Morse Decoder AI 1.1

The Morse Decoder AI is a specialized application designed for the interpretation of Morse code signals through artificial intelligence technology. Primarily intended for amateur radio operators and educational use, the application facilitates automatic decoding of audio signals into text by listening via microphone or line input. It is suitable for both casual and more experienced users seeking to enhance their decoding capabilities.

The underlying neural network has been trained to decode signals transmitted at speeds ranging from 10 to 40 words per minute within a frequency span of 200 Hz to 900 Hz. This ensures reliable performance across a variety of typical Morse code transmission scenarios.

The software offers two operational modes: the default direct mode and a tone filtering mode. In direct mode, the neural network attempts to decode Morse signals within the 250 Hz to 900 Hz range, functioning effectively when signals are received with reasonable clarity and strength, typically corresponding to signal levels of 7 dB to 9 dB on the S-meter.

The tone filtering mode is designed to facilitate decoding amid challenging conditions such as weak or noisy signals. It employs band-pass filtering prior to neural network processing, which enhances the detection of faint signals affected by radio interference. This mode necessitates precise knowledge of the CW tone frequency, which can be set according to individual transceiver configurations—using tools such as YAESU's ZIN/SPOT button or ICOM's AUTOTUNE function. Available filter bandwidth options include 25Hz, 50Hz, and 150Hz; notably, the 25Hz filter may enable successful decoding of very weak signals when the exact CW tone frequency is known.

The application provides two neural network configurations—Network A and Network B—accessible via the user interface. Network A is recommended for use with stable signals characterized by consistent dot and dash durations, while Network B accommodates transmission with variable pulse lengths, such as those produced by straight keys. Users can switch between these options in real time to evaluate decoding performance under differing signal conditions.

An integrated sound level indicator assists operators in maintaining optimal input levels, suggesting a target around -7 dB for effective decoding. Proper signal amplitude ensures that audio levels are neither too faint nor excessively strong. It is important to note that higher frequency components tend to produce quieter audio signals compared to lower frequencies.

The application also offers customization options through various color themes, allowing users to tailor the interface for personal comfort and improved usability during extended operation.