Morse Code Decoder

Decoder or translator: which one you need

Two different jobs, and most sites offer only the second one under both names.

• This decoder reads sound. Beeps from a file, a video played next to your phone, a radio, or someone tapping a key. It works out the tone frequency in hertz, the sending speed in words per minute and the text, and none of that is written down anywhere beforehand.
• The Morse code translator reads written dots and dashes. If you already have ... --- ... as text, paste it there and it comes back as plain text instantly. It also converts text into Morse code, plays it as audio beeps and flashes it as light.

So: heard it, use this page. Read it somewhere, use the translator. If what you have is a blinking light rather than a sound, the light decoder reads that through the camera, and a photo of dots and dashes goes into the image decoder.

What the decoder actually does

Four steps, and each one can go wrong on its own.

1. Find the tone. Morse on the air, what radio amateurs call CW, is a single steady pitch, usually between 400 and 900 Hz. The decoder scans that range and keeps the tone frequency that carries the most energy across the whole recording.
2. Measure that tone over time. Every five milliseconds it asks one question: how strong is this one frequency right now? That gives a curve which rises for every mark and falls in every gap. The picture above the expert settings is exactly that curve.
3. Decide where the tone is on. The threshold is not fixed. It is placed between the quiet and the loud part of the recording, so a faint tape and a loud one are read the same way.
4. Read the timing. The shortest marks are dots, three times that is a dash, and the gaps of one, three and seven units separate symbols, letters and words. From the length of one unit the decoder also states the sending speed in WPM: 60 milliseconds per unit is 20 words per minute, 120 milliseconds is 10.

The measurement in step two uses the Goertzel algorithm. It costs two multiplications per sample for a single frequency, where a full spectrum would cost far more and then throw almost all of it away. For a signal that consists of exactly one tone, that is the right tool.

Where it fails, and why

Machine-generated Morse decodes almost perfectly. Everything else depends on how the signal was made.

• Uneven hand sending. If the gap between two letters is shorter than the dash next to it, no decoder on earth can tell where one letter ended. A human operator uses the language to guess; this one does not guess.
• Two signals at once. The decoder locks onto one pitch. If another station sits within a few dozen hertz, both end up in the same measurement and the result is nonsense. The expert settings let you name the pitch yourself and shut the other one out.
• Fading and heavy noise. A signal that comes and goes takes the threshold with it. Short recordings of a strong passage work better than long ones full of holes.
• A recording of only dashes. This one is not a flaw but a real ambiguity: a string of dashes with letter gaps looks exactly like a string of dots at a third of the speed. Without more to go on, the decoder reads dots.

Getting a clean reading

• Feed it ten to twenty seconds. Long enough to see dots, dashes and word gaps; short enough that the speed does not drift.
• Hold the phone close to the speaker rather than turning the volume up. Room echo smears the edges of every mark.
• Watch the signal picture. If it shows a solid block instead of separate bars, the tone was found in the wrong place. Enter the pitch by hand in the expert settings.
• Only touch the threshold if the picture looks right and the text does not. Raise it when noise is being read as tone, lower it when quiet marks are being dropped.

The microphone path uses the raw signal: echo cancellation, noise suppression and automatic gain are all switched off. Those three are built for speech, and every one of them treats a steady tone as something to remove.