How malomn/dsit works
Main Components
A. Input Layer:
• Audio Input (.wav files)
• Transcription Input (.lbl/.seg files)
B. Core Processing Components:
• Preprocessing Module (preprocessing.py)
• CNN Model (models/cnn)
• Intelligibility Model (models/intelligibility)
• Severity Model (models/severity)
• ANPS Analysis Module (anps.py)
C. Data Storage:
• Cache System (cache.py)
• Static Data Files (data/*.json)
D. Output Generation:
• JSON Output Generator
Component Relationships
[Input Layer]
• ↓
[Preprocessing]
• ↓
[Model Processing Layer]
• → [CNN Model]
• → [Intelligibility Model]
• → [Severity Model]
• ↓
[Analysis Layer]
• → [ANPS Analysis]
• → [Confusion Matrix Generation]
• ↓
[Output Generation]
Important Details
A. Data Flow Annotations:
• Audio signal flow
• Feature extraction paths
• Model inference paths
• Score generation paths
B. Key Components:
• Data storage showing the various JSON configuration files
• Processing pipeline showing the transformation of raw audio to final scores
Specific Implementation Notes
• How the cache system interacts with the processing pipeline
• Where the phonetic trait analysis occurs
• Confusion matrix generation process
• Relationship between the various .json data files and the processing components
Output Format Details
• Intelligibility scores
• Severity scores
• Confusion matrix
• ANPS scores
• ANPS: Automatic Neuronal Phonetic Score
• CNN: Convolutional Neural Network
• DSIT: Deep Speech Intelligibility Toolkit
This system design should emphasize the pipeline nature of the project while clearly showing how each component contributes to the final speech analysis output.