fix: move ElevenLabs speech-to-text from the deprecated scribe_v1 to scribe_v2

javascript/src/voice/__tests__/stt.test.ts

@@ -144,7 +144,7 @@ describe("resolveSttProvider — the provider/model router", () => {
   });
 
   it("resolves the built-in ElevenLabs provider", () => {
-    expect(resolveSttProvider("elevenlabs/scribe_v1")).toBeInstanceOf(
+    expect(resolveSttProvider("elevenlabs/scribe_v2")).toBeInstanceOf(
       ElevenLabsSTTProvider,
     );
   });

javascript/src/voice/adapters/__tests__/elevenlabs.test.ts

@@ -402,7 +402,7 @@ describeFeature(
             // concrete modelId so a wrong-model regression fails — call-count
             // alone would pass with the wrong scribe model.
             expect(fakeClient.speechToText.convert).toHaveBeenCalledWith(
-              expect.objectContaining({ modelId: "scribe_v1" }),
+              expect.objectContaining({ modelId: "scribe_v2" }),
             );
           },
         );

javascript/src/voice/stt/elevenlabs-stt.ts

@@ -2,7 +2,7 @@
  * ElevenLabs STT leaf — {@link ElevenLabsSTTProvider} (Scribe). Same
  * {@link STTProvider} contract as the OpenAI leaf, different backend.
  *
- * Uses the `scribe_v1` model via the `@elevenlabs/elevenlabs-js` SDK's
+ * Uses the `scribe_v2` model via the `@elevenlabs/elevenlabs-js` SDK's
  * `speechToText.convert`. PCM16/24 kHz audio is wrapped in a minimal WAV
  * container before posting (EL's endpoint expects a file payload, not raw
  * PCM). Only `text` crosses the {@link STTProvider} boundary — no

javascript/src/voice/stt/stt-provider.ts

@@ -32,7 +32,7 @@ export interface STTProvider {
 
 /**
  * A factory that builds a provider for a given litellm-style model spec
- * (e.g. `"openai/gpt-4o-transcribe"`, `"elevenlabs/scribe_v1"`). The
+ * (e.g. `"openai/gpt-4o-transcribe"`, `"elevenlabs/scribe_v2"`). The
  * `provider` segment selects the factory; the remainder is the model id.
  */
 export type SttProviderFactory = (model: string) => STTProvider;

javascript/src/voice/voice-models.ts

@@ -52,7 +52,7 @@ export const COMPOSABLE_VOICE_LLM_MODEL = "gpt-5.4-mini";
 export const ELEVENLABS_TTS_MODEL = "eleven_v3";
 
 /** ElevenLabs STT model — Python parity: `stt.py:84`. */
-export const ELEVENLABS_STT_MODEL = "scribe_v1";
+export const ELEVENLABS_STT_MODEL = "scribe_v2";
 
 /**
  * Sarah — premade EL voice, free-tier accessible as of 2026-05. Other

python/scenario/voice/stt.py

@@ -88,14 +88,14 @@ async def _transcribe_single(self, audio: AudioChunk) -> str:
 # ---------------------------------------------------------------- ElevenLabs STT
 
 ELEVENLABS_STT_ENDPOINT = "https://api.elevenlabs.io/v1/speech-to-text"
-ELEVENLABS_STT_MODEL = "scribe_v1"
+ELEVENLABS_STT_MODEL = "scribe_v2"
 
 
 class ElevenLabsSTTProvider(STTProvider):
     """
     STT implementation backed by the ElevenLabs REST speech-to-text API.
 
-    Uses the ``scribe_v1`` model. Audio is converted from the canonical
+    Uses the ``scribe_v2`` model. Audio is converted from the canonical
     PCM16/24kHz AudioChunk to a WAV byte payload before posting.
 
     Reads ``ELEVENLABS_API_KEY`` from the environment when ``api_key`` is not