Voice Mode v4.1.2 Release Notes

Version: 4.1.2 (Feature Release) Date: December 2025 Type: G2P enhancement and lexicon expansion

Summary

Voice Mode v4.1.2 delivers enhanced grapheme-to-phoneme (G2P) conversion with a multi-source fallback chain and significantly expands multi-language lexicon support with 1,384 total pronunciation entries.

New Features

EnhancedG2PService

A new G2P service with intelligent fallback chain for accurate pronunciation generation:

Fallback Chain (in priority order):

Medical Pronunciation Cache (50+ terms, confidence: 0.95)
- Pre-computed IPA for common medical terms
- Includes drugs, conditions, and procedures
CMUdict (English, confidence: 0.9)
- Carnegie Mellon Pronouncing Dictionary
- 134,000+ English words with ARPABET phonemes
- Automatic ARPABET-to-IPA conversion
gruut (Multi-language, confidence: 0.8)
- Pure Python G2P for multiple languages
- Supports: English, German, French, Spanish, Russian, Polish
espeak-ng (Fallback, confidence: 0.7)
- System TTS fallback for unsupported terms
- Broad language coverage including Arabic, CJK
Raw Term Fallback (Last resort, confidence: 0.3)
- Returns term wrapped in slashes: /term/
- Ensures no silent failures

Key Features:

Runtime caching for repeated lookups
Batch generation for multiple terms
Language-aware processing
Comprehensive statistics API

Usage:

from app.services.enhanced_g2p_service import EnhancedG2PService

g2p = EnhancedG2PService()
result = await g2p.generate("metformin", "en")
# G2PResult(term='metformin', phonemes='mɛtfɔːrmɪn', source='medical_cache', confidence=0.95)

ARPABET-to-IPA Conversion

100+ phoneme mappings for converting CMUdict ARPABET to IPA:

ARPABET	IPA	Example
AA	ɑ	father
AE	æ	cat
IY	i	beet
SH	ʃ	ship
TH	θ	think
AA1	ˈɑ	(primary stress)
AA2	ˌɑ	(secondary stress)

Lexicon Expansion

Total coverage increased to 1,384 pronunciation entries:

Language	Terms	Status	Notes
Arabic	485	Complete	Quranic vocabulary
English	852+334	Complete	General + Quranic transliteration
Spanish	210	Complete	Medical terminology
Chinese	160	Complete	Medical terminology
Japanese	55	Expanded	Medical + common terms
Korean	55	Expanded	Medical + common terms
Polish	55	Expanded	Medical + common terms
Russian	55	Expanded	Medical + common terms
Turkish	55	Expanded	Medical + common terms

Documentation Updates

Getting Started Guide

Added comprehensive Getting Started section to What's New v4.1:

Voice-First Input Bar usage
VAD preset selection guide
Quality badge and PHI indicator explanations
Thinking feedback configuration

Screenshot Requirements

Created docs/voice/screenshots/README.md with:

Capture status table for 5 required screenshots
Annotation guidelines and tools
Resolution and format requirements

VAD Preset Terminology

Aligned terminology across all documentation:

Old Term	New Term
Quiet	Sensitive
Normal	Balanced
Noisy	Relaxed

Bug Fixes

Turkish lexicon typo: Fixed "zatürree" → "zatürre" (pneumonia)
ElevenLabs test: Updated model count assertion for new Flash/Turbo v2.5 models

Test Results

======================== 869 passed, 34 skipped ========================

New Tests Added:

test_enhanced_g2p_service.py: 34 integration tests
- Medical cache tests (4)
- CMUdict tests (2)
- Multi-language tests (5)
- Fallback chain tests (7)
- Caching tests (2)
- Batch generation tests (2)
- ARPABET conversion tests (2)
- Edge case tests (5)
- Statistics tests (1)
- Result dataclass tests (4)

Dependencies

New dependencies added to requirements.txt:

cmudict>=1.0.12  # CMU Pronouncing Dictionary for English
gruut>=2.4.0     # Multi-language G2P with pure Python implementation

Installation

cd services/api-gateway
pip install -r requirements.txt

Upgrade Notes

This release is fully backward compatible with v4.1.1. No configuration changes required.

Optional Enhancements:

Install espeak-ng system package for broader language fallback support
Configure medical pronunciation cache for domain-specific terms

Released: December 4, 2025 Commit: 7047d4d PR: #165

Voice Mode v4.1.2 Release Announcement