How sagnik-barman/morse-cw-decoder works
This is a deep-learning Morse CW decoder, with a classical DSP decoder for comparison. Its main workflow trains acoustic models on procedurally generated, impaired audio, then transcribes recordings into text; evaluation compares model and baseline errors on synthetic and real audio. Real-data tooling also aligns beacon recordings to a schedule and produces labelled clips with SNR estimates. The graph preserves the documented training, decoding, language-model, and benchmark branches. Source excerpts and README statements support the shown links; unsampled implementations are represented only where documented.
Main entry points: infer.py Inference Helpers · benchmark_real.py Real-audio Benchmark · evaluate.py Synthetic Evaluation · decode_wav.py WAV Transcription
Components 21
Synthetic Training
- Label Text
morse/text_gen.py - Audio Synthesis
morse/synth.py - Training Data
morse/data.py - Model Training
scripts/train.py
Transcription
- CTC Acoustic Model
morse/model.py - Inference Helpers
morse/infer.py - CTC Decoding
morse/decode.py - DSP Baseline
morse/baseline.py - WAV Transcription
scripts/decode_wav.py - Morse Vocabulary
morse/alphabet.py
Real Data
- Beacon Schedule & SNR
morse/beacons.py - Beacon Recording
scripts/record_beacons.py - Beacon Clip Labelling
scripts/slice_beacons.py - Real-data Loader
morse/realdata.py
Evaluation
- CER / WER Metrics
morse/metrics.py - Synthetic Evaluation
scripts/evaluate.py - Real-audio Benchmark
scripts/benchmark_real.py - Character LM Builder
scripts/build_lm.py
Other components
- Operator
- Audio Recording
- Decoded Text
Connections 29
- Operator → Model Training: trains
README.md - Label Text → Audio Synthesis: supplies labels
morse/synth.py - Audio Synthesis → Training Data: generates clips
morse/data.py - Training Data → Model Training: feeds batches
scripts/train.py - Model Training → CTC Acoustic Model: trains
scripts/train.py - Operator → Audio Recording: provides
README.md - Audio Recording → WAV Transcription: transcribes
README.md - WAV Transcription → Inference Helpers: uses
scripts/decode_wav.py - Inference Helpers → CTC Acoustic Model: runs inference
morse/infer.py - Inference Helpers → CTC Decoding: decodes output
morse/infer.py - Inference Helpers → DSP Baseline: offers baseline
morse/infer.py - CTC Decoding → Morse Vocabulary: maps symbols
morse/decode.py - WAV Transcription → Decoded Text: returns transcript
README.md - Operator → Beacon Recording: records beacon
README.md - Beacon Recording → Beacon Schedule & SNR: uses schedule
scripts/record_beacons.py - Beacon Recording → Beacon Clip Labelling: records and slices
README.md - Beacon Clip Labelling → Beacon Schedule & SNR: aligns and measures
scripts/slice_beacons.py - Beacon Clip Labelling → Real-data Loader: writes labelled clips
README.md - Operator → Synthetic Evaluation: evaluates
README.md - Synthetic Evaluation → Training Data: uses synthetic sets
scripts/evaluate.py - Synthetic Evaluation → Inference Helpers: runs systems
scripts/evaluate.py - Synthetic Evaluation → CER / WER Metrics: computes errors
scripts/evaluate.py - Operator → Real-audio Benchmark: benchmarks recordings
README.md - Real-audio Benchmark → Real-data Loader: loads real sets
scripts/benchmark_real.py - Real-audio Benchmark → Inference Helpers: transcribes clips
scripts/benchmark_real.py - Real-audio Benchmark → DSP Baseline: compares baseline
scripts/benchmark_real.py - Real-audio Benchmark → CER / WER Metrics: scores outputs
scripts/benchmark_real.py - Operator → Character LM Builder: builds language model
README.md - Character LM Builder → CTC Decoding: builds LM for
scripts/build_lm.py