Speech and realtime AI
Voice as a product surface: transcription, synthesis, and agents a person can hold a conversation with. Written for roles naming “AI solutions for communication, audio/video streaming, chatbots”.
| # | File | Covers |
|---|---|---|
| 01 | Speech: STT and TTS | streaming vs batch ASR, WER and why it misleads, diarization, custom vocabulary, streaming TTS, cost per minute, testing |
| 02 | Realtime voice agents | cascaded vs speech-to-speech, the latency budget, turn detection and barge-in, WebRTC transport, Pipecat/LiveKit, tools and escalation, evaluation |
Related
- Agent loops and tool calling: ../10_agents_orchestration/
- Chatbot design: ../10_agents_orchestration/11_chatbot.md
- Evaluation: ../13_evaluation/
- Transparency obligations for synthetic media: ../14_guardrails_safety/
- WebSockets and WebRTC: ../../backend/12_protocols/websockets/ and ../../frontend/18_browser_internals/07_webrtc_and_webtransport.md