Back to catalog / Agent workflows
Build a low-latency spoken FAQ prototype
Transcribe user audio, get a Claude answer, and stream a short TTS reply.
Goal
Transcribe user audio, get a Claude answer, and stream a short TTS reply.
Steps
- Use STT to produce text and segment timing; measure end-to-end latency. [source12]
- Generate an FAQ-grounded Claude answer and synthesize it in TTS chunks. [source12]
- Check transcription, latency, and speech; disclose recording and retain test audio only. [source12]
How to verify
Compare a recorded question with STT text, FAQ evidence, and TTS audio; record STT/model/TTS and end-to-end latency.
Troubleshooting
If the answer leaves the FAQ, inspect transcription and retrieval; if latency rises, measure STT, Claude, and TTS separately.
Permissions
Prerequisites
- Anthropic and ElevenLabs API credentials, microphone permission, synthetic FAQ.
Permissions
- Audio is sent to Anthropic and ElevenLabs and may incur charges; use approved test voices and disclose processing.
Limitations
The cascaded STT/LLM/TTS path can lose prosody and add latency; out-of-scope questions need fallback or human response.
Pinned upstream source
- Repository
- anthropics/anthropic-cookbook
- File path
- third_party/ElevenLabs/low_latency_stt_claude_tts.ipynb
- Pinned commit
9c1e493ddd2b82df399d2a61b9c806c539e18be6- Source checked
- 2026-10-14 · Reviewed against the pinned upstream source
Repository metrics
These values describe the upstream repository, not popularity or likes for an individual skill or playbook.
- Stars
- Not collected
- Forks
- Not collected
- Observed at
- Not collected
No metrics snapshot is available yet.
License: MIT
Main topic
Turn ecommerce campaign materials into a reviewable weekly brief