First practice: the example
Press “Start with the example” in step 1. Listen to the melody in step 2; each note button also plays an individual note. Move down or up an octave if the range is uncomfortable. When ready, press “Continue to recording”. In step 3, put on headphones and press “Record first take”. Allow the microphone, wait for 3, 2, 1, and start when “Recording” appears. The take ends automatically and results appear. Continue with the journey without reloading the melody.
What you can practice
A short melody for voice or an instrument playing one note at a time, without accompaniment. Complete songs are not analyzed and vocals are not separated. The example is original HertzKit material. For your files or recordings, confirm that you are authorized to use them.
Your audio creates the reference
In “Melody”, open “Use my melody” and choose “My audio”. If you already have a reference, use “Change melody” to return to that selection.
- Confirm permission and open the file. Tick the box next to the selector. Audio can be up to 10 minutes and 128 MB, mono or stereo; your browser must be able to decode it.
- Choose a phrase. The initial selection covers up to eight seconds. Drag Start and End on the waveform or enter the times: selections can be 0.2–20 seconds. For longer files, change the scale and browse the waveform to find the passage.
- Check the tempo. Automatic estimation analyzes sound changes in a 5–20 second excerpt. Solo vocals or free timing may not have a steady beat. If it is missing or does not match, enter the BPM or open “Tap the beat manually” and tap 4–8 beats while listening. A failed estimate keeps your entered BPM. Once entered manually, changing the selection does not replace it. At faster tempos, you may need a shorter phrase to stay within 32 beats.
- Listen and analyze. Play, pause or stop the selection. When it is the phrase you want, press “Analyze this phrase”.
- Review the reference. In Preparation, listen to the estimated notes and your original audio. If they differ, open “Edit notes or choose a phrase” and correct the reference before practicing.
The handles also move with the arrow keys; Shift changes the step from 0.1 to 1 second. The selection does not modify your original file.
To create a reference with the microphone, choose “Record phrase”, confirm permission and BPM, then record up to 20 seconds without accompaniment. Press “Finish phrase” to analyze it. This is not a practice take yet: first review the detected melody in Preparation.
Import MIDI
When you download your exercise as MIDI, guitar and piano include the matching program for General MIDI players. MIDI contains notes and timing, not recorded samples. Piano and guitar and Continuous tone export as piano: choose the instrument in your DAW if you need a different sound.
MIDI is a note sequence, not an audio recording. It has its own importer and a 1 MB limit. A compatible melody track is selected; incompatible tracks are omitted, not notes within an accepted track. Chords, pitch bend, sustain and format 2 are not supported. Leading silence is removed and constant tempo is used; original instruments, expression and tempo changes are not all retained. If there are multiple compatible tracks, choose another in step 2.
Adjust the reference
Slowing down makes practice longer. You can keep those takes as WAV. To take them into Sessions as .hksession, the practiced phrase must last no more than 20 seconds; choose a shorter fragment before starting if it becomes longer.
The recreated reference uses the selected instrument; original audio has separate playback. In Practice, bars show notes and timing, and a line follows playback. After recording, your latest take is available beside the controls. These bars are not a measurement of your voice or instrument.
Up to 64 notes from C2 to C6 within 32 beats, with a maximum of 8 beats per note. One beat is a quarter note; middle C is C4. Start times and durations must not overlap. Conversion preserves detected note starts, durations and rests, and removes leading silence. Review the result before practicing. An overlong or out-of-range phrase is rejected with a message, not silently shortened. Choose classical guitar, grand piano, piano and guitar, or a continuous tone for the reference, with adjustable volume. A4 defaults to 440 Hz, adjustable from 400 to 480 Hz. Rhythm variations use editable rules, not generative AI.
Record with a guide or from memory
The audio guide starts enabled: it plays the melody while you sing or play, so you do not need to memorize it. Use headphones; with speakers the microphone may measure the guide and confuse the result. Disable the guide to practice from memory and keep or remove the click.
During the take you see the reference note, progress and microphone level. The note on screen is what you should perform, not live detection of what you are singing.
In step 3, open “Guide, click and microphone”, then “Microphone and level” to change the input. It is also available when creating a reference with “Record phrase”. “Automatic (system)” uses your device’s default input; choose another or refresh the list after connecting it. Names appear after permission and the active input is reported. “Test microphone” checks incoming signal for five seconds and releases the input. Choose the input before recording: the selector locks during a take. “Cancel” discards an incomplete take.
Understand the result
“Higher” and “lower” describe the estimated note center relative to the reference. Distance is measured in cents; 100 cents is one semitone. “No stable reading” means insufficient or unstable signal and does not count as your mistake. If only part of the phrase is measured, it is explicitly not a complete evaluation. Typical deviation is the median of the absolute note-center offsets from their targets, not an artistic grade. Percentages describe analyzable signal frames, not correct notes. Your take has its own player and is separate from the original audio.
Optional timing comparison
Enable onset comparison in advanced options. Positive values can indicate late entry and negative values early entry, but device delay also affects them. Use “Measure acoustic delay · 5 s” in advanced options with the microphone near the speakers: this measures an acoustic round trip, not a professional calibration. If it cannot obtain three consistent readings, it keeps the previous value. You can also enter compensation measured separately. Return to headphones for practice and confirm the reference again after changing the setting. Repeated notes without a clear attack may have no onset reading. If notes are too short, practice more slowly.
