Maintained by Mayer · Last updated 2026-09-28
The pitch, key and tempo tools on Online Pitch Changer run inside your browser. Your file is decoded on your device, changed by one shared audio engine and saved back to your device; nothing is uploaded. This page explains each step in plain words and publishes the accuracy we measured in our own tests, including the limits.
What happens when you open a file
The file tools, such as the Online Key Changer, the Audio Pitch Changer and the Pitch Shifter, all follow the same five steps.
- You choose a file or record your voice. The page reads it from your device.
- Your browser decodes it with its own decoder, at the file’s own sample rate. That is why a format can open in one browser and not in another (older Safari cannot open OGG, for example).
- The engine starts in a Web Worker, a background thread, so the page stays responsive. Most tools render a short section first, then the whole file. In our desktop tests a 3-minute song started playing in under 0.9 seconds and was fully ready in 4 to 7 seconds. Phones are slower; we have not measured them yet.
- The preview is the download. In our tests the sound being played matched the downloaded WAV sample by sample, within 1 step of the 16-bit scale.
- You save WAV or MP3. WAV keeps the sample rate and channels of your file. MP3 is encoded on your device by the open-source lamejs encoder, which loads only when you ask for an MP3. Close the tab and the audio is gone.
How the engine changes pitch, tempo and key
Pitch without changing speed. A band-limited resampler changes pitch and length together, like a record played faster. A phase vocoder with phase locking (the method described by Laroche and Dolson in 1999) then stretches the sound back to its original length. Each semitone multiplies every frequency by 21/12 (about 1.0595), so +12 semitones doubles them. Tools shift up to ±12 or ±24 semitones, plus cents (100 cents = 1 semitone).
Tempo. The time-stretch on its own changes the length and keeps the pitch; that is how the Pitch and Tempo Changer changes speed. Its “linked” mode uses the resampler alone, so pitch and speed move together. See pitch vs tempo.
Key and tempo detection. The engine measures how strongly each of the 12 notes sounds, removes harmonics and compares the result with published key profiles (Krumhansl–Kessler, Temperley, Albrecht–Shanahan). You get the key, the relative key and a confidence score. Tempo comes from the rhythm of note onsets.
Formant correction. Shifting a voice also moves its resonances (formants), which causes the “chipmunk” sound. An optional correction keeps them in place; it is on by default in the Vocal Pitch Changer. A result is turned down only if it would clip; nothing is compressed.
Video, YouTube, microphone and theory tools
- Video Pitch Changer: the sound goes through the engine and the picture is copied into the new file untouched, not re-encoded.
- YouTube Pitch Changer: nothing is downloaded. The video plays in YouTube’s official embedded player; when you press Start, your browser asks to share this tab’s sound, and a real-time shifter on the page changes it. This needs Chrome or Edge on a computer; phones, Safari and Firefox cannot share a tab’s sound. YouTube sets its own cookies as soon as the player loads.
- Microphone tools: the Vocal Range Test and the Voice Type Test run a small pitch detector (YIN) on the live signal; a note counts only after you hold it steady for half a second.
- Theory tools such as the Semitone Calculator do music maths on the page and never touch your audio.
Measured accuracy
How we measured. On 27 and 28 September 2026 each tool ran on a test copy of WordPress (with the site’s theme and Yoast SEO) in Microsoft Edge, headless, at desktop and phone widths. We used synthetic test signals (a 440 Hz and 660 Hz stereo tone, click bursts at a known tempo, songs synthesised in all 24 keys, synthetic vowels) plus three real recordings for key detection. Microphone tools got a simulated microphone playing a known file. Every download was decoded again and measured outside the tool. “Worst” is the largest error in the test set.
