Speech Transcription and Recording Assistant

ASTRA, the Adaptive Speech Transcription and Recording Assistant, is a hybrid Windows desktop application designed to turn live meetings, interviews, hearings, trainings, consultations, and uploaded recordings into organized, reviewable, and exportable documentation. Before transcription begins, ASTRA prepares audio locally using FFmpeg and Silero Voice Activity Detection. Silent and non-speech portions are skipped, while useful speech is isolated and compressed before online transmission. This reduces upload size, unnecessary AI processing, and provider usage. Long recordings are divided into manageable sections, allowing users to monitor progress, replay audio, retry failed parts, resume interrupted work, and avoid restarting an entire transcription because of one failed section. Users can choose between online and offline processing. Offline mode runs Whisper locally for privacy, poor connectivity, or reduced cloud dependence. Online mode connects to the ASTRA Server through a license-protected API. The server validates access, accepts individual or batched audio clips, creates asynchronous transcription jobs, and returns job status while processing continues. It can route requests across multiple configured speech-to-text providers and automatically try another provider when the preferred service becomes unavailable. This server layer keeps provider credentials away from the desktop app and allows models or providers to be changed without rebuilding the client. After transcription, local Sherpa-ONNX speaker diarization adds anonymous speaker labels and keeps conversations easier to follow across sections. ASTRA also supports transcript polishing, summaries, timestamps, playback, speech-detection logs, processing status, and exportable output. The result is a practical transcription workflow that combines local privacy, cloud performance, provider resilience, and efficient AI resource usage for real documentation work.