sarvam-ai-sdk 0.1.1 → 0.1.3

package/README.md

@@ -1,6 +1,6 @@
 # AI SDK - Sarvam Provider
 
-The **[Sarvam provider](https://ai-sdk.dev/providers/ai-sdk-providers/sarvam)** for the [AI SDK](https://ai-sdk.dev/docs)
+The **[Sarvam provider](https://v4.ai-sdk.dev/providers/ai-sdk-providers/sarvam)** for the [AI SDK](https://v4.ai-sdk.dev/docs)
 contains language model support for the Sarvam chat completion, Text-to-Speech and Speech-to-Text APIs.
 
 ## Setup
@@ -11,6 +11,9 @@ The **[Sarvam](http://sarvam.ai)** provider is available in the `sarvam-ai-sdk`
 npm i sarvam-ai-sdk
 ```
 
+> [!WARNING]
+> This package only works with Vercel AI-SDK v4, not latest v6. Make sure to install `ai@4` in your project.
+
 ## Provider Instance
 
 You can import the default provider instance `sarvam` from `sarvam-ai-sdk`:
@@ -31,7 +34,7 @@ import { sarvam } from 'sarvam-ai-sdk';
 import { generateText } from 'ai';
 
 const { text } = await generateText({
-    model: sarvam("sarvam-m"),
+	model: sarvam("sarvam-30b"),
     prompt: "Translate this to malayalam: 'Keep cooking, guys'",
 });
 
@@ -46,7 +49,7 @@ import { experimental_generateSpeech as generateSpeech } from "ai";
 import { writeFile } from "fs/promises";
 
 const { audio } = await generateSpeech({
-    model: sarvam.speech("bulbul:v2", "ml-IN"),
+    model: sarvam.speech("bulbul:v3", "ml-IN"),
     text: "പാചകം തുടരൂ, സുഹൃത്തുക്കളേ",
 });
 
@@ -62,13 +65,26 @@ import { experimental_transcribe as transcribe } from "ai";
 import { readFile } from "fs/promises";
 
 const { text } = await transcribe({
-    model: sarvam.transcription("saarika:v2", "ml-IN")
+    model: sarvam.transcription("saarika:v2.5", "ml-IN"),
     audio: await readFile("./src/transcript-test.wav"),
 });
 
 console.log(text); // പാചകം തുടരും സുഹൃത്തുക്കളെ
 ```
 
+```ts
+import { sarvam } from "sarvam-ai-sdk";
+import { experimental_transcribe as transcribe } from "ai";
+import { readFile } from "fs/promises";
+
+const { text } = await transcribe({
+    model: sarvam.transcription("saaras:v3", "en-IN"),
+    audio: await readFile("./src/transcript-test.wav"),
+});
+
+console.log(text); // Pachakam thudaroo, suhruthukkale.
+```
+
 ## Speech-to-Text-Translate
 
 ```ts
@@ -77,7 +93,7 @@ import { experimental_transcribe as transcribe } from "ai";
 import { readFile } from "fs/promises";
 
 const result = await transcribe({
-    model: sarvam.speechTranslation("saaras:v2"),
+    model: sarvam.speechTranslation("saaras:v2.5"),
     audio: await readFile("./src/transcript-test.wav"),
 });
 
@@ -141,7 +157,7 @@ console.log(result.text); // ml-IN
 ## Tool Calling
 
 > [!WARNING]
-> Latest `sarvam-m` model isn't trained on native tool calling feature (aka JSON mode). So we simulate this with prompt engineering technique.
+> Latest `sarvam` models isn't trained on native tool calling feature (aka JSON mode). So we simulate this with prompt engineering technique.
 
