Vocal Analyzer

About Vocal Analyzer

Free browser tools that measure your voice: speaking pitch, live frequency, vocal range and voice type, plus a pitch tracker for voice training. No account, no install, no upload.

By Soluto, independent developer, not a clinician · Published

Who makes this

Vocal Analyzer is built and maintained by Soluto, an independent developer, not a clinician. The pages are written from the published sources listed below; no clinician or voice teacher has reviewed them. That is why the tools only measure and never give a diagnosis or a training plan — for those, see a speech-language pathologist or a qualified voice teacher.

How pitch is measured

Your microphone signal (or the file you choose) is decoded by your browser, reduced to about 16,000 samples a second and cut into short overlapping windows of about 85 ms. Each window goes through YIN (de Cheveigné & Kawahara, Journal of the Acoustical Society of America, 2002), which finds the repeating period of a voiced sound. Windows that are too quiet or have no clear period — silence, breath, “s” and “f” sounds — are skipped. A median over neighbouring windows removes single-window octave slips.

From the remaining values the analyzer reports your median, your 10th–90th percentile range and the share of voiced time in each band. The range tests record a note only after you hold it within about half a semitone for 0.3 seconds.

Accuracy

Synthetic tones. On voice-like test tones from 82 Hz to 1,047 Hz, including tones with added noise, the detector is within 1% of the true frequency; these checks run on every change to the code.

Real speech. We compared the analyzer with Praat, the standard phonetics program, on 80 read-aloud recordings from 40 adult speakers (half women, half men) in the public LibriSpeech corpus. Where both programs found a pitch, they agreed to within 0.3% at the median, and only 2% of those moments differed by more than 20%. Over a whole recording, our median speaking pitch was within 2% of Praat’s for 57 of the 80 recordings and within 5% for 73, with a typical difference of 1.1%. When it differs it tends to read a little high, because it skips more of the quiet, falling ends of phrases than Praat does. Treat a result as accurate to a couple of percent — enough to follow your own progress, not to compare a few hertz between people.

Background sound, very breathy or creaky voice and Bluetooth headset microphones in call mode all add error. Look at results from 20–30 seconds of speech rather than single moments. So far the tools have been tested in desktop Chrome; if something misbehaves on your phone or browser, please tell us.

Where the reference ranges come from

Not medical advice

These tools measure; they do not diagnose. If your voice is hoarse for more than a couple of weeks, hurts when you speak, or you want to change it safely, see a doctor or a speech-language pathologist.

Privacy

Audio is analysed in your browser and is never sent anywhere: this site is static files with no server code that could receive it. Microphone access ends when you press Stop or leave the page. The voice-training page keeps a short history of your results (date and numbers, never audio) in this browser’s local storage, which you can clear on that page.

You can check this yourself: open your browser’s developer tools, go to the Network tab, record or upload something, and watch — no request carries your audio.

Contact

Found a bug or a wrong number? Email hello@vocalanalyzer.org.