Tarjumaan is an AI-powered multilingual voice translation system that converts Urdu speech to text and translates it into multiple languages (English, Arabic, Chinese, Spanish, French, German, Turkish, and Persian) with optional audio output.
- Created Streamlit-based web application with audio recording and file upload capabilities
- Integrated Google Speech Recognition for Urdu speech-to-text (free, no API key required)
- Implemented multi-language translation using deep-translator (Google Translate)
- Added text-to-speech functionality using gTTS for all supported languages
- Built comprehensive dataset utilities module for Common Voice and OPUS datasets
- Implemented proper RTL text support for Urdu, Arabic, and Persian scripts
- Fixed gTTS language code compatibility for Chinese variants (zh-cn, zh-tw)
- Preserved original file extensions for uploaded audio files
-
app.py - Main Streamlit application
- Audio input via recording or file upload (WAV, MP3, M4A, OGG, FLAC)
- Google Speech Recognition integration for Urdu (ur-PK)
- Multi-language translation engine
- Text-to-speech generation
- RTL text display support
-
dataset_utils.py - Dataset preprocessing utilities
- Audio normalization and augmentation
- Urdu text cleaning and preprocessing
- Parallel corpus creation
- Dataset splitting and statistics
- Information about Common Voice, OPUS, and Tatoeba datasets
-
.streamlit/config.toml - Streamlit configuration
- Server settings (port 5000, headless mode)
- Frontend: Streamlit
- Speech Recognition: Google Speech Recognition API (free)
- Translation: Google Translate via deep-translator
- Text-to-Speech: gTTS (Google Text-to-Speech)
- Audio Processing: SpeechRecognition, librosa, numpy, pandas, scipy
- No API keys required - uses free Google services
- Browser-based audio recording with st.audio_input
- File upload support for multiple audio formats
- Real-time speech-to-text transcription
- Multi-language translation (9 languages)
- Text-to-speech audio generation
- RTL text support for Arabic-script languages
- Dataset integration information
None documented yet.
None currently.
- Fine-tuned custom models using collected datasets
- Batch translation for multiple audio files
- Translation history with save/export functionality
- Dialect and accent customization
- Real-time streaming translation
- Support for additional languages