chat: fix and complete voice chat (STT/TTS) wiring — full endpoint URLs, live gateway key, engine defaults

The speech section shipped 6 Nov was doubly broken: LibreChat POSTs to
speech.*.openai.url VERBATIM in 0.8.8, so the base '/v1' URL 404'd, and the
STT_API_KEY/TTS_API_KEY env values held a key that no longer exists on the
gateway (401 token_not_found). Also:
- STT: tinfoil/voxtral-small-24b (token-priced; gateway bills ~$0.00008/request,
  125x under the flat $0.01/request whisper; same TEE provider, verified equal
  quality on dictation clips). Whisper remains allowed on the key as fallback.
- TTS voices: full provider speaker list (an unknown voice 400s at the gateway).
- speechTab: object form seeds Settings->Speech defaults (engineSTT/engineTTS
  external, voice casual_female) for members who never saved their own.
- new gateway key: alias librechat-speech, $5/30d, audio models only, injected
  as STT_API_KEY/TTS_API_KEY app env.

Verified on a throwaway librechat-test install then live (chat.inference.coop):
config/get serves the defaults; /api/files/speech/stt returns a correct
transcript; /api/files/speech/tts/manual returns audio. Known gap (gateway-side,
NOT this repo): aspeech rows log $0 spend — voxtral-tts is unmetered on the
gateway's aspeech path while the co-op pays ~$0.002/request.

librechat.yaml

@@ -133,27 +133,42 @@ interface:
           rgb-border-light: '74 82 64'
 
 
-# Speech (STT/TTS) via the co-op gateway. Uses a dedicated LiteLLM virtual key
+# Speech (STT/TTS) via the co-op gateway. LibreChat POSTs to `url` VERBATIM
-# (alias: librechat-speech, $5/30d budget, restricted to the audio models) so
+# (no path is appended), so the FULL endpoint paths are required here.
-# speech never runs on the gateway master key. The gateway is a public
+# Uses a dedicated LiteLLM virtual key (alias: librechat-speech, $5/30d
+# budget, restricted to the audio models) injected as STT_API_KEY/TTS_API_KEY
+# app env — speech never runs on the master key. The gateway is a public
 # endpoint, so no SSRF allowedAddresses exemption is needed.
+# STT: tinfoil/voxtral-small-24b — token-priced (~$0.00008/typical dictation
+# request on the gateway) vs the flat $0.01/request of
+# tinfoil/whisper-large-v3-turbo; same TEE provider, equal quality on short
+# clips (verified round-trip 2026-11-07). Whisper stays allowed on the key as
+# a one-line fallback swap. TTS: tinfoil/voxtral-tts, $0.002/request at cost;
+# voices are the provider's supported speakers (an unknown name 400s).
 speech:
   tts:
     openai:
-      url: "https://gateway.inference.coop/v1"
+      url: "https://gateway.inference.coop/v1/audio/speech"
-      apiKey: "${TTS_API_KEY}"
+      apiKey: ${TTS_API_KEY}
       model: "tinfoil/voxtral-tts"
-      voices: ["casual_female"]
+      voices: ["ar_male", "casual_female", "casual_male", "cheerful_female", "de_female", "de_male", "es_female", "es_male", "fr_female", "fr_male", "hi_female", "hi_male", "it_female", "it_male", "neutral_female", "neutral_male", "nl_female", "nl_male", "pt_female", "pt_male"]
   stt:
     openai:
-      url: "https://gateway.inference.coop/v1"
+      url: "https://gateway.inference.coop/v1/audio/transcriptions"
-      apiKey: "${STT_API_KEY}"
+      apiKey: ${STT_API_KEY}
-      model: "tinfoil/whisper-large-v3-turbo"
+      model: "tinfoil/voxtral-small-24b"
+  # Seeded client defaults (Settings -> Speech): members who never saved
+  # their own speech settings get the External engines out of the box;
+  # a member's own saved settings always win. "Automatic playback" stays
+  # opt-in per member (set textToSpeech.automaticPlayback: true to change).
   speechTab:
     conversationMode: false
     advancedMode: false
-    speechToText: true
+    speechToText:
-    textToSpeech: true
+      engineSTT: external
+    textToSpeech:
+      engineTTS: external
+      voice: casual_female
 
 # Web search: self-hosted SearXNG (search.inference.coop, keyless JSON API)
 # for search, keyless Keenable for page scraping, no reranker. The instance