How Transcription Works

How Transcription Works

Local Processing with whisper.cpp

Myna uses whisper.cpp, an optimised local implementation of OpenAI's Whisper speech recognition model. Transcription runs entirely on your computer — your audio is never sent to any server.

Dual-Channel Transcription

Because Myna records system audio and microphone audio as separate tracks, it transcribes each channel independently:

• System audio track — transcribed and labelled as Remote (other participants)
• Microphone track — transcribed and labelled as You

The two transcripts are then interleaved by timestamp to produce a natural, conversation-style transcript with clear speaker labels.

Timestamps and Audio Seeking

Every line in the transcript includes second-level timestamps. Click any line in the transcript view to jump to that exact moment in the audio playback.

Output Files

Transcription produces two files alongside your recording:

• transcript.json — structured transcript with timestamps and speaker labels
• transcript.txt — plain text version for easy reading and searching

Both files are standard formats that can be opened with any text editor.