Audio Spectrogram Generator

• Choose Audio file, then select Generate spectrogram.
• Inspect the generated grayscale preview; low frequencies are at the bottom and brightness uses the disclosed fixed −100 to 0 dB range.
• Download result to save spectrogram.zip, containing spectrogram.png and spectrogram.json. Cancel stops the current worker; Reset clears the file and result.

A coherent 750 Hz sine at 48000 Hz falls at bin 32 of the 2048-point frequency grid; an opposite-phase stereo pair cancels when averaged.

• One nonempty WAV, MP3 or FLAC up to 64 MiB, at most 60 seconds, 1–2 channels and source rates 8–96 kHz. Format support still depends on this browser decoding the primary track.
• Uses the primary packet timeline; MP3 encoder delay/padding may remain. No gapless-trimming promise. Decode goes directly to 48 kHz Float32 analysis without a PCM16 intermediate.
• Fixed periodic 2048 Hann window, 512-frame hop, minimal complete tail coverage and zero padding only when needed. At most 5622 windows and 2048 time columns; every window belongs to one column and linear powers average before logarithm.
• Normalization is coherent-gain peak-amplitude-squared dBFS: factors 1 at DC/Nyquist and 4 at interior bins. It is not PSD, RMS power, SPL, LUFS or a pitch/BPM/music/medical inference. Display saturates outside −100 to 0 dB.
• Existing admission caps of 16 million combined source/output samples and 200 million resampler products can require shorter high-rate stereo clips. At most 2099200 pixels, 256 KiB JSON, 20 MiB ZIP and a 120-second worker deadline.