Chapter 8 — Validator and Rebuild Workflow

Developing a speech synthesis system involves many interconnected components. In the Bishnupriya Manipuri Dictionary and Language Science Project, several tools work together to convert dictionary entries into spoken audio.

Because these components depend on one another, even a small inconsistency can cause the system to fail.

For this reason, the project includes a validator and rebuild workflow designed to detect errors, repair inconsistencies, and maintain synchronization between the dictionary, phonological conversion rules, and diphone audio files.

1. The Need for Validation

A diphone-based speech system depends on the availability of correctly labeled audio files.

If a required diphone is missing, the speech system cannot produce the correct pronunciation.

Common problems include:

• missing diphone audio files
• incorrect filename mappings
• inconsistent IPA conversion rules
• mismatched diphone generation algorithms

Without systematic validation, these issues can accumulate and make the speech system unreliable.

2. Diphone Validator Tool

To detect such problems, the project includes a diphone validator tool.

The validator analyzes the diphone sequence generated from a dictionary word and compares it with the diphone audio files available in the system.

The validator can report:

• missing diphone files
• extra or unused diphones
• incorrect filename formats
• coverage statistics

This information helps identify exactly which audio segments must be recorded or corrected.

3. Coverage Analysis

One of the most useful outputs of the validator is diphone coverage analysis.

Coverage measures the percentage of diphone transitions required by the dictionary that are already available as audio recordings.

For example:

Total diphones required: 520

Diphones recorded: 468

Coverage: 90%

Missing diphones: 52

Coverage analysis helps prioritize which diphones must be recorded next.

4. Synchronization Problems

During development, a major challenge was ensuring that all components of the system used the same conversion rules.

Several pages within the system performed similar tasks, including:

• IPA conversion
• phoneme extraction
• diphone generation
• safe filename mapping

If these components used slightly different rules, the diphone sequences generated on one page could differ from those generated on another page.

Such inconsistencies often produced missing diphone errors even when the audio files existed.

5. Unifying Conversion Rules

To resolve synchronization problems, the project introduced a unified conversion module.

This module performs several tasks:

• BPM orthography to IPA conversion
• IPA to phoneme tokenization
• diphone generation
• safe filename mapping

All pages in the system now rely on this shared module.

This ensures that every component generates identical diphone sequences for the same word.

6. Rebuild Workflow

When diphone recordings are updated or conversion rules change, the diphone system must be rebuilt.

The rebuild workflow typically follows these steps:

1. Update dictionary entries
2. Generate IPA pronunciation
3. Extract phoneme sequences
4. Generate diphone sequences
5. Compare diphones with audio files
6. Identify missing diphones
7. Record or generate missing segments
8. Re-run validation
9. Deploy updated diphone inventory

This structured process ensures that the speech system remains consistent and reliable.

7. Automating the Workflow

To simplify maintenance, several automation tools were developed for the project.

These tools can:

• analyze dictionary entries in batch mode
• generate diphone inventories automatically
• detect missing audio files
• produce reports for recording sessions

Automation greatly reduces the manual effort required to maintain the speech system.

8. Importance for Future Development

The validator and rebuild workflow is essential for maintaining a sustainable speech system.

Without such tools, the system could easily become inconsistent as new words and recordings are added.

By integrating validation and rebuild procedures into the development process, the project ensures that the Bishnupriya Manipuri speech system remains scalable and maintainable.