A cut time becomes a sample boundary
Read the actual output bounds beside the requested seconds.
Time labels and audio frames
Digital audio is a sequence of samples at a sample rate. Stereo has two sample values per frame, one for each channel. WavePatch selects the same frame interval in both channels so the channels stay aligned.
A requested start rounds up to the next decoded sample boundary; the end rounds down and is exclusive. The audit gives the first frame, last exclusive frame and actual seconds. At 24,000 Hz, a one-second interval contains 24,000 frames per channel. The generated eight-second example is mono at that rate.
Why a one-millisecond field is not a lossless MP3 edit
The file is first decoded by the browser, which may resample it to the AudioContext rate. Selection operates on that decoded buffer. Export writes new 16-bit PCM WAV samples, not the original MP3 packets. Source encoding, metadata and exact original sample rate are not preserved as a lossless cut.
Keep enough context
For speech, inspect the sound around a word rather than choosing a number in isolation. The source waveform is a peak overview, not a transcript or a way to detect words automatically. Preview the selected region and test the downloaded file in the application that will use it.
Technical reference: MDN decodeAudioData.