Voice Assistant Bottlenecks: ASR Diarization and TTS Latency
dangerous_inference · reddit · 2026-08-05
Developers discussed the engineering challenges of ASR and TTS in building AI voice assistants.
- Latency: As main inference speeds up, ASR and TTS now account for 60% of total interaction latency.
- ASR Pain Point: Using Qwen3 1.7b ASR fails to isolate the target voice in noisy environments; a solution with speaker diarization is urgently needed.
- TTS Stack: Uses Chatterbox TTS Turbo for fast voice cloning and prosody tags like [cough] and [laugh], allowing the assistant to switch voices mid-response for great expressiveness.
More from coding & agent
- Anthropic Releases Free 1-Hour Workshop on Loop Engineering — goyalshaliniuk · 2026-08-05
- OpenAI's Codex Shows Monopolistic Trend, Squeezing Native Agent Frameworks — vista8 · 2026-08-05
- Microsoft's AgentStream: Evaluating Self-Evolving LLM Agents in Streaming Tasks — microsoft · 2026-08-05
- AI Agent Learns New Skills Autonomously: 8-Agent Loop Scours GitHub for Workflows — alexcovo_eth · 2026-08-05
- Pitfall: Claude API Skill Descriptions May Trigger Safety Classifiers — voooooogel · 2026-08-05
- Dev Workflow: 90% Codex, Drops Kimi K3 for GPT-5.6 Luna Max — DeryaTR_ · 2026-08-05