How timing is calculated
Timing comparison is approximate: it does not guarantee precise measurement of small timing differences.
What to do after a take
The journey keeps your first take in this tab. Review the selected fragment and press “Practice this fragment”. Open “Adjust fragment or speed” if you want to change notes or practice more slowly. “Return to the full phrase” restores exactly the notes, speed, guide, click and timing options from your first take. Record the full phrase again to compare.
A suggested fragment appears only when every note has a stable reading, capture level stayed stable, there is no clipping warning or large register discrepancy, and at least one note has 25 cents of deviation. It includes the preceding note when possible. This is a starting point for listening, not a diagnosis or a universal tuning tolerance. Without suitable readings, choose by ear or repeat the first take with a clearer signal. Missing readings are not performance mistakes.
Listen to the first and final recordings. A numerical difference requires matching reference settings, capture input, sample rate and level, stable readings for every note, no clipping or register warning, and your confirmation of the same instrument or voice, position and conditions. The result compares median absolute pitch deviation; it does not grade musical quality or prove that the exercise caused a change.
After finishing the practice journey, listen to the first and final recordings and press “Download this practice summary”. This TXT keeps the before and after. Either full take can also be downloaded as WAV. “Download this take’s details · TXT” in the collapsed details contains only the latest recording. An optional observation stays in page memory and is included in the TXT summary or .hksession activity you download. Recordings are not uploaded or automatically saved. Changing the reference or its options starts a new comparison. Starting a new comparison or leaving the page removes those temporary journey recordings; downloads remain.
To continue the same practice across days, download the journey’s .hksession and open it in Sessions. It already includes the first and final takes in one round. Start another round for the next practice and choose an “Another day” take. These imported recordings are compared by listening: Sessions does not verify their input settings. Practice history still stores summaries without audio; it does not replace this download.
Share a useful report with your teacher
After a take, open “Latest recording and measurements”, then “Save or share a visual report” to see the preview. This report describes one take; it does not replace the final journey summary. Add a title and observation, download PNG or PDF, or share the PNG through your device when the browser supports it. PDF uses a light background for printing and splits long phrases into pages; no notes are omitted.
The report identifies each note, estimated deviation in cents, onset in milliseconds when enabled and the proportion of analyzable signal. Missing values and “No stable reading” are not your mistakes. An incomplete take receives no full-phrase pitch summary.
Report comparison uses an earlier complete take in this session with the same phrase, tempo, tuning, compensation, guide, click and audio input. If input gain changed during capture, there is a clipping warning or the browser cannot identify the input, that comparison is unavailable. You must also confirm the same microphone, position, instrument and conditions. This does not certify identical acoustics or artistic progress; the existing history retains the scope explained in this guide.
Personal name, original file name and audio are excluded by default. Any title and observation you enter are included: review them before sending. The report is not published or uploaded to HertzKit. Downloaded or shared files are controlled by you and your chosen app; clearing site history does not delete those copies. Sharing requires no HertzKit account and is optional.
Saving and privacy
Practice does not require an account. Its exercises are saved in this browser, not in My account. Notes, settings and summaries do not include recordings. To continue on another device, export the library as JSON and import it there; download audio separately if you want to keep it.
Under “My exercises · save or resume”, open “Continue this exercise another day” to enable saving for the current exercise. It keeps notes, selected phrase, tempo, tuning, practice options, compensation and up to 20 summaries. Disabling it removes that copy; “Delete current exercise and history” also clears session history.
When restoring your own melody, confirm the permission shown in Preparation before continuing. Opening a library copy respects your existing automatic-saving choice; it does not enable saving on its own.
If the browser blocks storage, you can still practice and download your files. Takes stay temporarily in the page and are lost when you leave; download them first if needed. Analysis does not upload your files to a server.
Keep several exercises and see your progress
Open “My exercises · save or resume”, then “My saved exercises”; save up to 20 named copies with notes, settings and history, without audio. Later changes do not modify saved copies. Export a JSON backup to keep them or move them to another device; importing retains existing exercises and rejects conflicting copies. The chart only shows fully measured takes with the same phrase and settings: fewer cents means less deviation from the reference, not higher artistic quality. Download your latest take as WAV before leaving the page. Deleting the current automatic save does not delete your named exercises.
Adjust the microphone level
The “Microphone level in HertzKit” control adjusts the signal inside the page: 100% preserves incoming amplitude, 75% reduces it, and 50% is approximately −6 dB. Mute with 0% or increase up to 200% (+6 dB). Start at 100%, test the loudest passage and leave headroom below 0 dBFS. The first meter shows the signal received from the device; the second shows the adjusted signal. If the first reaches full scale or you hear source distortion, lower the system or interface gain; reducing the web control cannot repair already clipped audio. Increasing gain also increases noise: it does not make a noisy signal more accurate. The setting affects subsequent recording or analysis, not uploaded files or earlier takes. It lasts only while this page remains open.
Measurement limits
Chords, noise, multiple instruments, wide vibrato, distortion and fast attacks can confuse detection. This is not an automatic teacher or certified transcription. Audio tempo is an editable suggestion; meter is not detected automatically. To record, allow microphone access in a compatible browser. Switching apps, locking the phone or losing the microphone cancels an incomplete take. Rights to the material remain the user’s responsibility.