xai TTS: documented API parameters (optimize_streaming_latency, speed, text_normalization) cannot be passed

[moamen-bytesai]

Plugin / version

livekit-plugins-xai 1.6.0, livekit-agents 1.6.0, Python 3.13

Verified on the latest release: inspected the livekit-plugins-xai 1.6.0 wheel from PyPI (uploaded 2026-06-11) and the current main branch (livekit-plugins/livekit-plugins-xai/livekit/plugins/xai/tts.py) — TTS.__init__ and _connect_ws are unchanged; none of the parameters below are exposed. Behavior was originally observed in production on 1.5.17, which is code-identical.

Problem

xAI's TTS WebSocket endpoint (wss://api.x.ai/v1/tts) accepts several connection query parameters documented in the xAI TTS docs:

• optimize_streaming_latency (0/1/2) — first-chunk latency mode
• speed (0.7–1.5) — speaking-rate multiplier
• text_normalization (bool) — spoken-form normalization of numbers/abbreviations

The plugin's TTS.__init__ only accepts api_key, voice, language, tokenizer, http_session (keyword-only, no **kwargs), and _connect_ws sends only voice, language, codec, sample_rate. There is no way to set any of the parameters above — passing them raises TypeError: __init__() got an unexpected keyword argument.

Why it matters

optimize_streaming_latency in particular has a large effect on real-time voice agents. Measured against the live endpoint (same text, voice ara, 4 runs each):

level	generation TTFB
0 (plugin's effective behavior)	mean 741 ms
2	mean 379 ms

For a voice-agent use case this nearly halves TTS time-to-first-byte, but plugin users are stuck on the slowest setting.

Expected behavior

TTS.__init__ exposes these documented parameters (and forwards them as WS query params), similar to how other TTS plugins expose their providers' generation options.

[nightcityblade]

Hi, I'd like to work on this. I'll submit a PR shortly.