How windywhu/speech-digit-recognition works
This is a Python desktop application for recognizing Mandarin digits zero through nine, distributed with pretrained CNN and RBF-SVM models. The main user flow is WAV or microphone input into the desktop window, model-specific preprocessing and inference, then displayed predictions, charts, playback, and exportable history. A separate research workflow prepares speaker-isolated datasets, trains both model families, and evaluates split, robustness, and external-speaker performance. The GUI excerpt is partial, so its detailed chart and history internals are represented at the documented capability level.
Main entry points: 语音识别窗口.py Recognition window · cnn_best_model.pt Saved CNN model · best_model.joblib Saved SVM model · data Local WAV dataset
Components 18
Desktop recognition
- Recognition window
speech-digit-recognition/语音识别窗口.py - Audio input and preprocessing
speech-digit-recognition/语音识别窗口.py - Charts and history
speech-digit-recognition/语音识别窗口.py
Recognition models
- CNN probabilities
speech-digit-recognition/CNN方法/predict_probs.py - RBF-SVM prediction
speech-digit-recognition/传统方法/run_experiment.py - CNN training
speech-digit-recognition/CNN方法/train_cnn.py - SVM training
speech-digit-recognition/传统方法/run_experiment.py - Saved CNN model
speech-digit-recognition/CNN方法/outputs/cnn_best_model.pt - Saved SVM model
speech-digit-recognition/传统方法/outputs/best_model.joblib
Training and evaluation
- CNN split evaluation
speech-digit-recognition/CNN方法/evaluate_splits.py - SVM split evaluation
speech-digit-recognition/传统方法/evaluate_splits.py - Preprocessing analysis
speech-digit-recognition/传统方法/preprocess_stats.py - Robustness evaluation
speech-digit-recognition/evaluate_robustness.py - External-speaker evaluation
speech-digit-recognition/evaluate_external_speakers.py
Dataset preparation
- Speaker-isolated split
speech-digit-recognition/prepare_dataset_split.py - Local WAV dataset
speech-digit-recognition/data
Other components
- User
- WAV / microphone
Connections 21
- User → Recognition window: uses
README.md - WAV / microphone → Recognition window: provides audio
README.md - Recognition window → Audio input and preprocessing: loads and records
README.md - Recognition window → CNN probabilities: requests CNN inference
speech-digit-recognition/语音识别窗口.py - Recognition window → RBF-SVM prediction: requests SVM inference
speech-digit-recognition/语音识别窗口.py - Recognition window → Charts and history: presents results
README.md - CNN probabilities → Saved CNN model: loads weights
speech-digit-recognition/CNN方法/predict_probs.py - RBF-SVM prediction → Saved SVM model: loads model
- Local WAV dataset → Speaker-isolated split: supplies recordings
README.md - Speaker-isolated split → CNN training: provides split
README.md - Speaker-isolated split → SVM training: provides split
README.md - CNN training → Saved CNN model: writes checkpoint
README.md - SVM training → Saved SVM model: writes model
README.md - CNN training → CNN split evaluation: provides checkpoint and cache
README.md - SVM training → SVM split evaluation: provides model and features
README.md - Local WAV dataset → Preprocessing analysis: supplies WAV files
speech-digit-recognition/传统方法/preprocess_stats.py - Local WAV dataset → Robustness evaluation: supplies held-out audio
speech-digit-recognition/evaluate_robustness.py - Saved CNN model → Robustness evaluation: evaluated by
speech-digit-recognition/evaluate_robustness.py - Saved SVM model → Robustness evaluation: evaluated by
speech-digit-recognition/evaluate_robustness.py - Saved CNN model → External-speaker evaluation: evaluated by
speech-digit-recognition/evaluate_external_speakers.py - Saved SVM model → External-speaker evaluation: evaluated by
speech-digit-recognition/evaluate_external_speakers.py