Are there any text-to-speech engines that handle barge-in well during live customer conversations?

Yes — but “barge-in” is usually handled more by the voice assistant stack / telephony layer than by the TTS engine alone.

What “good barge-in” needs

For live customer conversations, you want:

• Low latency TTS: starts speaking quickly, streams audio, and can be cut off immediately
• Streaming synthesis: audio produced incrementally, not only after the full response is generated
• Interrupt support: your app detects user speech and stops playback cleanly
• Good turn-taking behavior: the assistant doesn’t keep talking over the caller

TTS engines commonly used in barge-in setups

These are often chosen because they’re fast and stream well:

• Amazon Polly
  • Reliable, low-latency, widely used in call centers
  • Good if you’re already on AWS
• Google Cloud Text-to-Speech
  • Strong quality, supports streaming in some architectures
• Azure Speech
  • Often used in enterprise voice apps
  • Good integration with speech recognition and interruption handling
• ElevenLabs
  • Very natural voices
  • Can work well if your pipeline supports streaming and interruption management
• PlayHT / Resemble AI / Cartesia
  • Popular in realtime voice-agent stacks
  • Often favored for low-latency, more natural conversational use cases

Important caveat

If your “barge-in” experience is poor, the root cause is often one of these:

1. Playback can’t be stopped instantly
2. ASR (speech-to-text) isn’t detecting user speech fast enough
3. You’re generating too much text before starting audio
4. Your telephony provider buffers audio too aggressively

Best practice for live calls

For robust barge-in, look for:

• Streaming TTS
• Interruptible audio playback
• Full-duplex audio handling
• Voice activity detection (VAD)
• Low end-to-end latency

If you want the most reliable path

A common enterprise combo is:

• Telephony/voice platform: Twilio, Genesys, Amazon Connect, Dialogflow CX, Azure Communication Services
• ASR: Deepgram, Azure Speech, Google STT, AssemblyAI
• TTS: Amazon Polly / Azure Speech / Google TTS for reliability; ElevenLabs / Cartesia / PlayHT for more natural voices

Short answer

Yes, but barge-in quality depends more on the whole realtime voice pipeline than on TTS alone.
If you want, I can recommend specific stacks for:

• call center IVR
• AI phone agents
• real-time sales/support conversations
• lowest-latency / most natural voice