| Tool | What we measured | Worst result | Tested |
|---|---|---|---|
| Online Key Changer | 11 WAV and 2 MP3 exports, −12 to +12 semitones; 3 chords | Pitch 0.041 cents (tones), 0.09 cents (chords); length 0.000 % | 2026-09-27 |
| Audio Pitch Changer | 11 WAV and 2 MP3 exports; 7 input formats; 48 and 96 kHz files | Pitch 0.080 cents; length 0.000 % | 2026-09-27 |
| MP3 Pitch Changer | 4 converted MP3s, 128 to 320 kbps | Pitch 0.045 cents; length +0.32 % | 2026-09-27 |
| Pitch Shifter | 11 WAV exports, −24 to +24 semitones, ±50 cents | Pitch 0.099 cents; length 0.000 % | 2026-09-27 |
| Pitch and Tempo Changer | 15 WAV exports at 50 to 200 % speed; a real 120 BPM song set to 100 BPM | Pitch 0.285 cents; tempo 0.018 %; song re-measured at 100.00 BPM | 2026-09-27 |
| Change Pitch Without Changing Speed | 8 WAV exports and 1 MP3 | Pitch 0.045 cents; tempo 0.011 %; length 0.000 % | 2026-09-28 |
| Karaoke Pitch Changer | 8 WAV exports, −6 to +6 keys; vocal reduce on a synthetic mix | Pitch 0.055 cents; length 0.000 %; centre vocal 37.5 dB quieter | 2026-09-27 |
| Vocal Pitch Changer | 10 WAV exports; formant correction on a synthetic vowel; a microphone recording | Pitch 0.041 cents; vowel shape error 7.6 dB off, 2.0 dB on (+4 semitones) | 2026-09-27 |
| Video Pitch Changer | 18 exported videos (MP4, MOV, WebM; 0.8× and 1.25× speed) | Picture identical in 18 of 18; pitch 0.080 cents; sound within one frame of the picture length | 2026-09-28 |
| YouTube Pitch Changer | Shifter on 50 test tones; 7 settings through Edge’s real tab sharing | Pitch under 0.001 cents (shifter), 0.81 cents (in the browser); sound about 0.2 s behind the picture | 2026-09-28 |
| Tool | What we measured | Result | Tested |
|---|---|---|---|
| Song Key Finder | 24 synthetic songs, 4 songs with drums, 3 real recordings, 2 hummed tunes | 24 of 24 keys correct; drum songs and hummed tunes correct; real recordings: see limits | 2026-09-27 |
| Vocal Range Test | 45 notes E2–C6, voice-like signals, octave traps, noise; a full microphone test | Worst 1.25 cents, 0 octave errors; G2–E5 range read within 3.5 cents | 2026-09-28 |
| Voice Type Test | 810 detector cases; voice-type rules against a separately typed table | All within 0.20 cents, 0 octave errors; all voice-type cases correct | 2026-09-28 |
| Semitone Calculator | All 576 key pairs, 11,664 note pairs, 873 frequencies | 0 mismatches; frequencies within 0.005 Hz | 2026-09-27 |
| Transpose Music Online | 41,515 chord-parsing checks; key guess on 96 progressions | 0 failures; 96 of 96 | 2026-09-27 |
| Music Key Chart | 2,072 chart values against textbook tables | All correct | 2026-09-27 |
| Major to Minor Converter | 11,244 theory checks; one −3 semitone export | 0 failures; 0.00 cents | 2026-09-27 |
The pitch changer on the homepage runs on an older build that has not been through these tests yet; it will be listed here once it has.
What we have not tested yet
- Real iPhones, Android phones and Safari. Phone layouts were checked in Edge’s phone emulation only.
- Firefox, except the Online Key Changer (Firefox 156: exports and key detection passed).
- Real microphones, voices and rooms. Noise and echo will make results less steady than our clean test signal.
- Listening tests: we measured pitch, length, clicks and level, not sound by ear.
Limits you will hear
- Full mixes move together. A pitch shift moves every instrument and voice. Karaoke vocal reduce removes only sound in the centre of a stereo mix, so real songs vary.
- Big shifts change the character. A voice shifted far up sounds thinner, far down heavier, because its resonances move. Formant correction helps; no tool can add detail the original file lacks.
- Key detection is an estimate. A test song that moves from C major to E major got 18 % confidence as a whole, while each half alone gave the right key. A real “Canon in D” recording read as D♭ major because it is tuned about a semitone low: the result follows the audio, not the title.
- Tempo can read double. In engine tests a 72 BPM drum loop was reported as 144 BPM; correct it by hand if needed.
- MP3 is lossy. Choose WAV if you will edit the file again.
- YouTube lag. The live sound is about 0.2 s behind the picture: about 0.16 s is the browser’s tab sharing, about 0.04 s our shifter. We aimed for 0.15 s and missed; the tool shows the real figure.
- Memory. Most audio tools take files up to 200 MB and 10 minutes, the video tool up to 500 MB and 15 minutes. Very old phones may still run out of memory.
If a tool will not start or a file will not open, see troubleshooting.
Why nothing is uploaded
The tool code makes no network request while it works; the tests of every audio, video and microphone tool above, except YouTube (which streams from YouTube by design), confirmed that no request left the site. The site itself loads Google Fonts, and Google Analytics only after you accept it in the cookie banner. Data security explains how each kind of tool handles your file.
Found a number that looks wrong?
Write to support@vocaltoollab.com with the tool’s address, your browser, device and file type. Mayer checks the report and corrects this page if a number is wrong. Our editorial guidelines describe how pages are checked and corrected.