refactor(voice-call): use config for realtime tuning

extensions/openai/realtime-transcription-provider.test.ts

@@ -1,13 +1,7 @@
-import { afterEach, describe, expect, it } from "vitest";
+import { describe, expect, it } from "vitest";
 import { buildOpenAIRealtimeTranscriptionProvider } from "./realtime-transcription-provider.js";
 
 describe("buildOpenAIRealtimeTranscriptionProvider", () => {
-  const originalEnv = { ...process.env };
-
-  afterEach(() => {
-    process.env = { ...originalEnv };
-  });
-
   it("normalizes OpenAI config defaults", () => {
     const provider = buildOpenAIRealtimeTranscriptionProvider();
     const resolved = provider.resolveConfig?.({
@@ -26,15 +20,19 @@ describe("buildOpenAIRealtimeTranscriptionProvider", () => {
     });
   });
 
-  it("reads provider-owned env fallbacks", () => {
-    process.env.REALTIME_TRANSCRIPTION_MODEL = "gpt-4o-transcribe";
-    process.env.SILENCE_DURATION_MS = "900";
-    process.env.VAD_THRESHOLD = "0.45";
-
+  it("keeps provider-owned transcription settings configurable via raw provider config", () => {
     const provider = buildOpenAIRealtimeTranscriptionProvider();
     const resolved = provider.resolveConfig?.({
       cfg: {} as never,
-      rawConfig: {},
+      rawConfig: {
+        providers: {
+          openai: {
+            model: "gpt-4o-transcribe",
+            silenceDurationMs: 900,
+            vadThreshold: 0.45,
+          },
+        },
+      },
     });
 
     expect(resolved).toEqual({

extensions/openai/realtime-transcription-provider.ts

@@ -57,21 +57,9 @@ function normalizeProviderConfig(
         value: raw?.openaiApiKey,
         path: "plugins.entries.voice-call.config.streaming.openaiApiKey",
       }),
-    model:
-      trimToUndefined(raw?.model) ??
-      trimToUndefined(raw?.sttModel) ??
-      trimToUndefined(process.env.REALTIME_TRANSCRIPTION_MODEL) ??
-      trimToUndefined(process.env.STREAMING_STT_MODEL),
-    silenceDurationMs:
-      asNumber(raw?.silenceDurationMs) ??
-      (typeof process.env.SILENCE_DURATION_MS === "string"
-        ? Number.parseInt(process.env.SILENCE_DURATION_MS, 10)
-        : undefined),
-    vadThreshold:
-      asNumber(raw?.vadThreshold) ??
-      (typeof process.env.VAD_THRESHOLD === "string"
-        ? Number.parseFloat(process.env.VAD_THRESHOLD)
-        : undefined),
+    model: trimToUndefined(raw?.model) ?? trimToUndefined(raw?.sttModel),
+    silenceDurationMs: asNumber(raw?.silenceDurationMs),
+    vadThreshold: asNumber(raw?.vadThreshold),
   };
 }
 

extensions/openai/realtime-voice-provider.test.ts

@@ -1,24 +1,22 @@
-import { afterEach, describe, expect, it } from "vitest";
+import { describe, expect, it } from "vitest";
 import { buildOpenAIRealtimeVoiceProvider } from "./realtime-voice-provider.js";
 
 describe("buildOpenAIRealtimeVoiceProvider", () => {
-  const originalEnv = { ...process.env };
-
-  afterEach(() => {
-    process.env = { ...originalEnv };
-  });
-
-  it("normalizes provider-owned env fallbacks", () => {
-    process.env.REALTIME_VOICE_MODEL = "gpt-realtime";
-    process.env.REALTIME_VOICE_VOICE = "verse";
-    process.env.REALTIME_VOICE_TEMPERATURE = "0.6";
-    process.env.SILENCE_DURATION_MS = "850";
-    process.env.VAD_THRESHOLD = "0.35";
-
+  it("normalizes provider-owned voice settings from raw provider config", () => {
     const provider = buildOpenAIRealtimeVoiceProvider();
     const resolved = provider.resolveConfig?.({
       cfg: {} as never,
-      rawConfig: {},
+      rawConfig: {
+        providers: {
+          openai: {
+            model: "gpt-realtime",
+            voice: "verse",
+            temperature: 0.6,
+            silenceDurationMs: 850,
+            vadThreshold: 0.35,
+          },
+        },
+      },
     });
 
