HERTZKIT / FROM A FILE TO A DECISION

Prepare your audio for useful results

A six-minute track can help measure loudness and still be a poor reference for practicing a phrase. Choose the file and excerpt around the question you want to answer.

Reviewed October 27, 2026 · With an audio example

On this page9 sections

AN EXAMPLE TO LISTEN TO

Remove the wait before the music starts

If your mono or stereo PCM WAV exceeds the limit, open it in the Trimmer and load just the seconds you want to measure. Export that section and open it in the Analyzer.

This piano starts after one second of silence. Listen to the original and then the trim: the notes stay the same; the beginning and length change.

Waveform of the ten-second example, marking the excerpt from 0.9 to 9.5 seconds.
The marked area is the excerpt we keep. A margin protects the attack and the ending.

Original · 10 seconds

Trim · 8.6 seconds

To do it yourself: open the trimmer, choose “Try an example”, then set Start to 0.9 and End to 9.5. Listen to the trim before downloading.

Try this trim →

VCSL piano · CC0. Arrangement and figure by HertzKit.

1. Write down what you need to check

Before loading a song, finish this sentence: “I want to know…”. Its tempo? Whether the chorus disappears in mono? Which notes a melody plays? A specific question helps keep a technical reading separate from a judgment about the music.

Keep the original and use a copy. Name it by version and purpose: song-v03-chorus.wav. When comparing two mixes, export matching beginnings and endings. A difference in volume or passage can make you choose for the wrong reason.

2. Choose how much audio you need

  • Tempo: a passage with drums or repeating attacks is usually more useful than a beatless introduction. The BPM detector accepts 5 seconds to 10 minutes. Check its suggestion with a metronome; try half-time and double-time too.
  • Mix: use the complete song for integrated loudness in Mix review. Then listen to the flagged regions. A chorus-only LUFS reading and a whole-song reading are not interchangeable.
  • A/B: load two versions of the same passage into the comparison tool. Loudness matching does not fix timing offsets between recordings. Change one thing at a time and write down what you hear.
  • Notes to MIDI: start with 10–30 seconds of an isolated voice or instrument in the transcriber. This short excerpt makes note review easier; it is not the input maximum. Listen alongside the original and correct the notes in your editor.
  • Practice: you can load up to 10 minutes, then select a 0.2–20 second phrase in HertzKit Practice. It practices one note at a time; loading a song does not turn the complete song into an assessment.
  • Separation: choose a short excerpt in the separator to check the result; it counts as one start within the hourly limit. Listen for artifacts. Separation does not establish authorship or grant permission to publish an extracted voice.

3. Minutes, megabytes and memory are different

HertzKit audio input accepts up to 10 minutes and 128 MB per file, mono or stereo. Both limits apply. MIDI and JSON have their own limits. The editor supports a joined result up to 10 minutes; the speed changer also limits output to 10 minutes, so an eight-minute track cannot be exported at half speed.

An MP3’s file size does not tell you how much memory analysis needs. Six minutes of stereo at 44,100 samples per second, represented as 32-bit numbers, requires about 121 MiB of decoded audio: 360 × 44,100 × 2 × 4 bytes. Calculations may require copies and additional memory. A small MP3 can become a large job.

On a phone, close other tabs and start with the passage you need. If the browser closes the page or cannot finish, trim a copy in your editor. Increasing an input limit does not increase physical device memory. Separation and transcription also need a connection to load their model initially.

4. Choose a format without inventing quality

If you have the recording project, export a mono or stereo WAV copy at its original sample rate. Avoid upsampling just for analysis. Converting MP3 to WAV, 16 to 24 bits or 44.1 to 96 kHz does not restore lost information.

For BPM, an existing MP3 may be sufficient. For small mix changes, prefer equivalent exports from the project. A format that works on a computer may not decode in another browser; try PCM WAV or MP3 if it fails. The converter exports WAV or MP3 and must also be able to decode its input.

A ten-minute, stereo, 24-bit WAV at 44.1 kHz occupies about 151 MiB and exceeds the file limit even though its duration is allowed. Trim the passage or export an appropriate working copy. Keep your original master.

5. Read progress by stage

Check duration → read → decode → calculate → result ready. Reading shows bytes read; measurable calculations show their own progress. Decoding uses a waiting indicator because the browser does not provide a reliable percentage. A stage reaching 100% does not mean all stages are complete.

When a result appears, review its units and limitations. “No clear pulse” does not mean the song is bad: it may contain rubato, few attacks or several rhythmic patterns. “Could not decode” describes format reading, not artistic quality.

For export, wait until the file is ready and request its download. Then check your browser’s downloads list and open the file. Check duration, beginning, ending, channels and unexpected silence. The page can hand a file to the browser but cannot confirm it was saved in your folder.

A ten-minute exercise

  1. Choose 30 seconds of your music with a clear pulse and estimate BPM. Try the metronome over the audio; note whether the reading, half-time or double-time fits.
  2. Export two versions of the same passage with one EQ difference. Compare at matched loudness. Describe the change before calling it better.
  3. Download a result and reopen it. If the reading does not change a recording, mixing or practice decision, rethink your question.

Before sharing

Use your own recordings or material you are authorized to process and publish. Separation, transcription and conversion do not remove rights in the composition, performance or recording. Agree on the use of other people’s material before publishing. Measurements do not certify rights or guarantee distributor acceptance.

Technical references

MDN: decoding and audio context sample rate · MDN: scope of the download attribute. Memory examples use duration × sample rate × channels × bytes per sample; they estimate audio storage, not total page memory.