FAQ

Why does my pitch look feminine but my formants look masculine (or vice versa)?

Because pitch and resonance are independent cues. F0 is controlled mostly by how tense and thick your vocal folds are; formant frequencies come from the shape and length of your vocal tract. You can shift one without shifting the other. In the perception literature, this combination places the voice in the overlap zone where listener agreement is lower. That is acoustically normal, not a bug. See Voice and gender perception.

Why are my numbers different from another voice-analysis tool?

Several reasons, usually in combination:

• Different aggregation. Whole-recording median vs per-vowel mean vs long-term average will all differ on the same audio.
• Different algorithms. Pitch extraction (autocorrelation vs YIN-style), formant extraction (Burg LPC vs peak-picking), and voice-quality corrections (Iseli-Alwan vs uncorrected) all give different numbers.
• Different band definitions. Each tool picks different boundaries from different source studies.
• Different pre-processing. Some tools apply pre-emphasis, high-pass filters, or normalisation; Vocal Analyzer does not by default.

None of this means one tool is "right" and the other is "wrong". Voice analysis is a field with many valid choices; compare numbers against themselves (this tool, across sessions) rather than against another tool's numbers directly.

Does recording quality really matter that much?

Yes for voice quality, HNR, and sibilants — noticeably. Low SNR and low sample rate are the biggest offenders; see Recording tips and Methodology. Pitch is the most robust metric under degraded conditions, which is why the SNR warning explicitly calls pitch out as "usually the most robust metric".

What is a realistic target?

Vocal Analyzer deliberately does not prescribe target values. The reference bands exist so you can see where a recording lands; they are not "scores to achieve". Two reasons for this framing:

1. Published bands are descriptive, not prescriptive — they describe what listeners heard in specific corpora, not what listeners "should hear" for any given speaker.
2. Targets without context mislead. A pitch of 200 Hz in a pressed, hyperarticulated voice lands very differently from 200 Hz in a natural connected speech recording. A voice teacher is the right partner for turning "where does my voice land?" into "what do I practice?".

Can I compare across sessions?

Yes — with the same formant mode and similar recording conditions.

Good comparisons:

• Same hardware, same room, same target text, same formant mode.
• Track pitch median, dispersion, and voice-quality-ladder trends across a dozen sessions over weeks.

Bad comparisons:

• Whole-recording dispersion in one session vs per-vowel dispersion in another — different aggregation.
• Uncalibrated intensity across sessions — see Intensity.
• A WAV from an earlier session vs an MP3 from a later session.

Is my audio uploaded anywhere?

No. Analysis runs on your own device in both versions.

• In the browser version, the analysis runs inside your browser, and recordings and results are stored in that browser's storage for the site. The app downloads its own software from the site but does not send your recordings, results or settings to it.
• In the installed app, the web UI runs on your machine (make up or make dev-serve); uploads go to a local temporary directory; sessions persist under your local platform-data directory. No network calls leave your machine during analysis.

See Storage and privacy.

Does this tool know what gender I am?

No. It measures acoustic cues that correlate with listener judgement of perceived gender in English-speaking adult corpora. It does not infer, predict, or determine anyone's gender, identity, or self-perception. See What is Vocal Analyzer?.

Can I use this for languages other than English?

You can, but with caveats:

• Pitch (F0) measurement is language-agnostic — semitones are semitones. The bands are calibrated against English adult speech, so they may not overlay cleanly for other languages.
• Formant measurement is language-agnostic, but the per-vowel aggregation uses an English vowel class set.
• Tonal languages will have F0 dynamics the tool is not designed for — the statistics still compute, but the perceptual framing does not fit tonal systems.