AI Voice Bots

Use this skill to build, ship, and tune voice bots — phone IVR, real-time speech agents, and voice-first customer interactions — using pure Python frameworks.

This skill owns the voice-specific pipeline: STT, TTS, telephony platforms, latency engineering, and voice quality. For conversation design, persona, and escalation patterns, use ai-bot-builder.

Default posture for Jul 2026: Pipecat as default framework, sub-700ms total turn latency, streaming STT→LLM→TTS pipeline, deterministic VAD, and voice quality metrics from day one. S2S (speech-to-speech) via OpenAI Realtime API or Gemini Live is available as a latency-first alternative — see Framework Selection and references/s2s-and-native-voice-apis.md.

When to Use This Skill

• Building a voice bot for phone, IVR, or real-time speech
• Choosing a telephony platform (Twilio, Vapi, Bland.ai, Retell, Telnyx, Vonage)
• Choosing a voice pipeline framework (Pipecat, LiveKit Agents, Vocode)
• Engineering latency budgets for voice (TTFB, total turn latency)
• Selecting and configuring STT/TTS providers (Deepgram, ElevenLabs, Cartesia, Azure)
• Monitoring voice quality (MOS, WER, call completion rate)
• Designing IVR flows with DTMF and voice hybrid
• Building outbound dialing campaigns