Implement TDNN Model for Javanese Speech Recognition

[Menjadianjay]

Requirements:

• The TDNN model must be implemented for consistent and reliable evaluation

• Accept as input the processed sentences/words from:

  • lexicon + HMM pipeline
  • n-gram language model

• Support feature inputs (e.g., MFCC / filterbank)

• Must be compatible with the current ASR pipeline

• Provide training and inference capability

• Evaluation must include:

  • Word Error Rate (WER)
  • Accuracy

Notes:

• The model will be used for Javanese speech recognition
• Focus on improving temporal modeling compared to baseline methods
• Ensure modular integration with existing system