Recording tips

Vocal Analyzer works on any uploaded audio file, but the quality of the measurements depends heavily on the recording. This page walks through what to aim for so the dashboard shows you a clean signal instead of a signal filtered through room noise.

Microphone and room

• Any modern microphone is fine. A USB condenser, a laptop mic, or a phone earbuds mic all produce usable recordings. The key is consistency across sessions if you are tracking progress over time.
• Quiet room beats fancy mic. A laptop mic in a carpeted room usually gives cleaner harmonics-to-noise ratio than a studio mic in a room with a fan, a fridge, or an HVAC cycle.
• Avoid reverb-heavy rooms (bathrooms, stairwells). Long reflections bias voice-quality and formant estimates because the tool is designed for direct-path speech.

Distance and positioning

• Hold the microphone a consistent distance from your mouth across sessions. 10–20 cm (a hand span) is a good baseline for most mics.
• Slightly off-axis (the mic aimed at your chin, not at your lips) reduces plosive pops without losing high-frequency detail.
• Stay still while recording. Large head movements cause level swings that show up as intensity variation, not voice variation.

Caveat: Intensity values in Vocal Analyzer are uncalibrated. Comparing loudness across sessions is not meaningful even with a fixed mic distance — the tool's intensity metric is only useful as a within-recording relative cue.

Clipping

Watch the input-level meter in your recording app. Peaks should sit comfortably below the top of the meter. Clipping (flat tops on the waveform where the signal hits the maximum) irreversibly damages the harmonic content the analyzer needs — it biases voice quality, HNR, and formant estimates, and no post-processing can undo it.

If your app does not show levels, record a test clip and inspect the waveform. If the tops look chopped off, move the mic further away or lower the input gain.

Sample rate

• Record at 16 kHz or higher. Below 16 kHz, the analyzer skips resonance and voice-quality extraction entirely because the Nyquist limit puts the formants and high-frequency voice-quality cues into the noise floor. See Methodology for the warning codes that fire when the sample rate drops.
• 44.1 kHz or 48 kHz is the safe default — it matches what most recording apps and OS-level settings produce anyway.

Session length

• 3 to 30 seconds of continuous voiced speech is the practical sweet spot. Under ~3 seconds there are not enough voiced frames for stable median and percentile statistics; past ~30 seconds you hit diminishing returns on the per-metric medians.
• Read or speak, don't vocalize vowels alone. Connected speech gives more realistic formant and prosody estimates than sustained vowels. A short paragraph from a book or a description of your day works well.

Uploading or recording in the app

Both versions accept .wav, .mp3, .aiff, and .aif files. Upload them with Upload recordings on Home (or drop them on its Drop recordings to analyze area), or with Upload audio file in Studio. Other formats, such as .flac or .m4a, are not accepted; convert them to WAV first. The installed app accepts uploads up to 500 MB, but most voice recordings are well under 5 MB.

You can also record in the app, from your browser's microphone, in both versions. On the Practice page, press Start, speak, and press Stop. Save to practice log keeps the take, analyzes it like an uploaded recording, and adds it to your recordings on Home; Discard throws it away. See Live microphone preview for microphone requirements and how the live numbers differ from the saved analysis.

An external recorder (QuickTime, Voice Memos, Audacity, OBS, etc.) works too; save the file as WAV and upload it.

Note: Prefer an uncompressed format (.wav or .aiff) over .mp3 when possible. Lossy compression discards high-frequency detail that sibilant and voice-quality metrics rely on. The dashboard will still produce a result from an MP3, but the voice-quality ladder (see Voice quality) may sit one rung lower than on an equivalent WAV.

What to do if the dashboard looks wrong

If a metric lands in an unexpected band, check the top of the dashboard for a Measurement warning tile. The most common causes of surprising numbers are covered in Troubleshooting.

Next: Your first analysis.