Alignment layers

Vocal Analyzer can expose aligned segment layers alongside the core acoustic result. These layers are optional timing aids:

• Words come from transcript timestamps when transcription is enabled. They are available even when no phone-level aligner is configured.
• Phones are optional forced-alignment intervals such as S, Z, IH, or UW.
• Vowels are the subset of phone intervals that can feed the per-vowel formant path.

Alignment layers are available only in the installed app. The browser version does not transcribe or align, so its results have no word, phone or aligned vowel layers.

The acoustic measurements still come from Praat/Parselmouth. Alignment layers only say where to look in time; they do not replace the acoustic extractors.

Current backend scope

The built-in transcript backend reuses transcript word timestamps. The experimental mfa backend is an adapter for a separately installed Montreal Forced Aligner command plus caller-supplied dictionary and acoustic model paths. MFA is not bundled with the default install.

Phone and vowel layer support is currently scoped to English (en). Other languages can still use transcript word timestamps, but phone and vowel statuses are marked unavailable rather than guessed.

How alignment changes analysis

Alignment is additive:

• Sibilants can use aligned /s/ and /z/ phone intervals when they are present. That phone-gated path can raise confidence when the spectral detector and aligned phone layer agree.
• Per-vowel formants can use aligned vowel intervals when they are present. Whole-recording formants remain the continuity baseline for history, comparisons, and benchmark checks.
• If a requested phone/vowel layer is unavailable, the result carries a status reason such as phone_model_unavailable, unsupported_language, or alignment_backend_unavailable.

What alignment does not claim

Alignment layers are not pronunciation scoring, accent assessment, speaker verification, clinical screening, or listener-outcome prediction. They do not prove that a phone was produced correctly; they only provide timing intervals from an optional upstream alignment source. Recording quality, transcript errors, dictionary coverage, language mismatch, and aligner model choice can all move or remove segments.

Use aligned segments as a more precise descriptive lens when the status is available, and fall back to the non-aligned acoustic view when it is not.