 ```ts
 import { z } from "zod";
@@ -150,7 +166,7 @@ import { sarvam } from "sarvam-ai-sdk";
 
 
 const result = await generateText({
-  model: sarvam("sarvam-m", {
+  model: sarvam("sarvam-30b", {
     simulate: "tool-calling" // ⚠️ important
   }),
   tools: {
@@ -174,7 +190,7 @@ console.log(result.toolResults);
 ## Generate JSON object
 
 > [!WARNING]
-> Latest `sarvam-m` model isn't trained on native JSON object generation. So we simulate this with prompt engineering technique.
+> Latest `sarvam` models isn't trained on native JSON object generation. So we simulate this with prompt engineering technique.
 
 ```ts
 import { z } from "zod";
@@ -182,7 +198,7 @@ import { sarvam } from "sarvam-ai-sdk";
 import { generateObject } from 'ai';
 
 const { object } = await generateObject({
-  model: sarvam("sarvam-m", {
+  model: sarvam("sarvam-30b", {
     simulate: "json-object" // ⚠️ important
   }),
   schema: z.object({
@@ -198,6 +214,35 @@ const { object } = await generateObject({
 console.log(object);
 ```
 
+## All APIs
+
+```ts
+import { sarvam } from "sarvam-ai-sdk";
+
+// Text-to-Text + Chat Completion
+sarvam("sarvam-105b");
+sarvam.languageModel("sarvam-30b");
+
+// Text-to-Text + Transliteration
+sarvam.transliterate({ from: "en-IN", to: "ml-IN" });
+
+// Text-to-Text + Translation
+sarvam.translation({ from: "en-IN", to: "ml-IN" });
+
+// Text-to-Text + Language identification
+sarvam.languageIdentification();
+
+// Text-to-Speech
+sarvam.speech("bulbul:v3", "ml-IN");
+
+// Speech-to-Text + Transcribe to same language
+sarvam.transcription("saarika:v2.5");
+
+// Speech-to-Text + Translate to English
+sarvam.speechTranslation("saaras:v3");
+```
+
+
 ## Documentation
 
-Please check out the **[Sarvam provider documentation](https://ai-sdk.dev/providers/ai-sdk-providers/sarvam)** and **[Sarvam API documentation](https://docs.sarvam.ai)** for more information.
+Please check out the **[Sarvam provider documentation](https://v4.ai-sdk.dev/providers/ai-sdk-providers/sarvam)** and **[Sarvam API documentation](https://docs.sarvam.ai)** for more information.

package/dist/index.d.mts

@@ -2,7 +2,15 @@ import { LanguageModelV1, TranscriptionModelV1, SpeechModelV1 } from '@ai-sdk/pr
 import { FetchFunction } from '@ai-sdk/provider-utils';
 import { z } from 'zod';
 
-type SarvamChatModelId = "sarvam-m" | (string & {});
+/**
+ * @description Product models
+ */
+type SarvamChatModelId = "sarvam-30b" | "sarvam-30b-16k" | "sarvam-105b" | "sarvam-105b-32k" | SarvamChatLegacyModelId | (string & {});
+/**
+ * @description Legacy models
+ * @deprecated
+ */
+type SarvamChatLegacyModelId = "sarvam-m";
 interface SarvamChatSettings {
     /**
     * Whether to simulate artificial tool calling or JSON object generation, because Sarvam Models doen't support native Tool Calling or JSON Schmea.
@@ -42,9 +50,9 @@ interface SarvamChatSettings {
 type SarvamLanguageCode = z.infer<typeof SarvamLanguageCodeSchema>;
 declare const SarvamLanguageCodeSchema: z.ZodEnum<["hi-IN", "bn-IN", "kn-IN", "ml-IN", "mr-IN", "od-IN", "pa-IN", "ta-IN", "te-IN", "en-IN", "gu-IN"]>;
 
-type SarvamSpeechModelId = "bulbul:v1" | "bulbul:v2" | (string & {});
+type SarvamSpeechModelId = "bulbul:v2" | "bulbul:v3" | (string & {});
 type SarvamSpeechVoices = z.infer<typeof SpeakerSchema>;
-declare const SpeakerSchema: z.ZodDefault<z.ZodEnum<["meera", "pavithra", "maitreyi", "arvind", "amol", "amartya", "diya", "neel", "misha", "vian", "arjun", "maya", "anushka", "abhilash", "manisha", "vidya", "arya", "karun", "hitesh"]>>;
+declare const SpeakerSchema: z.ZodEnum<["abhilash", "karun", "hitesh", "anushka", "manisha", "vidya", "arya", "shubh", "aditya", "rahul", "rohan", "amit", "dev", "ratan", "varun", "manan", "sumit", "kabir", "aayan", "ashutosh", "advait", "anand", "tarun", "sunny", "mani", "gokul", "vijay", "mohit", "rehan", "soham", "ritu", "priya", "neha", "pooja", "simran", "kavya", "ishita", "shreya", "roopa", "amelia", "sophia", "tanya", "shruti", "suhani", "kavitha", "rupali"]>;
 /**
  * Configuration settings for Sarvam Text-to-Speech API.
  *
@@ -56,10 +64,10 @@ type SarvamSpeechSettings = {
     /**
      * The speaker voice to be used for the output audio.
      *
-     * @default "meera"
-     * @example "meera" (Default female voice for bulbul:v1)
-     * @example "arvind" (Male voice for bulbul:v1)
-     * @example "anushka" (Female voice for bulbul:v2)
+     * @default
+     * - "shubh" (Male voice for bulbul:v3)
+     * - "anushka" (Female voice for bulbul:v2)
+     * - "meera" (Female voice for bulbul:v1)
      */
     speaker?: SarvamSpeechVoices;
     /**
@@ -103,11 +111,74 @@ type SarvamSpeechSettings = {
      * @example false (Disable preprocessing)
      */
     enable_preprocessing?: boolean;
+    /**
+     * Specifies the audio codec for the output audio file.
+     * Different codecs offer various compression and quality characteristics.
+     */
+    output_audio_codec?: "mp3" | "linear16" | "mulaw" | "alaw" | "opus" | "flac" | "aac" | "wav";
+    /**
+     * Temperature controls how much randomness and expressiveness the TTS model uses while generating speech.
+     * Lower values produce more stable and consistent output,
+     * while higher values sound more expressive but may introduce artifacts or errors.
+     *
+     * Any number inbetween 0.01 - 2
+     * @default 0.6
+     *
+     * Note: This parameter is only supported for bulbul:v3. It has no effect on bulbul:v2.
+     */
+    temperature?: number;
+    /**
+     * The ID of a pronunciation dictionary to apply during synthesis.
+     * When provided, matching words in the input text will be replaced with their custom pronunciations before generating speech.
+     *
+     * Only supported by bulbul:v3.
+     */
+    dict_id?: string;
+    /**
+     * Enable caching for the request. When enabled, identical requests will return cached audio instead of regenerating.
+     *
+     * @default false
+     *
+     * Currently in beta and only available for bulbul:v1 and bulbul:v2 models.
+     */
+    enable_cached_responses?: boolean;
 };
 
-type SarvamTranscriptionModelId = "saarika:v2" | "saarika:v1" | "saarika:flash" | (string & {});
-type SarvamSpeechTranslationModelId = "saaras:v1" | "saaras:v2" | "saaras:turbo" | "saaras:flash" | (string & {});
+type SarvamTranscriptionModelId = "saaras:v3" | "saarika:v2.5" | (string & {});
+type SarvamSpeechTranslationModelId = "saaras:v3" | "saaras:v2.5" | (string & {});
+declare const SarvamProviderOptionsSchema: z.ZodObject<{
+    mode: z.ZodDefault<z.ZodEnum<["transcribe", "translate", "verbatim", "translit", "codemix"]>>;
+    with_timestamps: z.ZodDefault<z.ZodOptional<z.ZodNullable<z.ZodBoolean>>>;
+    with_diarization: z.ZodDefault<z.ZodOptional<z.ZodNullable<z.ZodBoolean>>>;
+    num_speakers: z.ZodOptional<z.ZodNullable<z.ZodNumber>>;
+}, "strip", z.ZodTypeAny, {
+    mode: "transcribe" | "translate" | "verbatim" | "translit" | "codemix";
+    with_timestamps: boolean | null;
+    with_diarization: boolean | null;
+    num_speakers?: number | null | undefined;
+}, {
+    mode?: "transcribe" | "translate" | "verbatim" | "translit" | "codemix" | undefined;
+    with_timestamps?: boolean | null | undefined;
+    with_diarization?: boolean | null | undefined;
+    num_speakers?: number | null | undefined;
+}>;
 type SarvamTranscriptionCallOptions = {
+    /**
+    * @default "transcribe"
+    *
+    * @description
+    * - `transcribe`: Standard transcription in the original language, `output`: Text in source language
+    * - `translate`: Transcribe and translate to English, `output`: English text
+    * - `verbatim`: Word-for-word transcription including filler words and repetitions, `output`: Verbatim text in source language
+    * - `translit`: Transcribe and transliterate to Roman script, `output`: Romanized text
+    * - `codemix`: Transcribe code-mixed speech (e.g., Hindi-English) naturally, `output`: Code-mixed text
+    */
+    mode?: z.infer<typeof SarvamProviderOptionsSchema.shape.mode>;
+    /**
+     * - Chunk-level timestamp support
+     * - Useful for subtitle alignment and audio navigation
+     * - Provides start and end times for each segment of text
+     */
     with_timestamps?: boolean;
     /**
      * Enables speaker diarization, which identifies and separates different speakers in the audio.
@@ -240,11 +311,19 @@ interface SarvamProvider {
     /**
     * Creates a model for text generation.
     */
-    (modelId: SarvamChatModelId, settings?: SarvamChatSettings): LanguageModelV1;
+    (
+    /**
+    * @description Sarvam-M (24B) is now a legacy model. But we recommend migrating to Sarvam-30B or Sarvam-105B for improved performance.
+    */
+    modelId: SarvamChatModelId, settings?: SarvamChatSettings): LanguageModelV1;
     /**
     * Creates an Sarvam chat model for text generation.
     */
-    languageModel(modelId: SarvamChatModelId, settings?: SarvamChatSettings): LanguageModelV1;
+    languageModel(
+    /**
+    * @description Sarvam-M (24B) is now a legacy model. But we recommend migrating to Sarvam-30B or Sarvam-105B for improved performance.
+    */
+    modelId: SarvamChatModelId, settings?: SarvamChatSettings): LanguageModelV1;
     /**
     * Creates a Sarvam model for transcription.
     */

package/dist/index.d.ts

@@ -2,7 +2,15 @@ import { LanguageModelV1, TranscriptionModelV1, SpeechModelV1 } from '@ai-sdk/pr
 import { FetchFunction } from '@ai-sdk/provider-utils';
 import { z } from 'zod';
 
-type SarvamChatModelId = "sarvam-m" | (string & {});
+/**
+ * @description Product models
+ */
+type SarvamChatModelId = "sarvam-30b" | "sarvam-30b-16k" | "sarvam-105b" | "sarvam-105b-32k" | SarvamChatLegacyModelId | (string & {});
+/**
+ * @description Legacy models
+ * @deprecated
+ */
+type SarvamChatLegacyModelId = "sarvam-m";
 interface SarvamChatSettings {
     /**
     * Whether to simulate artificial tool calling or JSON object generation, because Sarvam Models doen't support native Tool Calling or JSON Schmea.
@@ -42,9 +50,9 @@ interface SarvamChatSettings {
 type SarvamLanguageCode = z.infer<typeof SarvamLanguageCodeSchema>;
 declare const SarvamLanguageCodeSchema: z.ZodEnum<["hi-IN", "bn-IN", "kn-IN", "ml-IN", "mr-IN", "od-IN", "pa-IN", "ta-IN", "te-IN", "en-IN", "gu-IN"]>;
 
-type SarvamSpeechModelId = "bulbul:v1" | "bulbul:v2" | (string & {});
+type SarvamSpeechModelId = "bulbul:v2" | "bulbul:v3" | (string & {});
 type SarvamSpeechVoices = z.infer<typeof SpeakerSchema>;
-declare const SpeakerSchema: z.ZodDefault<z.ZodEnum<["meera", "pavithra", "maitreyi", "arvind", "amol", "amartya", "diya", "neel", "misha", "vian", "arjun", "maya", "anushka", "abhilash", "manisha", "vidya", "arya", "karun", "hitesh"]>>;
+declare const SpeakerSchema: z.ZodEnum<["abhilash", "karun", "hitesh", "anushka", "manisha", "vidya", "arya", "shubh", "aditya", "rahul", "rohan", "amit", "dev", "ratan", "varun", "manan", "sumit", "kabir", "aayan", "ashutosh", "advait", "anand", "tarun", "sunny", "mani", "gokul", "vijay", "mohit", "rehan", "soham", "ritu", "priya", "neha", "pooja", "simran", "kavya", "ishita", "shreya", "roopa", "amelia", "sophia", "tanya", "shruti", "suhani", "kavitha", "rupali"]>;
 /**
  * Configuration settings for Sarvam Text-to-Speech API.
  *
@@ -56,10 +64,10 @@ type SarvamSpeechSettings = {
     /**
      * The speaker voice to be used for the output audio.
      *
-     * @default "meera"
-     * @example "meera" (Default female voice for bulbul:v1)
-     * @example "arvind" (Male voice for bulbul:v1)
-     * @example "anushka" (Female voice for bulbul:v2)
+     * @default
+     * - "shubh" (Male voice for bulbul:v3)
+     * - "anushka" (Female voice for bulbul:v2)
+     * - "meera" (Female voice for bulbul:v1)
      */
     speaker?: SarvamSpeechVoices;
     /**
@@ -103,11 +111,74 @@ type SarvamSpeechSettings = {
      * @example false (Disable preprocessing)
      */
     enable_preprocessing?: boolean;
+    /**
+     * Specifies the audio codec for the output audio file.
+     * Different codecs offer various compression and quality characteristics.
+     */
+    output_audio_codec?: "mp3" | "linear16" | "mulaw" | "alaw" | "opus" | "flac" | "aac" | "wav";
+    /**
+     * Temperature controls how much randomness and expressiveness the TTS model uses while generating speech.
+     * Lower values produce more stable and consistent output,
+     * while higher values sound more expressive but may introduce artifacts or errors.
+     *
+     * Any number inbetween 0.01 - 2
+     * @default 0.6
+     *
+     * Note: This parameter is only supported for bulbul:v3. It has no effect on bulbul:v2.
+     */
+    temperature?: number;
+    /**
+     * The ID of a pronunciation dictionary to apply during synthesis.
+     * When provided, matching words in the input text will be replaced with their custom pronunciations before generating speech.
+     *
+     * Only supported by bulbul:v3.
+     */
+    dict_id?: string;
+    /**
+     * Enable caching for the request. When enabled, identical requests will return cached audio instead of regenerating.
+     *
+     * @default false
+     *
+     * Currently in beta and only available for bulbul:v1 and bulbul:v2 models.
+     */
+    enable_cached_responses?: boolean;
 };
 
-type SarvamTranscriptionModelId = "saarika:v2" | "saarika:v1" | "saarika:flash" | (string & {});
-type SarvamSpeechTranslationModelId = "saaras:v1" | "saaras:v2" | "saaras:turbo" | "saaras:flash" | (string & {});
+type SarvamTranscriptionModelId = "saaras:v3" | "saarika:v2.5" | (string & {});
+type SarvamSpeechTranslationModelId = "saaras:v3" | "saaras:v2.5" | (string & {});
+declare const SarvamProviderOptionsSchema: z.ZodObject<{
+    mode: z.ZodDefault<z.ZodEnum<["transcribe", "translate", "verbatim", "translit", "codemix"]>>;
+    with_timestamps: z.ZodDefault<z.ZodOptional<z.ZodNullable<z.ZodBoolean>>>;
+    with_diarization: z.ZodDefault<z.ZodOptional<z.ZodNullable<z.ZodBoolean>>>;
+    num_speakers: z.ZodOptional<z.ZodNullable<z.ZodNumber>>;
+}, "strip", z.ZodTypeAny, {
+    mode: "transcribe" | "translate" | "verbatim" | "translit" | "codemix";
+    with_timestamps: boolean | null;
+    with_diarization: boolean | null;
+    num_speakers?: number | null | undefined;
+}, {
+    mode?: "transcribe" | "translate" | "verbatim" | "translit" | "codemix" | undefined;
+    with_timestamps?: boolean | null | undefined;
+    with_diarization?: boolean | null | undefined;
+    num_speakers?: number | null | undefined;
+}>;
 type SarvamTranscriptionCallOptions = {
+    /**
+    * @default "transcribe"
+    *
+    * @description
+    * - `transcribe`: Standard transcription in the original language, `output`: Text in source language
+    * - `translate`: Transcribe and translate to English, `output`: English text
+    * - `verbatim`: Word-for-word transcription including filler words and repetitions, `output`: Verbatim text in source language
+    * - `translit`: Transcribe and transliterate to Roman script, `output`: Romanized text
+    * - `codemix`: Transcribe code-mixed speech (e.g., Hindi-English) naturally, `output`: Code-mixed text
+    */
+    mode?: z.infer<typeof SarvamProviderOptionsSchema.shape.mode>;
+    /**
+     * - Chunk-level timestamp support
+     * - Useful for subtitle alignment and audio navigation
+     * - Provides start and end times for each segment of text
+     */
     with_timestamps?: boolean;
     /**
      * Enables speaker diarization, which identifies and separates different speakers in the audio.
@@ -240,11 +311,19 @@ interface SarvamProvider {
     /**
     * Creates a model for text generation.
     */
-    (modelId: SarvamChatModelId, settings?: SarvamChatSettings): LanguageModelV1;
+    (
+    /**
+    * @description Sarvam-M (24B) is now a legacy model. But we recommend migrating to Sarvam-30B or Sarvam-105B for improved performance.
+    */
+    modelId: SarvamChatModelId, settings?: SarvamChatSettings): LanguageModelV1;
     /**
     * Creates an Sarvam chat model for text generation.
     */
-    languageModel(modelId: SarvamChatModelId, settings?: SarvamChatSettings): LanguageModelV1;
+    languageModel(
+    /**
+    * @description Sarvam-M (24B) is now a legacy model. But we recommend migrating to Sarvam-30B or Sarvam-105B for improved performance.
+    */
+    modelId: SarvamChatModelId, settings?: SarvamChatSettings): LanguageModelV1;
     /**
     * Creates a Sarvam model for transcription.
     */

package/dist/index.js

@@ -767,30 +767,72 @@ var sarvamChatChunkSchema = import_zod2.z.union([
 
 // src/sarvam-speech-model.ts
 var import_provider_utils5 = require("@ai-sdk/provider-utils");
+var import_zod4 = require("zod");
 
 // src/sarvam-speech-settings.ts
 var import_zod3 = require("zod");
 var SpeakerSchema = import_zod3.z.enum([
-  "meera",
-  "pavithra",
-  "maitreyi",
-  "arvind",
-  "amol",
-  "amartya",
-  "diya",
-  "neel",
-  "misha",
-  "vian",
-  "arjun",
-  "maya",
-  "anushka",
+  // male bulbul:v2
   "abhilash",
+  "karun",
+  "hitesh",
+  // female bulbul:v2
+  "anushka",
   "manisha",
   "vidya",
   "arya",
-  "karun",
-  "hitesh"
-]).default("meera");
+  // male bulbul:v3
+  "shubh",
+  "aditya",
+  "rahul",
+  "rohan",
+  "amit",
+  "dev",
+  "ratan",
+  "varun",
+  "manan",
+  "sumit",
+  "kabir",
+  "aayan",
+  "ashutosh",
+  "advait",
+  "anand",
+  "tarun",
+  "sunny",
+  "mani",
+  "gokul",
+  "vijay",
+  "mohit",
+  "rehan",
+  "soham",
+  // female bulbul:v3
+  "ritu",
+  "priya",
+  "neha",
+  "pooja",
+  "simran",
+  "kavya",
+  "ishita",
+  "shreya",
+  "roopa",
+  "amelia",
+  "sophia",
+  "tanya",
+  "shruti",
+  "suhani",
+  "kavitha",
+  "rupali"
+]);
+var outputAudioCodecSchema = import_zod3.z.enum([
+  "mp3",
+  "linear16",
+  "mulaw",
+  "alaw",
+  "opus",
+  "flac",
+  "aac",
+  "wav"
+]);
 var SarvamProviderOptionsSchema = import_zod3.z.object({
   speaker: SpeakerSchema,
   pitch: import_zod3.z.number().min(-0.75).max(0.75).default(0),
@@ -802,11 +844,14 @@ var SarvamProviderOptionsSchema = import_zod3.z.object({
     import_zod3.z.literal(22050),
     import_zod3.z.literal(24e3)
   ]).default(22050),
-  enable_preprocessing: import_zod3.z.boolean().default(false)
+  enable_preprocessing: import_zod3.z.boolean().default(false),
+  output_audio_codec: outputAudioCodecSchema.optional(),
+  temperature: import_zod3.z.number().min(0.01).max(2).default(0.6),
+  dict_id: import_zod3.z.string().optional(),
+  enable_cached_responses: import_zod3.z.boolean().default(false)
 }).partial();
 
 // src/sarvam-speech-model.ts
-var import_zod4 = require("zod");
 var SarvamSpeechModel = class {
   constructor(modelId, languageCode, config) {
     this.modelId = modelId;
@@ -821,7 +866,7 @@ var SarvamSpeechModel = class {
     text,
     voice,
     outputFormat = "wav",
-    // speed,
+    speed,
     // instructions,
     providerOptions
   }) {
@@ -837,48 +882,42 @@ var SarvamSpeechModel = class {
       schema: SarvamProviderOptionsSchema
     });
     const getSpeaker = () => {
-      if (sarvamOptions == null ? void 0 : sarvamOptions.speaker) return sarvamOptions.speaker;
       if (voice) {
         return SpeakerSchema.parse(voice);
       }
       switch (this.modelId) {
-        case "bulbul:v1":
-          return "meera";
         case "bulbul:v2":
           return "manisha";
+        case "bulbul:v3":
+          return "shubh";
       }
-      return "meera";
+      return "shubh";
     };
     const requestBody = {
       model: this.modelId,
       text,
       target_language_code: this.languageCode,
-      speaker: getSpeaker()
-      // response_format: "wav",
-      // speed,
-      // instructions,
+      speaker: getSpeaker(),
+      pace: speed
     };
     if (outputFormat) {
-      if (["mp3", "opus", "aac", "flac", "wav", "pcm"].includes(
-        outputFormat
-      )) {
-        requestBody.response_format = outputFormat;
+      const of = outputAudioCodecSchema.safeParse(outputFormat);
+      if (of.success) {
+        requestBody.output_audio_codec = of.data;
       } else {
         warnings.push({
           type: "unsupported-setting",
           setting: "outputFormat",
-          details: `Unsupported output format: ${outputFormat}. Using mp3 instead.`
+          details: `Unsupported output format: ${outputFormat}. Using wav instead.`
         });
       }
     }
     if (sarvamOptions) {
-      const speechModelOptions = {};
-      for (const key in speechModelOptions) {
-        const value = speechModelOptions[key];
+      Object.entries(sarvamOptions).forEach(([key, value]) => {
         if (value !== void 0) {
           requestBody[key] = value;
         }
-      }
+      });
     }
     return {
       requestBody,
@@ -934,6 +973,7 @@ var import_zod6 = require("zod");
 // src/sarvam-transcription-settings.ts
 var import_zod5 = require("zod");
 var SarvamProviderOptionsSchema2 = import_zod5.z.object({
+  mode: import_zod5.z.enum(["transcribe", "translate", "verbatim", "translit", "codemix"]).default("transcribe"),
   with_timestamps: import_zod5.z.boolean().nullish().default(false),
   with_diarization: import_zod5.z.boolean().nullish().default(false),
   num_speakers: import_zod5.z.number().int().nullish()
@@ -956,10 +996,6 @@ var SarvamTranscriptionModel = class {
     providerOptions
   }) {
     const warnings = [];
-    if (this.modelId === "saarika:v1" && this.languageCode === "unknown")
-      throw new Error(
-        "Language code unknown is not supported for model saarika:v1"
-      );
     const sarvamOptions = (0, import_provider_utils6.parseProviderOptions)({
       provider: "sarvam",
       providerOptions: {
@@ -974,22 +1010,13 @@ var SarvamTranscriptionModel = class {
     const blob = audio instanceof Blob ? audio : new Blob([audio], { type: mediaType });
     formData.append("file", blob);
     formData.append("model", this.modelId);
+    formData.append("language_code", this.languageCode);
     if (sarvamOptions) {
-      formData.append("language_code", this.languageCode);
-      formData.append(
-        "with_timestamps",
-        sarvamOptions.with_timestamps ? "true" : "false"
-      );
-      formData.append(
-        "with_diarization",
-        sarvamOptions.with_diarization ? "true" : "false"
-      );
-      if (sarvamOptions.num_speakers !== null && sarvamOptions.num_speakers !== void 0) {
-        formData.append(
-          "num_speakers",
-          sarvamOptions.num_speakers.toString()
-        );
-      }
+      Object.entries(sarvamOptions).forEach(([key, value]) => {
+        if (value) {
+          formData.append(key, String(value));
+        }
+      });
     }
     return {
       formData,