Resonance bands
Broad bands of energy may reflect vocal tract resonances. Vowels often involve changing resonance patterns, but filtering, reverberation, and other sound sources can create similar bands.
The acoustic field guide
Start with a feature.
Then ask what it supports.
A spectrogram maps time, frequency, and energy. It can help locate and describe sounds. A shape alone rarely identifies a word or a speaker.
Reading the workspace
Time moves from left to right; frequency increases upward. Brighter colors show stronger energy within the displayed range. The brightness control changes the display, while the original recording stays intact. Drag a box to select a time and frequency region. Playback plays the selected time interval across the original frequency range.
Periodic harmonics
Regularly spaced frequency lines can indicate a repeating source such as vocal fold vibration. Whispered speech often lacks this periodic structure; absence of harmonics does not establish absence of speech.
Frication and bursts
Diffuse energy can accompany fricatives; brief transients can accompany stop releases. Airflow, rubbing, clicks, and mechanical noise can produce overlapping patterns.
Timing and energy
Mark onset, duration, gaps, and changes in amplitude. Syllable boundaries require context and listening; a bright patch is not automatically a syllable.
A useful listening pass
- Listen once at normal speed without a suggested transcript.
- Mark a short interval and describe what you hear in neutral terms.
- Inspect the corresponding energy pattern. Record time and frequency bounds.
- Replay briefly at normal and slower speed. Keep uncertain sounds uncertain.
- Write a phonetic or word hypothesis separately, then export the record.
Repeated listening and model agreement can both reinforce an incorrect interpretation. Known speech samples and independent listeners are useful references. The current lab does not score speech authenticity or establish threats, identities, motives, diagnoses, or crimes.
Repair research
The next phase will compare reversible gain and noise reduction against the original, using known speech, noise-only controls, and blinded listening. Any inferred or reconstructed spectral content must be identified as a processing result. The existing OpenEar VST3 prototype applies frequency-band attenuation; it does not recognize phonemes or reconstruct missing dialogue.