live-speech-translation
EmmaHarutyunyan/live-speech-translation
Real-time Armenian-to-English speech transcription and translation with speaker identification and local AI processing.
speech-recognitionmachine-translationreal-timewhisperlocal-aiaudio-processingspeaker-identificationarmenian-language
Summary
A real-time speech translation system that transcribes Armenian speech from two audio sources (microphone and system audio), identifies speakers, and translates to English locally using Whisper and a translation model. Designed for live conversation scenarios like calls, it outputs timestamped, labeled transcriptions with latency breakdowns. The project includes detailed setup instructions, model selection rationale, test results, and acknowledges limitations like accuracy and echo issues.