Pair Synthesizer for NVDA

A Script and Formatting-Aware Speech Synthesizer Switcher

What is it?

Pair Synthesizer is a powerful NVDA add-on that automatically switches between different speech synthesizers based on text type. For example, you can configure Russian text to be spoken by RHVoice, English by eSpeak NG, and numbers by SAPI5 with clear digit pronunciation. It is particularly useful for:
- Multilingual users
- Working with technical texts containing numbers, formulas, code
- Reading text with various formatting (bold, italic, quotes)
- Improving clarity in mixed content

Key Features

• Automatic script detection: Latin, Cyrillic, Arabic, Hindi, Greek, Thai, Han (CJK), numbers, emojis/symbols
• Formatting recognition: Bold text, italic, quoted text, parenthesized text
• Conditional switching: Rules to activate specific engines upon detecting markers (e.g., reading code with a special voice)
• Global speech settings: Unified rate, pitch, and volume for all engines (optional)
• Advanced timing controls: Fine-tune switch delays, safety buffers, duration multipliers
• Full speech sequentiality: Guarantees speech from different engines never overlaps

Configuration

After installation, a new item Pair Synthesizer Settings… appears in the Tools menu.

Important: To change settings, ensure the current synthesizer is NOT Pair Synthesizer. Temporarily switch to another engine (e.g., SAPI5) in NVDA’s speech settings.

“Scripts” Tab

Configure which engine and voice to use for each text type:
- Latin Text (English, French, etc.)
- Cyrillic Text (Russian, Ukrainian, etc.)
- Arabic Text
- Hindi/Devanagari Text
- Numbers (0-9)
- Emojis & Symbols
- Greek Text
- Han Characters (CJK)
- Thai Text

For each type, select:
- Engine – speech synthesizer (e.g., SAPI5, eSpeak NG, RHVoice)
- Voice – specific voice for that engine

“Formatting” Tab

Settings for text with specific formatting:
- Strong – bold text (usually **text**)
- Em – italic text (*text*)
- Quote – quoted text
- Parens – parenthesized text

You can assign a separate engine/voice to make such text stand out intonationally.

“Global Settings” Tab

Performance & Timing Settings

• Switch Delay (ms) – base delay between engine switches (default 150 ms)
• Safety Buffer (ms) – additional “tail” buffer after each fragment finishes to guarantee no overlap (default 300 ms)
• Engine Switch Buffer (ms) – mandatory pause before switching to another engine (default 200 ms)

Global Speech Overrides

• Force Global Speed and Pitch on ALL Engines – apply global rate and pitch to all engines
• Global Rate – global speech rate (0–100)
• Global Pitch – global voice pitch (0–100)
• Mandatory Wait Ratio (Sequential Strictness) – mandatory wait ratio (0–100%). Determines what portion of the estimated fragment duration must be waited before potentially proceeding. 100% = strict sequential, 0% = aggressive switching.

Heuristic Timing Multipliers (Advanced)

For each text type, you can set a multiplier that affects the estimated speech duration. For example, Numbers have a multiplier of 15.0 because “123” is spoken as “one hundred twenty‑three”, taking much longer than three characters.

“Conditional Switching” Tab

Allows creating rules to activate a specific engine when conditions are met.

Add a rule:
- Rule Name – rule name
- Enabled – whether the rule is active
- Condition – condition type:
- Starts With – text starts with marker
- Starts & Ends With – text starts and ends with markers
- Contains – text contains marker
- Punctuation Only – text consists only of punctuation
- Underlined – text is underlined (works with NVDA’s FieldCommand)
- Word Count – segment word count exceeds threshold
- Start Marker / Text – starting marker/text for condition
- End Marker (if applicable) – ending marker (for Starts & Ends With)
- Return to Normal after – when to revert to normal engine:
- Immediately (One-shot) – right after the segment
- Finding End Marker – after finding the end marker
- Speaking X Words – after speaking the given word count
- Any Punctuation – after any punctuation
- Word Limit – word limit (for Speaking X Words)
- Target Engine – engine to switch to
- Target Voice – voice (optional)

Example rule: if text starts with “python", switch to eSpeak NG and read code until "” marker.

Usage

1. After configuration, in NVDA’s speech settings select Pair Synthesizer as the current synthesizer.
2. Now when reading text, NVDA will automatically switch between configured engines based on text type.
3. To force-stop speech, use the standard NVDA keystroke (Control).

Debugging and Logging

If issues occur (speech overlap, incorrect switching), check NVDA’s log. Pair Synthesizer logs detailed error messages. You can also increase Safety Buffer and Engine Switch Buffer for more stable operation on slower systems.

Technical Details

• Based on text segmentation using regular expressions.
• Uses multithreading for loading voices and processing speech without UI blocking.
• Caches synthesizer instances for fast switching.
• Implements a multi‑layer blocking mechanism to guarantee speech sequentiality.