OpenAI TTS: use wav for Groq speech

Made-with: Cursor

extensions/openai/speech-provider.test.ts

@@ -1,7 +1,14 @@
-import { describe, expect, it } from "vitest";
+import { afterEach, describe, expect, it, vi } from "vitest";
 import { buildOpenAISpeechProvider } from "./speech-provider.js";
 
 describe("buildOpenAISpeechProvider", () => {
+  const originalFetch = globalThis.fetch;
+
+  afterEach(() => {
+    globalThis.fetch = originalFetch;
+    vi.restoreAllMocks();
+  });
+
   it("normalizes provider-owned speech config from raw provider config", () => {
     const provider = buildOpenAISpeechProvider();
     const resolved = provider.resolveConfig?.({
@@ -16,6 +23,7 @@ describe("buildOpenAISpeechProvider", () => {
             voice: "alloy",
             speed: 1.25,
             instructions: " Speak warmly ",
+            responseFormat: " WAV ",
           },
         },
       },
@@ -28,6 +36,7 @@ describe("buildOpenAISpeechProvider", () => {
       voice: "alloy",
       speed: 1.25,
       instructions: "Speak warmly",
+      responseFormat: "wav",
     });
   });
 
@@ -67,4 +76,61 @@ describe("buildOpenAISpeechProvider", () => {
       handled: false,
     });
   });
+
+  it("uses wav for Groq-compatible OpenAI TTS endpoints", async () => {
+    const provider = buildOpenAISpeechProvider();
+    const fetchMock = vi.fn(async (_url: string, init?: RequestInit) => {
+      expect(init?.body).toBeTruthy();
+      const body = JSON.parse(String(init?.body)) as { response_format?: string };
+      expect(body.response_format).toBe("wav");
+      return new Response(new Uint8Array([1, 2, 3]), { status: 200 });
+    });
+    globalThis.fetch = fetchMock as unknown as typeof fetch;
+
+    const result = await provider.synthesize({
+      text: "hello",
+      cfg: {} as never,
+      providerConfig: {
+        apiKey: "sk-test",
+        baseUrl: "https://api.groq.com/openai/v1",
+        model: "canopylabs/orpheus-v1-english",
+        voice: "daniel",
+      },
+      target: "audio-file",
+      timeoutMs: 1_000,
+    });
+
+    expect(result.outputFormat).toBe("wav");
+    expect(result.fileExtension).toBe(".wav");
+    expect(result.voiceCompatible).toBe(false);
+  });
+
+  it("honors explicit responseFormat overrides and clears voice-note compatibility when not opus", async () => {
+    const provider = buildOpenAISpeechProvider();
+    const fetchMock = vi.fn(async (_url: string, init?: RequestInit) => {
+      expect(init?.body).toBeTruthy();
+      const body = JSON.parse(String(init?.body)) as { response_format?: string };
+      expect(body.response_format).toBe("wav");
+      return new Response(new Uint8Array([1, 2, 3]), { status: 200 });
+    });
+    globalThis.fetch = fetchMock as unknown as typeof fetch;
+
+    const result = await provider.synthesize({
+      text: "hello",
+      cfg: {} as never,
+      providerConfig: {
+        apiKey: "sk-test",
+        baseUrl: "https://proxy.example.com/openai/v1",
+        model: "canopylabs/orpheus-v1-english",
+        voice: "daniel",
+        responseFormat: "wav",
+      },
+      target: "voice-note",
+      timeoutMs: 1_000,
+    });
+
+    expect(result.outputFormat).toBe("wav");
+    expect(result.fileExtension).toBe(".wav");
+    expect(result.voiceCompatible).toBe(false);
+  });
 });

extensions/openai/speech-provider.ts

@@ -21,6 +21,10 @@ import {
   openaiTTS,
 } from "./tts.js";
 
+const OPENAI_SPEECH_RESPONSE_FORMATS = ["mp3", "opus", "wav"] as const;
+
+type OpenAiSpeechResponseFormat = (typeof OPENAI_SPEECH_RESPONSE_FORMATS)[number];
+
 type OpenAITtsProviderConfig = {
   apiKey?: string;
   baseUrl: string;
@@ -28,6 +32,7 @@ type OpenAITtsProviderConfig = {
   voice: string;
   speed?: number;
   instructions?: string;
+  responseFormat?: OpenAiSpeechResponseFormat;
 };
 
 type OpenAITtsProviderOverrides = {
@@ -36,6 +41,57 @@ type OpenAITtsProviderOverrides = {
   speed?: number;
 };
 
