How our audio pitch changer works

Pitch Practice helps you move an existing recording into a more comfortable key, slow a passage down, and keep a useful practice copy.

What happens to a file

1. Your browser reads the selected file and decodes it into sample data. No audio upload is used.
2. Practice mode uses centred time-domain stretching with source-time anchoring and bandlimited resampling in a worker. Music mode uses Signalsmith Stretch with staged time stretching, bandlimited pitch resampling and source-guided stereo processing. If its WebAssembly engine cannot initialize, a local SoundTouchJS compatibility engine is used. Both run locally. Semitones and cents control pitch; speed controls duration independently. Compare both modes on your recording: neither algorithm is transparent on every voice or mix.
3. The changed track is rendered in browser memory. Playback and export use that same rendered audio. If processing creates peaks above the output range, the entire result is attenuated to avoid clipping.
4. WAV export writes your choice of 16-bit or 24-bit PCM with the decoded sample rate and channel count. MP3 export uses lamejs at your chosen 128–320 kbps in a browser worker.

What stays the same—and what can change

At 100% speed, the output has the same number of decoded samples as the input. A speed of 75% creates a track roughly one third longer. Mono stays mono and stereo stays stereo. Browser decoding can resample compressed files; standard PCM WAV sample rates from 8–96 kHz are read from the file header when available. MP3 export resamples unsupported encoder rates to 44.1 kHz.

WAV is uncompressed. Choose 24-bit to retain more of the rendered audio’s numerical precision, especially in quiet passages, or 16-bit for a smaller file. The source is decoded and processed before export; choosing 24-bit does not recover missing detail or preserve an original floating-point format. MP3 is lossy. Downloads include delay and padding metadata for compatible players; other players may still expose encoder padding. For loud exports, the encoder lowers the MP3 level to reserve 2 dB of sample-peak headroom; playback and WAV levels stay as rendered. This reduces encoding overshoots but is not a universal true-peak guarantee. Neither format restores detail already lost in the source.

Listen before you keep it

Large shifts can change a voice's character or blur transients. Dense mixes, percussion and extreme shifts can expose artifacts. Compare small changes against the original and choose the version that works for your practice.

Deliberately limited

The current tool accepts one mono/stereo file at a time, up to 500 MB and 60 minutes, with a further decoded-memory limit. It does not identify musical keys, extract melodies, separate stems, tune individual notes or return edited video. File support depends on the codecs your browser can decode.