fast-transcription

Contact center call transcription with speaker identification uses stereo channel separation or mono diarization to label speakers automatically. It returns per-channel or per-speaker transcripts along with a combined transcript for faster-than-real-time review.

Stereo channel transcription separates agent and customer audio onto distinct channels for accurate speaker labeling. This approach processes both channels simultaneously to deliver faster-than-real-time results.

Azure Speech Fast Transcription supports mono diarization to recognize individual speaker turns on single-channel recordings. It returns a combined transcript with identified speakers at faster-than-real-time speeds using API version 2025-10-15.

Real-time call transcription requires an Azure Speech endpoint and API key configured in your environment. You set AZURE_AI_SPEECH_ENDPOINT and AZURE_AI_SPEECH_KEY to authenticate transcription requests.

Install Python dependencies from requirements.txt including requests, python-dotenv, and pydub. Configure Azure credentials in your environment, then call TranscriptionClient().transcribe_stereo(...) or transcribe_with_diarization(...) to begin.

Use stereo channel transcription when agent and customer audio are recorded on separate channels for precise speaker labeling. Mono diarization is better for single-channel recordings where speaker turns must be identified algorithmically.