Conversational Voice · Streaming Text-to-Speech · Audio & Speech Models
Independent technical coverage of real-time voice AI infrastructure.
Latest research

Conversational Voice
Voice Agent Handoff to Human Agents in Contact Centers
Context-rich handoffs spare callers from repeating themselves to human agents.
- Porcupine vs openWakeWord vs Custom Models for Embedded Wake WordsHardware limits and licensing, not benchmarks, decide which wake word engine fits.
- Acoustic Echo Cancellation in Full-Duplex Voice AI SystemsHow voice AI stops hearing itself, from DSP filters to neural models.
- Conversation State Management in Long-Running Voice AI SessionsMost voice agents ship, then lose the thread; bounded context fixes it.
- Sentence Boundary Detection for Streaming TTS Chunk SplittingHow to split streaming text for speech without breaking prosody.
Conversational Voice
View all →- Voice Agent Handoff to Human Agents in Contact CentersContext-rich handoffs spare callers from repeating themselves to human agents.
- Porcupine vs openWakeWord vs Custom Models for Embedded Wake WordsHardware limits and licensing, not benchmarks, decide which wake word engine fits.
- Acoustic Echo Cancellation in Full-Duplex Voice AI SystemsHow voice AI stops hearing itself, from DSP filters to neural models.
- Conversation State Management in Long-Running Voice AI SessionsMost voice agents ship, then lose the thread; bounded context fixes it.
Streaming Text-to-Speech
View all →- Sentence Boundary Detection for Streaming TTS Chunk SplittingHow to split streaming text for speech without breaking prosody.
- Streaming TTS Voice Persona Consistency Across Long-Form OutputLong-form TTS systems lose voice consistency at scale, requiring new detection methods and fixes.
- Streaming TTS Punctuation Handling for Natural Pause GenerationStreaming TTS systems use punctuation as real-time pacing commands to balance speed and naturalness.
- SSML Support Across Commercial TTS Streaming APIsDevelopers need to pick a TTS provider based on actual SSML support, not marketing claims.
Audio & Speech Models
View all →- Phoneme-Level Timestamp Alignment in Speech Recognition APIsPhoneme-level timestamps require a separate alignment model, not just better transcription.
- Noise Robustness Testing Methods for Production ASR DeploymentsProduction ASR systems fail in ways that standard benchmarks never measure.
- Real-Time ASR Adaptation for Domain-Specific TerminologySpecialized speech systems need runtime vocabulary injection and fine-tuning to match domain audio.
- Streaming ASR Word-Level Confidence Scores for Downstream NLP UseConfidence scores let downstream systems judge which ASR words to trust.
Features
View all →- Text Segmentation Heuristics That Reduce Streaming TTS Artifact BoundariesStreaming TTS seams come from early commitment, state resets, and capacity limits.
- How Loudspeaker Feedback Reaches the Microphone and What Cancels ItUnderstanding how acoustic loops form and why placement matters most.
- How Voice Agents Persist Session State Across Multi-Turn ConversationsVoice agents must store conversation memory outside the LLM to handle multi-turn dialogue.
- Barge-in and turn-taking: how conversational voice systems handle interruptionWhy voice assistants struggle to know when you're interrupting them.