@ai-sdk/openai 1.3.8 → 1.3.9

package/CHANGELOG.md

@@ -1,5 +1,15 @@
 # @ai-sdk/openai
 
+## 1.3.9
+
+### Patch Changes
+
+- 013faa8: core (ai): change transcription model mimeType to mediaType
+- 013faa8: fix (provider/openai): increase transcription model resilience
+- Updated dependencies [013faa8]
+  - @ai-sdk/provider@1.1.2
+  - @ai-sdk/provider-utils@2.2.6
+
 ## 1.3.8
 
 ### Patch Changes

package/dist/index.js

@@ -1687,7 +1687,7 @@ var OpenAITranscriptionModel = class {
   }
   getArgs({
     audio,
-    mimeType,
+    mediaType,
     providerOptions
   }) {
     const warnings = [];
@@ -1699,7 +1699,7 @@ var OpenAITranscriptionModel = class {
     const formData = new FormData();
     const blob = audio instanceof Uint8Array ? new Blob([audio]) : new Blob([(0, import_provider_utils7.convertBase64ToUint8Array)(audio)]);
     formData.append("model", this.modelId);
-    formData.append("file", new File([blob], "audio", { type: mimeType }));
+    formData.append("file", new File([blob], "audio", { type: mediaType }));
     if (openAIOptions) {
       const transcriptionModelOptions = {
         include: openAIOptions.include,
@@ -1721,10 +1721,14 @@ var OpenAITranscriptionModel = class {
     };
   }
   async doGenerate(options) {
-    var _a, _b, _c;
+    var _a, _b, _c, _d, _e, _f;
     const currentDate = (_c = (_b = (_a = this.config._internal) == null ? void 0 : _a.currentDate) == null ? void 0 : _b.call(_a)) != null ? _c : /* @__PURE__ */ new Date();
     const { formData, warnings } = this.getArgs(options);
-    const { value: response, responseHeaders } = await (0, import_provider_utils7.postFormDataToApi)({
+    const {
+      value: response,
+      responseHeaders,
+      rawValue: rawResponse
+    } = await (0, import_provider_utils7.postFormDataToApi)({
       url: this.config.url({
         path: "/audio/transcriptions",
         modelId: this.modelId
@@ -1738,46 +1742,37 @@ var OpenAITranscriptionModel = class {
       abortSignal: options.abortSignal,
       fetch: this.config.fetch
     });
-    let language;
-    if (response.language && response.language in languageMap) {
-      language = languageMap[response.language];
-    }
+    const language = response.language != null && response.language in languageMap ? languageMap[response.language] : void 0;
     return {
       text: response.text,
-      segments: response.words.map((word) => ({
+      segments: (_e = (_d = response.words) == null ? void 0 : _d.map((word) => ({
         text: word.word,
         startSecond: word.start,
         endSecond: word.end
-      })),
+      }))) != null ? _e : [],
       language,
-      durationInSeconds: response.duration,
+      durationInSeconds: (_f = response.duration) != null ? _f : void 0,
       warnings,
       response: {
         timestamp: currentDate,
         modelId: this.modelId,
         headers: responseHeaders,
-        body: response
-      },
-      // When using format `verbose_json` on `whisper-1`, OpenAI includes the things like `task` and enhanced `segments` information.
-      providerMetadata: {
-        openai: {
-          transcript: response
-        }
+        body: rawResponse
       }
     };
   }
 };
 var openaiTranscriptionResponseSchema = import_zod6.z.object({
   text: import_zod6.z.string(),
-  language: import_zod6.z.string().optional(),
-  duration: import_zod6.z.number().optional(),
+  language: import_zod6.z.string().nullish(),
+  duration: import_zod6.z.number().nullish(),
   words: import_zod6.z.array(
     import_zod6.z.object({
       word: import_zod6.z.string(),
       start: import_zod6.z.number(),
       end: import_zod6.z.number()
     })
-  )
+  ).nullish()
 });
 
 // src/responses/openai-responses-language-model.ts