+function normalizeOpenAISpeechResponseFormat(
+  value: unknown,
+): OpenAiSpeechResponseFormat | undefined {
+  const next = trimToUndefined(typeof value === "string" ? value : undefined)?.toLowerCase();
+  if (!next) {
+    return undefined;
+  }
+  if (
+    OPENAI_SPEECH_RESPONSE_FORMATS.includes(next as (typeof OPENAI_SPEECH_RESPONSE_FORMATS)[number])
+  ) {
+    return next as OpenAiSpeechResponseFormat;
+  }
+  throw new Error(`Invalid OpenAI speech responseFormat: ${next}`);
+}
+
+function isGroqSpeechBaseUrl(baseUrl: string): boolean {
+  try {
+    const hostname = new URL(baseUrl).hostname.toLowerCase();
+    return hostname === "groq.com" || hostname.endsWith(".groq.com");
+  } catch {
+    return false;
+  }
+}
+
+function resolveSpeechResponseFormat(
+  baseUrl: string,
+  target: "audio-file" | "voice-note",
+  configuredFormat?: OpenAiSpeechResponseFormat,
+): OpenAiSpeechResponseFormat {
+  if (configuredFormat) {
+    return configuredFormat;
+  }
+  if (isGroqSpeechBaseUrl(baseUrl)) {
+    return "wav";
+  }
+  return target === "voice-note" ? "opus" : "mp3";
+}
+
+function responseFormatToFileExtension(
+  format: OpenAiSpeechResponseFormat,
+): ".mp3" | ".opus" | ".wav" {
+  switch (format) {
+    case "opus":
+      return ".opus";
+    case "wav":
+      return ".wav";
+    default:
+      return ".mp3";
+  }
+}
+
 function normalizeOpenAIProviderConfig(
   rawConfig: Record<string, unknown>,
 ): OpenAITtsProviderConfig {
@@ -54,6 +110,7 @@ function normalizeOpenAIProviderConfig(
     voice: trimToUndefined(raw?.voice) ?? "coral",
     speed: asFiniteNumber(raw?.speed),
     instructions: trimToUndefined(raw?.instructions),
+    responseFormat: normalizeOpenAISpeechResponseFormat(raw?.responseFormat),
   };
 }
 
@@ -66,6 +123,8 @@ function readOpenAIProviderConfig(config: SpeechProviderConfig): OpenAITtsProvid
     voice: trimToUndefined(config.voice) ?? normalized.voice,
     speed: asFiniteNumber(config.speed) ?? normalized.speed,
     instructions: trimToUndefined(config.instructions) ?? normalized.instructions,
+    responseFormat:
+      normalizeOpenAISpeechResponseFormat(config.responseFormat) ?? normalized.responseFormat,
   };
 }
 
@@ -171,7 +230,11 @@ export function buildOpenAISpeechProvider(): SpeechProviderPlugin {
       if (!apiKey) {
         throw new Error("OpenAI API key missing");
       }
-      const responseFormat = req.target === "voice-note" ? "opus" : "mp3";
+      const responseFormat = resolveSpeechResponseFormat(
+        config.baseUrl,
+        req.target,
+        config.responseFormat,
+      );
       const audioBuffer = await openaiTTS({
         text: req.text,
         apiKey,
@@ -186,8 +249,8 @@ export function buildOpenAISpeechProvider(): SpeechProviderPlugin {
       return {
         audioBuffer,
         outputFormat: responseFormat,
-        fileExtension: responseFormat === "opus" ? ".opus" : ".mp3",
-        voiceCompatible: req.target === "voice-note",
+        fileExtension: responseFormatToFileExtension(responseFormat),
+        voiceCompatible: req.target === "voice-note" && responseFormat === "opus",
       };
     },
     synthesizeTelephony: async (req) => {

extensions/openai/tts.ts

@@ -112,7 +112,7 @@ export async function openaiTTS(params: {
   voice: string;
   speed?: number;
   instructions?: string;
-  responseFormat: "mp3" | "opus" | "pcm";
+  responseFormat: "mp3" | "opus" | "pcm" | "wav";
   timeoutMs: number;
 }): Promise<Buffer> {
   const { text, apiKey, baseUrl, model, voice, speed, instructions, responseFormat, timeoutMs } =