges on the waveform and spectrogram.
[ 01 / Paradigm shift ]
Target vs Reality
Current
Trusting Your Ears
A convincing voice passes a casual listen. Edits and synthetic speech are hard to hear and hard to prove after the fact.
Target
Measured on the Spectrogram
Sensors measure continuity, hum and encoding behaviour, and the report marks the time ranges with the spectral evidence beside them.
[ 02 / Method ]
Built for scrutiny
01Preserve
The original recording is hashed and sealed.
02Decode
A reference copy is made while codec details are kept.
03Measure
Spectral, noise and hum sensors run independently.
04Mark
Time ranges are annotated with spectra and notes.
Audio Analysis Chain
Control planeAnalysis plane
01Decode & Normalise
PCM · METADATA
Decodes to a reference representation while preserving the original stream and codec details.
02Spectral Sensors
SPECTROGRAM · PHASE
Examines spectral continuity, band limits and phase behaviour for signs of synthesis or editing.
03Noise Floor
CONTINUITY
Checks that background noise and room response stay consistent through the recording.
04Network Hum
ENF · RE-ENCODING
Looks for electrical-network hum patterns and traces of repeated encoding.
05Annotated Report
TIMELINE · SPECTRA
Marks time ranges of interest on the waveform and spectrogram with supporting measurements.
Control plane / CUSTODY LOG
Case Ledger
Links the source recording to every spectrogram and measurement.
Hover or tap a stage to inspect
[ 03 / Core systems ]
Core sensors
01 / SPECTRAL01 / 04
SPECTRALCONTINUITY
Natural speech has characteristic spectral and phase behaviour. Edits and vocoders can disturb it.
SPECTROGRAM · PHASE · BAND LIMITS
Inspect the signal's structure.
MEASURES
Spectral smoothness, band limits and phase coherence over time.
REPORTS
Time ranges with abrupt or unnatural spectral behaviour.
LIMITS
Telephone and conferencing codecs alter spectra, so reports state the channel assumptions.
02 / NOISE02 / 04
NOISE FLOORCONTINUITY
Rooms and microphones leave a steady background. Splices often interrupt it.
ROOM RESPONSE · BACKGROUND
Listen to what is under the voice.
MEASURES
Background level and character, and changes in room response.
REPORTS
Points where background changes without an acoustic reason.
LIMITS
Real environments change, so the report shows the change and leaves the interpretation to the examiner.
03 / NETWORK HUM03 / 04
ELECTRICALNETWORK
Mains hum can be matched to an expected pattern for the recording time and region.
ENF · TIMELINE
Use physics as a witness.
MEASURES
Hum frequency variation across the recording.
REPORTS
Discontinuities in the hum track that may indicate cuts.
LIMITS
Not every recording captures hum, and in that case the sensor reports no result.
04 / ENCODING04 / 04
RE-ENCODINGTRACES
Repeated compression leaves marks that single-pass recordings do not have.
CODEC · QUANTISATION
Count the passes.
MEASURES
Codec signatures, frame boundaries and quantisation patterns.
REPORTS
Evidence of multiple encodings and segments with different histories.
LIMITS
Messaging apps re-encode by design, so provenance matters when reading the result.
[ 04 / Reference patterns ]
Where it applies
INVESTIGATION
Disputed recorded call
Time-marked findings that can be played back against the spectrogram.
FINANCIAL
Voice-authenticated transactions
Screening for replayed or synthetic voices on high-risk calls.
COMPLIANCE
Recorded instructions
Checks that archived recordings were not edited after the fact.
[ 05 / Delivery & deployment ]
From evidence to report
Reports carry annotated waveforms and spectrograms, plus the sensor settings. Audio never leaves your environment.
Audio forensics for recordings that may be edited or synthesised. We examine spectral continuity, background noise and encoding traces, and mark the exact time ranges on the waveform and spectrogram.
What changes compared with the current approach?+
A convincing voice passes a casual listen. Edits and synthetic speech are hard to hear and hard to prove after the fact. Sensors measure continuity, hum and encoding behaviour, and the report marks the time ranges with the spectral evidence beside them.
Can it run on-premise or air-gapped?+
Yes. Analysis runs on your own hardware, and the platform can be installed with no connectivity. Reports are exported as signed PDF and machine-readable JSON.
How do I start?+
Request a briefing by form, WhatsApp (+60 17-773 7302) or email (kumar@spearcompute.com). We scope a pilot on samples you provide.