Fix Java installation error - Replace openjdk-8-jdk with default-jdk for Debian bookworm compatibility - OpenJDK 8 not available in Debian bookworm repositories - Use default-jdk which provides OpenJDK 17 (compatible with UFPAlign)
Implementar UFPAlign com segmentação de sílabas - Substituir MFA por UFPAlign para alinhamento específico do português brasileiro - Adicionar interface para mostrar início e fim de cada sílaba - Exibir informações detalhadas de fonemas, palavras e sílabas - Usar repositório oficial do UFPAlign da UFPA - Interface otimizada para visualização de segmentação silábica
Fix MFA dictionary model name - Add portuguese_mfa dictionary download to Dockerfile - Update app.py to use correct portuguese_mfa models for both dictionary and acoustic - Ensure all three required models (dictionary, acoustic, g2p) are downloaded
Add UFPAlign Docker implementation - Fix MFA Portuguese model names in original Dockerfile - Add comprehensive UFPAlign Docker setup with Kaldi toolkit - Include Docker Compose configuration for multi-container deployment - Add Makefile with convenient commands for container management - Support for interactive shell, API service, and batch processing - Complete documentation for UFPAlign Brazilian Portuguese alignment
Fix Python 3.12 compatibility issues - Remove version pinning for packages to use latest compatible versions - Fix AttributeError: module pkgutil has no attribute ImpImporter - Use Python 3.12 compatible package versions
Use official Montreal Forced Aligner Docker container - Switch to mmcauliffe/montreal-forced-aligner:latest base image - Drastically simplify Dockerfile and avoid all dependency issues - Official container already includes MFA, Kaldi, OpenFST, pynini, etc.
Switch to conda-forge for Montreal Forced Aligner installation - Avoid compilation issues with pre-built packages from conda-forge - Remove unavailable libopenfst-dev packages from Ubuntu 20.04
Fix download_model.sh syntax error and create missing word_boundary.int - Remove broken sed command that corrupted download_model.sh - Generate word_boundary.int from phones.txt during build - Create proper phone boundary mappings for Kaldi - Show word_boundary.int content during build for verification
Pre-extract data.tar.gz during Docker build to fix word_boundary.int issue - Download and extract data files during build instead of runtime - Modify download_model.sh to skip data download if directory exists - Show data directory structure during build for verification - This should resolve the missing word_boundary.int file error
Add comprehensive debugging for UFPAlign data extraction issue - Log directory structure before and after UFPAlign execution - Check for data/lang/phones/word_boundary.int file existence - Identify where data extraction/copying is failing - Debug both source UFPAlign/data and target egs/UFPAlign/s5/data directories
Fix UFPAlign data extraction and environment setup - Add UFPALIGN_DIR environment variable as expected by UFPAlign script - Grant user sudo privileges for file extraction operations - UFPAlign script uses sudo for directory creation and file operations - This should resolve word_boundary.int missing file error
Fix Kaldi egs directory permissions for UFPAlign - Grant user write permissions to /opt/kaldi/egs directory - UFPAlign needs to create working directories in KALDI_ROOT/egs/UFPAlign/s5 - Resolves 'mkdir: Permission denied' error during alignment execution
Fix externally-managed-environment error for Python 3.11+ - Add --break-system-packages flag to all pip install commands - Required for Python 3.11+ which prevents global package installation by default - Safe to use in Docker containers where we want global package installation
Fix Java package compatibility for Debian Bookworm - Replace openjdk-8-jdk with default-jdk for Debian Bookworm compatibility - The kaldiasr/kaldi:latest image uses Debian Bookworm where openjdk-8-jdk is not available - default-jdk provides the same functionality and is compatible with UFPAlign
Enable UFPAlign native automatic model downloading - Remove manual download endpoint and UI - UFPAlign automatically downloads models via utils/download_model.sh on first use - Cleaner implementation following official UFPAlign behavior - Added wget dependency for downloading FalaBrasil annotator lib - Models download automatically when first alignment is requested
Follow official UFPAlign Dockerfile structure with runtime model download - Use official UFPAlign build approach from GitHub - Add runtime model download endpoint /download-models - Update web interface with download models button - Include unzip package for model extraction - Set Portuguese Brazilian locale as per official requirements
Implement robust multi-method download with fallback handling - Replace single gdown with comprehensive script trying multiple methods - Add proper error handling and graceful fallback to test structure - Extensive logging and status reporting