     expect(resolved).toEqual({

extensions/openai/realtime-voice-provider.ts

@@ -103,25 +103,11 @@ function normalizeProviderConfig(
       value: raw?.apiKey,
       path: "plugins.entries.voice-call.config.realtime.providers.openai.apiKey",
     }),
-    model: trimToUndefined(raw?.model) ?? trimToUndefined(process.env.REALTIME_VOICE_MODEL),
-    voice: (trimToUndefined(raw?.voice) ?? trimToUndefined(process.env.REALTIME_VOICE_VOICE)) as
-      | OpenAIRealtimeVoice
-      | undefined,
-    temperature:
-      asNumber(raw?.temperature) ??
-      (typeof process.env.REALTIME_VOICE_TEMPERATURE === "string"
-        ? Number.parseFloat(process.env.REALTIME_VOICE_TEMPERATURE)
-        : undefined),
-    vadThreshold:
-      asNumber(raw?.vadThreshold) ??
-      (typeof process.env.VAD_THRESHOLD === "string"
-        ? Number.parseFloat(process.env.VAD_THRESHOLD)
-        : undefined),
-    silenceDurationMs:
-      asNumber(raw?.silenceDurationMs) ??
-      (typeof process.env.SILENCE_DURATION_MS === "string"
-        ? Number.parseInt(process.env.SILENCE_DURATION_MS, 10)
-        : undefined),
+    model: trimToUndefined(raw?.model),
+    voice: trimToUndefined(raw?.voice) as OpenAIRealtimeVoice | undefined,
+    temperature: asNumber(raw?.temperature),
+    vadThreshold: asNumber(raw?.vadThreshold),
+    silenceDurationMs: asNumber(raw?.silenceDurationMs),
     prefixPaddingMs: asNumber(raw?.prefixPaddingMs),
     azureEndpoint: trimToUndefined(raw?.azureEndpoint),
     azureDeployment: trimToUndefined(raw?.azureDeployment),

extensions/voice-call/src/config.test.ts

@@ -273,12 +273,6 @@ describe("normalizeVoiceCallConfig", () => {
 });
 
 describe("resolveVoiceCallConfig", () => {
-  const originalEnv = { ...process.env };
-
-  afterEach(() => {
-    process.env = { ...originalEnv };
-  });
-
   it("keeps legacy streaming OpenAI fields inside providers.openai without forcing provider selection", () => {
     const resolved = resolveVoiceCallConfig({
       enabled: true,
@@ -301,14 +295,13 @@ describe("resolveVoiceCallConfig", () => {
     });
   });
 
-  it("maps realtime instructions from the legacy env hook without altering provider selection", () => {
-    process.env.REALTIME_VOICE_INSTRUCTIONS = "Stay concise.";
-
+  it("preserves configured realtime instructions without env indirection", () => {
     const resolved = resolveVoiceCallConfig({
       enabled: true,
       provider: "twilio",
       realtime: {
         enabled: true,
+        instructions: "Stay concise.",
       },
     });
 

extensions/voice-call/src/config.ts

@@ -605,12 +605,6 @@ export function resolveVoiceCallConfig(config: VoiceCallConfigInput): VoiceCallC
   resolved.streaming = mergeLegacyStreamingOpenAICompat(resolved.streaming);
 
   resolved.realtime = mergeLegacyRealtimeOpenAICompat(resolved.realtime);
-  if (
-    typeof resolved.realtime.instructions !== "string" &&
-    typeof process.env.REALTIME_VOICE_INSTRUCTIONS === "string"
-  ) {
-    resolved.realtime.instructions = process.env.REALTIME_VOICE_INSTRUCTIONS;
-  }
 
   return normalizeVoiceCallConfig(resolved);
 }