Chapter 7 — Designing the Bishnupriya Manipuri Diphone System

The creation of a diphone-based speech synthesis system requires a careful understanding of the phonological structure of a language. In the Bishnupriya Manipuri Dictionary and Language Science Project, the diphone system forms the core mechanism that enables automatic pronunciation playback.

Rather than recording every possible word in the language, the diphone approach records transitions between phonemes. These transitions can then be combined to synthesize the pronunciation of many different words.

1. Phoneme Inventory of the Language

The first step in designing a diphone system is identifying the phoneme inventory of the language.

Phonemes are the minimal sound units that distinguish meaning between words.

Based on phonological analysis, Bishnupriya Manipuri includes several categories of phonemes:

• vowels
• long vowels
• nasal vowels
• stops
• fricatives
• nasals
• approximants

These phonemes form the basic building blocks from which diphones are constructed.

2. What is a Diphone?

A diphone represents the transition between two adjacent phonemes.

Speech sounds are not isolated units. When a speaker moves from one phoneme to another, the acoustic signal changes gradually.

Diphones capture these transitions.

For example:

Phoneme sequence:
d – i – ʃ – aː

Diphone sequence:
#-d
d-i
i-ʃ
ʃ-aː
aː-#

The symbol "#" represents the beginning or end of a word.

By recording these transitions, a speech synthesis system can reconstruct the pronunciation of many words.

3. Determining the Required Diphones

A key question in building a diphone system is determining how many diphones are required to cover the phonological structure of the language.

In theory, if a language contains N phonemes, the number of possible diphones is approximately:

N × N

However, many of these combinations do not occur in actual words.

Therefore, the diphone inventory is usually constructed by analyzing real lexical data from a dictionary or corpus.

In the present project, dictionary entries serve as the primary source for generating diphone combinations.

4. Extracting Diphones from Dictionary Words

After dictionary words are converted into IPA, the phoneme sequence of each word can be analyzed.

From this sequence, the system automatically generates diphone pairs.

For example:

Word: দিশা

IPA: diʃaː

Phonemes:
d i ʃ aː

Generated diphones:
#-d
d-i
i-ʃ
ʃ-aː
aː-#

Repeating this process for thousands of dictionary words produces a large inventory of diphone transitions.

5. Safe Filename System

IPA symbols contain special characters that are not always suitable for file naming.

To ensure compatibility with web servers and operating systems, the project introduces a safe filename mapping.

In this system, each IPA symbol is converted into a standardized ASCII representation.

For example:

IPA diphone: ʃ-aː

Safe filename: sh-aa.wav

This mapping ensures that diphone audio files can be stored and accessed reliably within the speech synthesis system.

6. Diphone Coverage Analysis

A major challenge in diphone-based systems is ensuring that all required diphones have corresponding audio recordings.

To address this issue, the project includes a diphone coverage analyzer.

This tool compares the diphones generated from dictionary words with the diphone audio files available in the system.

The analyzer identifies:

• missing diphones
• unused diphones
• coverage percentage

This information helps guide the recording of additional audio segments.

7. Validator Workflow

Once diphone audio files are recorded, they must be validated before they can be used by the speech system.

The validator checks several properties:

• correct diphone labeling
• consistent filename format
• audio sample rate
• bit depth
• channel configuration

Ensuring technical consistency prevents playback errors and improves the quality of synthesized speech.

8. Integration with the Dictionary

The diphone system is closely integrated with the digital dictionary.

When a user clicks the audio button for a dictionary entry, the system performs the following steps:

1. Convert word to IPA
2. Extract phoneme sequence
3. Generate diphone sequence
4. Locate corresponding audio files
5. Play diphone audio in sequence

Through this process, the dictionary becomes an interactive pronunciation system.

9. Advantages of the Diphone Approach

The diphone method offers several advantages for languages with limited digital resources.

• requires fewer recordings than word-based systems
• supports large vocabulary coverage
• works with rule-based pronunciation systems
• can be implemented with relatively simple software

For the Bishnupriya Manipuri language, this approach provides a practical path toward speech technology development.