@broberg/ai-sdk 0.25.0 → 0.25.1

package/dist/index.d.ts

@@ -1353,7 +1353,9 @@ declare const transcribeInputSchema: z.ZodObject<{
     /** Bias toward brand/jargon terms (Azure phraseList, F029.3); others ignore it. */
     phrases: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
     /** Opt-in timestamps (F036) — a single granularity or an array. Omit → the
-     *  current { text, usage } shape, unchanged. Pass ["word","segment"] for both. */
+     *  current { text, usage } shape, unchanged. Pass ["word","segment"] for both.
+     *  Supported by Whisper (openai) + Azure fast-transcription (F036.1); other
+     *  adapters (Voxtral/mistral) ignore it and return text only. */
     timestamps: z.ZodOptional<z.ZodUnion<[z.ZodEnum<["word", "segment"]>, z.ZodArray<z.ZodEnum<["word", "segment"]>, "many">]>>;
 }, "strip", z.ZodTypeAny, {
     audio: string | Uint8Array<ArrayBuffer>;
@@ -2043,8 +2045,8 @@ declare const falStubAdapter: ProviderAdapter;
  *  wires the live adapters. */
 declare const stubProviders: Record<string, ProviderAdapter>;
 
-declare const VERSION: "0.25.0";
-declare const SDK_TAG: "@broberg/ai-sdk@0.25.0";
+declare const VERSION: "0.25.1";
+declare const SDK_TAG: "@broberg/ai-sdk@0.25.1";
 
 /** Built-in defaults. Every entry is overridable via AiConfig.defaults or a
  *  per-call override.

package/dist/index.js

@@ -1534,7 +1534,26 @@ function azureAdapter(config = {}) {
       outputTokens: 0
     });
     usage.costUsd = minutes * (config.sttPricePerMin ?? AZURE_STT_PRICE_PER_MIN);
-    return { text, usage };
+    const result = { text, usage };
+    if (req.timestamps && req.timestamps.length > 0 && data.phrases) {
+      const toSec = (ms) => (ms ?? 0) / 1e3;
+      if (req.timestamps.includes("word")) {
+        const words = data.phrases.flatMap((p) => p.words ?? []);
+        if (words.length > 0) {
+          result.words = words.map((w) => ({
+            word: w.text ?? "",
+            start: toSec(w.offsetMilliseconds),
+            end: toSec((w.offsetMilliseconds ?? 0) + (w.durationMilliseconds ?? 0))
+          }));
+        }
+      }
+      result.segments = data.phrases.map((p) => ({
+        text: p.text ?? "",
+        start: toSec(p.offsetMilliseconds),
+        end: toSec((p.offsetMilliseconds ?? 0) + (p.durationMilliseconds ?? 0))
+      }));
+    }
+    return result;
   }
   return { name: "azure", tts, transcribe };
 }
@@ -2538,7 +2557,9 @@ var transcribeInputSchema = z.object({
   /** Bias toward brand/jargon terms (Azure phraseList, F029.3); others ignore it. */
   phrases: z.array(z.string()).optional(),
   /** Opt-in timestamps (F036) — a single granularity or an array. Omit → the
-   *  current { text, usage } shape, unchanged. Pass ["word","segment"] for both. */
+   *  current { text, usage } shape, unchanged. Pass ["word","segment"] for both.
+   *  Supported by Whisper (openai) + Azure fast-transcription (F036.1); other
+   *  adapters (Voxtral/mistral) ignore it and return text only. */
   timestamps: z.union([z.enum(["word", "segment"]), z.array(z.enum(["word", "segment"]))]).optional(),
   ...callOptions
 });
@@ -2586,8 +2607,8 @@ var aiConfigSchema = z.object({
 });
 
 // src/version.ts
-var VERSION = "0.25.0";
-var SDK_TAG = "@broberg/ai-sdk@0.25.0";
+var VERSION = "0.25.1";
+var SDK_TAG = "@broberg/ai-sdk@0.25.1";
 
 // src/cost/sinks/upmetrics.ts
 function upmetricsSink(config) {