Skip to content

Add voice customer-service example (Qwen/vLLM + switchable browser/Kokoro TTS)#1267

Open
kumare3 wants to merge 2 commits into
mainfrom
example/voice-customer-service-demo
Open

Add voice customer-service example (Qwen/vLLM + switchable browser/Kokoro TTS)#1267
kumare3 wants to merge 2 commits into
mainfrom
example/voice-customer-service-demo

Conversation

@kumare3

@kumare3 kumare3 commented Jun 27, 2026

Copy link
Copy Markdown
Contributor

A browser voice customer-service agent on Flyte.

  • Brain: Qwen2.5-3B on vLLM (L4), OpenAI-compatible, served via a FastAPI UI app that proxies /api/chat.
  • STT: in-browser (Web Speech API); TTS: runtime-switchable — browser speechSynthesis vs. server-side Kokoro-82M (/api/tts) — with barge-in and an on-screen latency comparison.
  • Experimental third engine: a combined LLM+TTS Qwen2.5-Omni (vllm-omni) — code-complete but its talker stage doesn't serve yet; documented in the README as a starting point.
  • Also adds examples/genai/nemotron_omni_voice (Nemotron-Omni serving + Python mic client).

🤖 Generated with Claude Code

kumare3 and others added 2 commits June 27, 2026 18:48
…ble TTS)

examples/genai/voice_customer_service: a browser voice agent backed by
Qwen2.5-3B on vLLM, with a runtime-switchable TTS engine (browser
speechSynthesis vs. server-side Kokoro), barge-in, and live latency metrics.
Includes an experimental, currently-blocked Qwen2.5-Omni (vllm-omni) combined
LLM+TTS engine, documented in the README.

Also adds examples/genai/nemotron_omni_voice (Nemotron-Omni serving + mic client).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Signed-off-by: Ketan Umare <kumare3@users.noreply.github.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Signed-off-by: Ketan Umare <kumare3@users.noreply.github.com>
@kumare3 kumare3 force-pushed the example/voice-customer-service-demo branch from ed904f3 to cea2ddf Compare June 28, 2026 01:49
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant