@ai-sdk/provider 2.1.0-beta.4 → 2.1.0-beta.5

package/CHANGELOG.md

@@ -1,5 +1,12 @@
 # @ai-sdk/provider
 
+## 2.1.0-beta.5
+
+### Patch Changes
+
+- 046aa3b: feat(provider): speech model v3 spec
+- e8109d3: feat: tool execution approval
+
 ## 2.1.0-beta.4
 
 ### Patch Changes

package/dist/index.d.mts

@@ -1016,12 +1016,24 @@ interface LanguageModelV3ToolResultPart {
      */
     providerOptions?: SharedV3ProviderOptions;
 }
+/**
+ * Result of a tool call.
+ */
 type LanguageModelV3ToolResultOutput = {
     type: 'text';
     value: string;
 } | {
     type: 'json';
     value: JSONValue;
+} | {
+    /**
+     * Type when the user has denied the execution of the tool call.
+     */
+    type: 'execution-denied';
+    /**
+     * Optional reason for the execution denial.
+     */
+    reason?: string;
 } | {
     type: 'error-text';
     value: string;
@@ -2524,8 +2536,8 @@ type LanguageModelV2Middleware = {
     }) => PromiseLike<Awaited<ReturnType<LanguageModelV2['doStream']>>>;
 };
 
-type SpeechModelV2ProviderOptions = Record<string, Record<string, JSONValue>>;
-type SpeechModelV2CallOptions = {
+type SpeechModelV3ProviderOptions = Record<string, Record<string, JSONValue>>;
+type SpeechModelV3CallOptions = {
     /**
      * Text to convert to speech.
      */
@@ -2564,7 +2576,7 @@ type SpeechModelV2CallOptions = {
      * }
      * ```
      */
-    providerOptions?: SpeechModelV2ProviderOptions;
+    providerOptions?: SpeechModelV3ProviderOptions;
     /**
      * Abort signal for cancelling the operation.
      */
@@ -2580,9 +2592,9 @@ type SpeechModelV2CallOptions = {
  * Warning from the model provider for this call. The call will proceed, but e.g.
  * some settings might not be supported, which can lead to suboptimal results.
  */
-type SpeechModelV2CallWarning = {
+type SpeechModelV3CallWarning = {
     type: 'unsupported-setting';
-    setting: keyof SpeechModelV2CallOptions;
+    setting: keyof SpeechModelV3CallOptions;
     details?: string;
 } | {
     type: 'other';
@@ -2590,9 +2602,9 @@ type SpeechModelV2CallWarning = {
 };
 
 /**
- * Speech model specification version 2.
+ * Speech model specification version 3.
  */
-type SpeechModelV2 = {
+type SpeechModelV3 = {
     /**
      * The speech model must specify which speech model interface
      * version it implements. This will allow us to evolve the speech
@@ -2600,7 +2612,7 @@ type SpeechModelV2 = {
      * implementation versions can be handled as a discriminated union
      * on our side.
      */
-    readonly specificationVersion: 'v2';
+    readonly specificationVersion: 'v3';
     /**
      * Name of the provider for logging purposes.
      */
@@ -2612,7 +2624,7 @@ type SpeechModelV2 = {
     /**
      * Generates speech audio from text.
      */
-    doGenerate(options: SpeechModelV2CallOptions): PromiseLike<{
+    doGenerate(options: SpeechModelV3CallOptions): PromiseLike<{
         /**
          * Generated audio as an ArrayBuffer.
          * The audio should be returned without any unnecessary conversion.
@@ -2624,7 +2636,7 @@ type SpeechModelV2 = {
         /**
          * Warnings for the call, e.g. unsupported settings.
          */
-        warnings: Array<SpeechModelV2CallWarning>;
+        warnings: Array<SpeechModelV3CallWarning>;
         /**
          * Optional request information for telemetry and debugging purposes.
          */
@@ -2664,8 +2676,8 @@ type SpeechModelV2 = {
     }>;
 };
 
-type TranscriptionModelV2ProviderOptions = Record<string, Record<string, JSONValue>>;
-type TranscriptionModelV2CallOptions = {
+type TranscriptionModelV3ProviderOptions = Record<string, Record<string, JSONValue>>;
+type TranscriptionModelV3CallOptions = {
     /**
   Audio data to transcribe.
   Accepts a `Uint8Array` or `string`, where `string` is a base64 encoded audio file.
@@ -2691,7 +2703,7 @@ type TranscriptionModelV2CallOptions = {
   }
   ```
    */
-    providerOptions?: TranscriptionModelV2ProviderOptions;
+    providerOptions?: TranscriptionModelV3ProviderOptions;
     /**
   Abort signal for cancelling the operation.
    */
@@ -2707,9 +2719,9 @@ type TranscriptionModelV2CallOptions = {
 Warning from the model provider for this call. The call will proceed, but e.g.
 some settings might not be supported, which can lead to suboptimal results.
  */
-type TranscriptionModelV2CallWarning = {
+type TranscriptionModelV3CallWarning = {
     type: 'unsupported-setting';
-    setting: keyof TranscriptionModelV2CallOptions;
+    setting: keyof TranscriptionModelV3CallOptions;
     details?: string;
 } | {
     type: 'other';
@@ -2717,9 +2729,9 @@ type TranscriptionModelV2CallWarning = {
 };
 
 /**
-Transcription model specification version 2.
+Transcription model specification version 3.
  */
-type TranscriptionModelV2 = {
+type TranscriptionModelV3 = {
     /**
   The transcription model must specify which transcription model interface
   version it implements. This will allow us to evolve the transcription
@@ -2727,7 +2739,7 @@ type TranscriptionModelV2 = {
   implementation versions can be handled as a discriminated union
   on our side.
      */
-    readonly specificationVersion: 'v2';
+    readonly specificationVersion: 'v3';
     /**
   Name of the provider for logging purposes.
      */
@@ -2739,7 +2751,7 @@ type TranscriptionModelV2 = {
     /**
   Generates a transcript.
      */
-    doGenerate(options: TranscriptionModelV2CallOptions): PromiseLike<{
+    doGenerate(options: TranscriptionModelV3CallOptions): PromiseLike<{
         /**
          * The complete transcribed text from the audio.
          */
@@ -2775,7 +2787,7 @@ type TranscriptionModelV2 = {
         /**
     Warnings for the call, e.g. unsupported settings.
          */
-        warnings: Array<TranscriptionModelV2CallWarning>;
+        warnings: Array<TranscriptionModelV3CallWarning>;
         /**
     Optional request information for telemetry and debugging purposes.
          */
@@ -2801,7 +2813,7 @@ type TranscriptionModelV2 = {
             /**
       Response headers.
             */
-            headers?: SharedV2Headers;
+            headers?: SharedV3Headers;
             /**
       Response body.
             */
@@ -2859,7 +2871,7 @@ interface ProviderV3 {
   
   @returns {TranscriptionModel} The transcription model associated with the id
     */
-    transcriptionModel?(modelId: string): TranscriptionModelV2;
+    transcriptionModel?(modelId: string): TranscriptionModelV3;
     /**
   Returns the speech model with the given id.
   The model id is then passed to the provider function to get the model.
@@ -2868,9 +2880,301 @@ interface ProviderV3 {
   
   @returns {SpeechModel} The speech model associated with the id
     */
-    speechModel?(modelId: string): SpeechModelV2;
+    speechModel?(modelId: string): SpeechModelV3;
 }
 
+type SpeechModelV2ProviderOptions = Record<string, Record<string, JSONValue>>;
+type SpeechModelV2CallOptions = {
+    /**
+     * Text to convert to speech.
+     */
+    text: string;
+    /**
+     * The voice to use for speech synthesis.
+     * This is provider-specific and may be a voice ID, name, or other identifier.
+     */
+    voice?: string;
+    /**
+     * The desired output format for the audio e.g. "mp3", "wav", etc.
+     */
+    outputFormat?: string;
+    /**
+     * Instructions for the speech generation e.g. "Speak in a slow and steady tone".
+     */
+    instructions?: string;
+    /**
+     * The speed of the speech generation.
+     */
+    speed?: number;
+    /**
+     * The language for speech generation. This should be an ISO 639-1 language code (e.g. "en", "es", "fr")
+     * or "auto" for automatic language detection. Provider support varies.
+     */
+    language?: string;
+    /**
+     * Additional provider-specific options that are passed through to the provider
+     * as body parameters.
+     *
+     * The outer record is keyed by the provider name, and the inner
+     * record is keyed by the provider-specific metadata key.
+     * ```ts
+     * {
+     *   "openai": {}
+     * }
+     * ```
+     */
+    providerOptions?: SpeechModelV2ProviderOptions;
+    /**
+     * Abort signal for cancelling the operation.
+     */
+    abortSignal?: AbortSignal;
+    /**
+     * Additional HTTP headers to be sent with the request.
+     * Only applicable for HTTP-based providers.
+     */
+    headers?: Record<string, string | undefined>;
+};
+
+/**
+ * Warning from the model provider for this call. The call will proceed, but e.g.
+ * some settings might not be supported, which can lead to suboptimal results.
+ */
+type SpeechModelV2CallWarning = {
+    type: 'unsupported-setting';
+    setting: keyof SpeechModelV2CallOptions;
+    details?: string;
+} | {
+    type: 'other';
+    message: string;
+};
+
+/**
+ * Speech model specification version 2.
+ */
+type SpeechModelV2 = {
+    /**
+     * The speech model must specify which speech model interface
+     * version it implements. This will allow us to evolve the speech
+     * model interface and retain backwards compatibility. The different
+     * implementation versions can be handled as a discriminated union
+     * on our side.
+     */
+    readonly specificationVersion: 'v2';
+    /**
+     * Name of the provider for logging purposes.
+     */
+    readonly provider: string;
+    /**
+     * Provider-specific model ID for logging purposes.
+     */
+    readonly modelId: string;
+    /**
+     * Generates speech audio from text.
+     */
+    doGenerate(options: SpeechModelV2CallOptions): PromiseLike<{
+        /**
+         * Generated audio as an ArrayBuffer.
+         * The audio should be returned without any unnecessary conversion.
+         * If the API returns base64 encoded strings, the audio should be returned
+         * as base64 encoded strings. If the API returns binary data, the audio
+         * should be returned as binary data.
+         */
+        audio: string | Uint8Array;
+        /**
+         * Warnings for the call, e.g. unsupported settings.
+         */
+        warnings: Array<SpeechModelV2CallWarning>;
+        /**
+         * Optional request information for telemetry and debugging purposes.
+         */
+        request?: {
+            /**
+             * Response body (available only for providers that use HTTP requests).
+             */
+            body?: unknown;
+        };
+        /**
+         * Response information for telemetry and debugging purposes.
+         */
+        response: {
+            /**
+             * Timestamp for the start of the generated response.
+             */
+            timestamp: Date;
+            /**
+             * The ID of the response model that was used to generate the response.
+             */
+            modelId: string;
+            /**
+             * Response headers.
+             */
+            headers?: SharedV2Headers;
+            /**
+             * Response body.
+             */
+            body?: unknown;
+        };
+        /**
+         * Additional provider-specific metadata. They are passed through
+         * from the provider to the AI SDK and enable provider-specific
+         * results that can be fully encapsulated in the provider.
+         */
+        providerMetadata?: Record<string, Record<string, JSONValue>>;
+    }>;
+};
+
+type TranscriptionModelV2ProviderOptions = Record<string, Record<string, JSONValue>>;
+type TranscriptionModelV2CallOptions = {
+    /**
+  Audio data to transcribe.
+  Accepts a `Uint8Array` or `string`, where `string` is a base64 encoded audio file.
+       */
+    audio: Uint8Array | string;
+    /**
+  The IANA media type of the audio data.
+  
+  @see https://www.iana.org/assignments/media-types/media-types.xhtml
+     */
+    mediaType: string;
+    /**
+  Additional provider-specific options that are passed through to the provider
+  as body parameters.
+  
+  The outer record is keyed by the provider name, and the inner
+  record is keyed by the provider-specific metadata key.
+  ```ts
+  {
+  "openai": {
+  "timestampGranularities": ["word"]
+  }
+  }
+  ```
+   */
+    providerOptions?: TranscriptionModelV2ProviderOptions;
+    /**
+  Abort signal for cancelling the operation.
+   */
+    abortSignal?: AbortSignal;
+    /**
+  Additional HTTP headers to be sent with the request.
+  Only applicable for HTTP-based providers.
+   */
+    headers?: Record<string, string | undefined>;
+};
+
+/**
+Warning from the model provider for this call. The call will proceed, but e.g.
+some settings might not be supported, which can lead to suboptimal results.
+ */
+type TranscriptionModelV2CallWarning = {
+    type: 'unsupported-setting';
+    setting: keyof TranscriptionModelV2CallOptions;
+    details?: string;
+} | {
+    type: 'other';
+    message: string;
+};
+
+/**
+Transcription model specification version 2.
+ */
+type TranscriptionModelV2 = {
+    /**
+  The transcription model must specify which transcription model interface
+  version it implements. This will allow us to evolve the transcription
+  model interface and retain backwards compatibility. The different
+  implementation versions can be handled as a discriminated union
+  on our side.
+     */
+    readonly specificationVersion: 'v2';
+    /**
+  Name of the provider for logging purposes.
+     */
+    readonly provider: string;
+    /**
+  Provider-specific model ID for logging purposes.
+     */
+    readonly modelId: string;
+    /**
+  Generates a transcript.
+     */
+    doGenerate(options: TranscriptionModelV2CallOptions): PromiseLike<{
+        /**
+         * The complete transcribed text from the audio.
+         */
+        text: string;
+        /**
+         * Array of transcript segments with timing information.
+         * Each segment represents a portion of the transcribed text with start and end times.
+         */
+        segments: Array<{
+            /**
+             * The text content of this segment.
+             */
+            text: string;
+            /**
+             * The start time of this segment in seconds.
+             */
+            startSecond: number;
+            /**
+             * The end time of this segment in seconds.
+             */
+            endSecond: number;
+        }>;
+        /**
+         * The detected language of the audio content, as an ISO-639-1 code (e.g., 'en' for English).
+         * May be undefined if the language couldn't be detected.
+         */
+        language: string | undefined;
+        /**
+         * The total duration of the audio file in seconds.
+         * May be undefined if the duration couldn't be determined.
+         */
+        durationInSeconds: number | undefined;
+        /**
+    Warnings for the call, e.g. unsupported settings.
+         */
+        warnings: Array<TranscriptionModelV2CallWarning>;
+        /**
+    Optional request information for telemetry and debugging purposes.
+         */
+        request?: {
+            /**
+      Raw request HTTP body that was sent to the provider API as a string (JSON should be stringified).
+      Non-HTTP(s) providers should not set this.
+             */
+            body?: string;
+        };
+        /**
+    Response information for telemetry and debugging purposes.
+         */
+        response: {
+            /**
+      Timestamp for the start of the generated response.
+            */
+            timestamp: Date;
+            /**
+      The ID of the response model that was used to generate the response.
+            */
+            modelId: string;
+            /**
+      Response headers.
+            */
+            headers?: SharedV2Headers;
+            /**
+      Response body.
+            */
+            body?: unknown;
+        };
+        /**
+    Additional provider-specific metadata. They are passed through
+    from the provider to the AI SDK and enable provider-specific
+    results that can be fully encapsulated in the provider.
+         */
+        providerMetadata?: Record<string, Record<string, JSONValue>>;
+    }>;
+};
+
 /**
  * Provider for language, text embedding, and image generation models.
  */
@@ -2926,4 +3230,4 @@ interface ProviderV2 {
     speechModel?(modelId: string): SpeechModelV2;
 }
 
-export { AISDKError, APICallError, type EmbeddingModelV2, type EmbeddingModelV2Embedding, type EmbeddingModelV3, type EmbeddingModelV3Embedding, EmptyResponseBodyError, type ImageModelV2, type ImageModelV2CallOptions, type ImageModelV2CallWarning, type ImageModelV2ProviderMetadata, type ImageModelV3, type ImageModelV3CallOptions, type ImageModelV3CallWarning, type ImageModelV3ProviderMetadata, InvalidArgumentError, InvalidPromptError, InvalidResponseDataError, type JSONArray, type JSONObject, JSONParseError, type JSONValue, type LanguageModelV2, type LanguageModelV2CallOptions, type LanguageModelV2CallWarning, type LanguageModelV2Content, type LanguageModelV2DataContent, type LanguageModelV2File, type LanguageModelV2FilePart, type LanguageModelV2FinishReason, type LanguageModelV2FunctionTool, type LanguageModelV2Message, type LanguageModelV2Middleware, type LanguageModelV2Prompt, type LanguageModelV2ProviderDefinedTool, type LanguageModelV2Reasoning, type LanguageModelV2ReasoningPart, type LanguageModelV2ResponseMetadata, type LanguageModelV2Source, type LanguageModelV2StreamPart, type LanguageModelV2Text, type LanguageModelV2TextPart, type LanguageModelV2ToolCall, type LanguageModelV2ToolCallPart, type LanguageModelV2ToolChoice, type LanguageModelV2ToolResultOutput, type LanguageModelV2ToolResultPart, type LanguageModelV2Usage, type LanguageModelV3, type LanguageModelV3CallOptions, type LanguageModelV3CallWarning, type LanguageModelV3Content, type LanguageModelV3DataContent, type LanguageModelV3File, type LanguageModelV3FilePart, type LanguageModelV3FinishReason, type LanguageModelV3FunctionTool, type LanguageModelV3Message, type LanguageModelV3Middleware, type LanguageModelV3Prompt, type LanguageModelV3ProviderDefinedTool, type LanguageModelV3Reasoning, type LanguageModelV3ReasoningPart, type LanguageModelV3ResponseMetadata, type LanguageModelV3Source, type LanguageModelV3StreamPart, type LanguageModelV3Text, type LanguageModelV3TextPart, type LanguageModelV3ToolCall, type LanguageModelV3ToolCallPart, type LanguageModelV3ToolChoice, type LanguageModelV3ToolResult, type LanguageModelV3ToolResultOutput, type LanguageModelV3ToolResultPart, type LanguageModelV3Usage, LoadAPIKeyError, LoadSettingError, NoContentGeneratedError, NoSuchModelError, type ProviderV2, type ProviderV3, type SharedV2Headers, type SharedV2ProviderMetadata, type SharedV2ProviderOptions, type SharedV3Headers, type SharedV3ProviderMetadata, type SharedV3ProviderOptions, type SpeechModelV2, type SpeechModelV2CallOptions, type SpeechModelV2CallWarning, TooManyEmbeddingValuesForCallError, type TranscriptionModelV2, type TranscriptionModelV2CallOptions, type TranscriptionModelV2CallWarning, TypeValidationError, UnsupportedFunctionalityError, getErrorMessage, isJSONArray, isJSONObject, isJSONValue };
+export { AISDKError, APICallError, type EmbeddingModelV2, type EmbeddingModelV2Embedding, type EmbeddingModelV3, type EmbeddingModelV3Embedding, EmptyResponseBodyError, type ImageModelV2, type ImageModelV2CallOptions, type ImageModelV2CallWarning, type ImageModelV2ProviderMetadata, type ImageModelV3, type ImageModelV3CallOptions, type ImageModelV3CallWarning, type ImageModelV3ProviderMetadata, InvalidArgumentError, InvalidPromptError, InvalidResponseDataError, type JSONArray, type JSONObject, JSONParseError, type JSONValue, type LanguageModelV2, type LanguageModelV2CallOptions, type LanguageModelV2CallWarning, type LanguageModelV2Content, type LanguageModelV2DataContent, type LanguageModelV2File, type LanguageModelV2FilePart, type LanguageModelV2FinishReason, type LanguageModelV2FunctionTool, type LanguageModelV2Message, type LanguageModelV2Middleware, type LanguageModelV2Prompt, type LanguageModelV2ProviderDefinedTool, type LanguageModelV2Reasoning, type LanguageModelV2ReasoningPart, type LanguageModelV2ResponseMetadata, type LanguageModelV2Source, type LanguageModelV2StreamPart, type LanguageModelV2Text, type LanguageModelV2TextPart, type LanguageModelV2ToolCall, type LanguageModelV2ToolCallPart, type LanguageModelV2ToolChoice, type LanguageModelV2ToolResultOutput, type LanguageModelV2ToolResultPart, type LanguageModelV2Usage, type LanguageModelV3, type LanguageModelV3CallOptions, type LanguageModelV3CallWarning, type LanguageModelV3Content, type LanguageModelV3DataContent, type LanguageModelV3File, type LanguageModelV3FilePart, type LanguageModelV3FinishReason, type LanguageModelV3FunctionTool, type LanguageModelV3Message, type LanguageModelV3Middleware, type LanguageModelV3Prompt, type LanguageModelV3ProviderDefinedTool, type LanguageModelV3Reasoning, type LanguageModelV3ReasoningPart, type LanguageModelV3ResponseMetadata, type LanguageModelV3Source, type LanguageModelV3StreamPart, type LanguageModelV3Text, type LanguageModelV3TextPart, type LanguageModelV3ToolCall, type LanguageModelV3ToolCallPart, type LanguageModelV3ToolChoice, type LanguageModelV3ToolResult, type LanguageModelV3ToolResultOutput, type LanguageModelV3ToolResultPart, type LanguageModelV3Usage, LoadAPIKeyError, LoadSettingError, NoContentGeneratedError, NoSuchModelError, type ProviderV2, type ProviderV3, type SharedV2Headers, type SharedV2ProviderMetadata, type SharedV2ProviderOptions, type SharedV3Headers, type SharedV3ProviderMetadata, type SharedV3ProviderOptions, type SpeechModelV2, type SpeechModelV2CallOptions, type SpeechModelV2CallWarning, type SpeechModelV3, type SpeechModelV3CallOptions, type SpeechModelV3CallWarning, TooManyEmbeddingValuesForCallError, type TranscriptionModelV2, type TranscriptionModelV2CallOptions, type TranscriptionModelV2CallWarning, type TranscriptionModelV3, type TranscriptionModelV3CallOptions, type TranscriptionModelV3CallWarning, TypeValidationError, UnsupportedFunctionalityError, getErrorMessage, isJSONArray, isJSONObject, isJSONValue };

package/dist/index.d.ts

@@ -1016,12 +1016,24 @@ interface LanguageModelV3ToolResultPart {
      */
     providerOptions?: SharedV3ProviderOptions;
 }
+/**
+ * Result of a tool call.
+ */
 type LanguageModelV3ToolResultOutput = {
     type: 'text';
     value: string;
 } | {
     type: 'json';
     value: JSONValue;
+} | {
+    /**
+     * Type when the user has denied the execution of the tool call.
+     */
+    type: 'execution-denied';
+    /**
+     * Optional reason for the execution denial.
+     */
+    reason?: string;
 } | {
     type: 'error-text';
     value: string;
@@ -2524,8 +2536,8 @@ type LanguageModelV2Middleware = {
     }) => PromiseLike<Awaited<ReturnType<LanguageModelV2['doStream']>>>;
 };
 
-type SpeechModelV2ProviderOptions = Record<string, Record<string, JSONValue>>;
-type SpeechModelV2CallOptions = {
+type SpeechModelV3ProviderOptions = Record<string, Record<string, JSONValue>>;
+type SpeechModelV3CallOptions = {
     /**
      * Text to convert to speech.
      */
@@ -2564,7 +2576,7 @@ type SpeechModelV2CallOptions = {
      * }
      * ```
      */
-    providerOptions?: SpeechModelV2ProviderOptions;
+    providerOptions?: SpeechModelV3ProviderOptions;
     /**
      * Abort signal for cancelling the operation.
      */
@@ -2580,9 +2592,9 @@ type SpeechModelV2CallOptions = {
  * Warning from the model provider for this call. The call will proceed, but e.g.
  * some settings might not be supported, which can lead to suboptimal results.
  */
-type SpeechModelV2CallWarning = {
+type SpeechModelV3CallWarning = {
     type: 'unsupported-setting';
-    setting: keyof SpeechModelV2CallOptions;
+    setting: keyof SpeechModelV3CallOptions;
     details?: string;
 } | {
     type: 'other';
@@ -2590,9 +2602,9 @@ type SpeechModelV2CallWarning = {
 };
 
 /**
- * Speech model specification version 2.
+ * Speech model specification version 3.
  */
-type SpeechModelV2 = {
+type SpeechModelV3 = {
     /**
      * The speech model must specify which speech model interface
      * version it implements. This will allow us to evolve the speech
@@ -2600,7 +2612,7 @@ type SpeechModelV2 = {
      * implementation versions can be handled as a discriminated union
      * on our side.
      */
-    readonly specificationVersion: 'v2';
+    readonly specificationVersion: 'v3';
     /**
      * Name of the provider for logging purposes.
      */
@@ -2612,7 +2624,7 @@ type SpeechModelV2 = {
     /**
      * Generates speech audio from text.
      */
-    doGenerate(options: SpeechModelV2CallOptions): PromiseLike<{
+    doGenerate(options: SpeechModelV3CallOptions): PromiseLike<{
         /**
          * Generated audio as an ArrayBuffer.
          * The audio should be returned without any unnecessary conversion.
@@ -2624,7 +2636,7 @@ type SpeechModelV2 = {
         /**
          * Warnings for the call, e.g. unsupported settings.
          */
-        warnings: Array<SpeechModelV2CallWarning>;
+        warnings: Array<SpeechModelV3CallWarning>;
         /**
          * Optional request information for telemetry and debugging purposes.
          */
@@ -2664,8 +2676,8 @@ type SpeechModelV2 = {
     }>;
 };
 
-type TranscriptionModelV2ProviderOptions = Record<string, Record<string, JSONValue>>;
-type TranscriptionModelV2CallOptions = {
+type TranscriptionModelV3ProviderOptions = Record<string, Record<string, JSONValue>>;
+type TranscriptionModelV3CallOptions = {
     /**
   Audio data to transcribe.
   Accepts a `Uint8Array` or `string`, where `string` is a base64 encoded audio file.
@@ -2691,7 +2703,7 @@ type TranscriptionModelV2CallOptions = {
   }
   ```
    */
-    providerOptions?: TranscriptionModelV2ProviderOptions;
+    providerOptions?: TranscriptionModelV3ProviderOptions;
     /**
   Abort signal for cancelling the operation.
    */
@@ -2707,9 +2719,9 @@ type TranscriptionModelV2CallOptions = {
 Warning from the model provider for this call. The call will proceed, but e.g.
 some settings might not be supported, which can lead to suboptimal results.
  */
-type TranscriptionModelV2CallWarning = {
+type TranscriptionModelV3CallWarning = {
     type: 'unsupported-setting';
-    setting: keyof TranscriptionModelV2CallOptions;
+    setting: keyof TranscriptionModelV3CallOptions;
     details?: string;
 } | {
     type: 'other';
@@ -2717,9 +2729,9 @@ type TranscriptionModelV2CallWarning = {
 };
 
 /**
-Transcription model specification version 2.
+Transcription model specification version 3.
  */
-type TranscriptionModelV2 = {
+type TranscriptionModelV3 = {
     /**
   The transcription model must specify which transcription model interface
   version it implements. This will allow us to evolve the transcription
@@ -2727,7 +2739,7 @@ type TranscriptionModelV2 = {
   implementation versions can be handled as a discriminated union
   on our side.
      */
-    readonly specificationVersion: 'v2';
+    readonly specificationVersion: 'v3';
     /**
   Name of the provider for logging purposes.
      */
@@ -2739,7 +2751,7 @@ type TranscriptionModelV2 = {
     /**
   Generates a transcript.
      */
-    doGenerate(options: TranscriptionModelV2CallOptions): PromiseLike<{
+    doGenerate(options: TranscriptionModelV3CallOptions): PromiseLike<{
         /**
          * The complete transcribed text from the audio.
          */
@@ -2775,7 +2787,7 @@ type TranscriptionModelV2 = {
         /**
     Warnings for the call, e.g. unsupported settings.
          */
-        warnings: Array<TranscriptionModelV2CallWarning>;
+        warnings: Array<TranscriptionModelV3CallWarning>;
         /**
     Optional request information for telemetry and debugging purposes.
          */
@@ -2801,7 +2813,7 @@ type TranscriptionModelV2 = {
             /**
       Response headers.
             */
-            headers?: SharedV2Headers;
+            headers?: SharedV3Headers;
             /**
       Response body.
             */
@@ -2859,7 +2871,7 @@ interface ProviderV3 {
   
   @returns {TranscriptionModel} The transcription model associated with the id
     */
-    transcriptionModel?(modelId: string): TranscriptionModelV2;
+    transcriptionModel?(modelId: string): TranscriptionModelV3;
     /**
   Returns the speech model with the given id.
   The model id is then passed to the provider function to get the model.
@@ -2868,9 +2880,301 @@ interface ProviderV3 {
   
   @returns {SpeechModel} The speech model associated with the id
     */
-    speechModel?(modelId: string): SpeechModelV2;
+    speechModel?(modelId: string): SpeechModelV3;
 }
 
+type SpeechModelV2ProviderOptions = Record<string, Record<string, JSONValue>>;
+type SpeechModelV2CallOptions = {
+    /**
+     * Text to convert to speech.
+     */
+    text: string;
+    /**
+     * The voice to use for speech synthesis.
+     * This is provider-specific and may be a voice ID, name, or other identifier.
+     */
+    voice?: string;
+    /**
+     * The desired output format for the audio e.g. "mp3", "wav", etc.
+     */
+    outputFormat?: string;
+    /**
+     * Instructions for the speech generation e.g. "Speak in a slow and steady tone".
+     */
+    instructions?: string;
+    /**
+     * The speed of the speech generation.
+     */
+    speed?: number;
+    /**
+     * The language for speech generation. This should be an ISO 639-1 language code (e.g. "en", "es", "fr")
+     * or "auto" for automatic language detection. Provider support varies.
+     */
+    language?: string;
+    /**
+     * Additional provider-specific options that are passed through to the provider
+     * as body parameters.
+     *
+     * The outer record is keyed by the provider name, and the inner
+     * record is keyed by the provider-specific metadata key.
+     * ```ts
+     * {
+     *   "openai": {}
+     * }
+     * ```
+     */
+    providerOptions?: SpeechModelV2ProviderOptions;
+    /**
+     * Abort signal for cancelling the operation.
+     */
+    abortSignal?: AbortSignal;
+    /**
+     * Additional HTTP headers to be sent with the request.
+     * Only applicable for HTTP-based providers.
+     */
+    headers?: Record<string, string | undefined>;
+};
+
+/**
+ * Warning from the model provider for this call. The call will proceed, but e.g.
+ * some settings might not be supported, which can lead to suboptimal results.
+ */
+type SpeechModelV2CallWarning = {
+    type: 'unsupported-setting';
+    setting: keyof SpeechModelV2CallOptions;
+    details?: string;
+} | {
+    type: 'other';
+    message: string;
+};
+
+/**
+ * Speech model specification version 2.
+ */
+type SpeechModelV2 = {
+    /**
+     * The speech model must specify which speech model interface
+     * version it implements. This will allow us to evolve the speech
+     * model interface and retain backwards compatibility. The different
+     * implementation versions can be handled as a discriminated union
+     * on our side.
+     */
+    readonly specificationVersion: 'v2';
+    /**
+     * Name of the provider for logging purposes.
+     */
+    readonly provider: string;
+    /**
+     * Provider-specific model ID for logging purposes.
+     */
+    readonly modelId: string;
+    /**
+     * Generates speech audio from text.
+     */
+    doGenerate(options: SpeechModelV2CallOptions): PromiseLike<{
+        /**
+         * Generated audio as an ArrayBuffer.
+         * The audio should be returned without any unnecessary conversion.
+         * If the API returns base64 encoded strings, the audio should be returned
+         * as base64 encoded strings. If the API returns binary data, the audio
+         * should be returned as binary data.
+         */
+        audio: string | Uint8Array;
+        /**
+         * Warnings for the call, e.g. unsupported settings.
+         */
+        warnings: Array<SpeechModelV2CallWarning>;
+        /**
+         * Optional request information for telemetry and debugging purposes.
+         */
+        request?: {
+            /**
+             * Response body (available only for providers that use HTTP requests).
+             */
+            body?: unknown;
+        };
+        /**
+         * Response information for telemetry and debugging purposes.
+         */
+        response: {
+            /**
+             * Timestamp for the start of the generated response.
+             */
+            timestamp: Date;
+            /**
+             * The ID of the response model that was used to generate the response.
+             */
+            modelId: string;
+            /**
+             * Response headers.
+             */
+            headers?: SharedV2Headers;
+            /**
+             * Response body.
+             */
+            body?: unknown;
+        };
+        /**
+         * Additional provider-specific metadata. They are passed through
+         * from the provider to the AI SDK and enable provider-specific
+         * results that can be fully encapsulated in the provider.
+         */
+        providerMetadata?: Record<string, Record<string, JSONValue>>;
+    }>;
+};
+
+type TranscriptionModelV2ProviderOptions = Record<string, Record<string, JSONValue>>;
+type TranscriptionModelV2CallOptions = {
+    /**
+  Audio data to transcribe.
+  Accepts a `Uint8Array` or `string`, where `string` is a base64 encoded audio file.
+       */
+    audio: Uint8Array | string;
+    /**
+  The IANA media type of the audio data.
+  
+  @see https://www.iana.org/assignments/media-types/media-types.xhtml
+     */
+    mediaType: string;
+    /**
+  Additional provider-specific options that are passed through to the provider
+  as body parameters.
+  
+  The outer record is keyed by the provider name, and the inner
+  record is keyed by the provider-specific metadata key.
+  ```ts
+  {
+  "openai": {
+  "timestampGranularities": ["word"]
+  }
+  }
+  ```
+   */
+    providerOptions?: TranscriptionModelV2ProviderOptions;
+    /**
+  Abort signal for cancelling the operation.
+   */
+    abortSignal?: AbortSignal;
+    /**
+  Additional HTTP headers to be sent with the request.
+  Only applicable for HTTP-based providers.
+   */
+    headers?: Record<string, string | undefined>;
+};
+
+/**
+Warning from the model provider for this call. The call will proceed, but e.g.
+some settings might not be supported, which can lead to suboptimal results.
+ */
+type TranscriptionModelV2CallWarning = {
+    type: 'unsupported-setting';
+    setting: keyof TranscriptionModelV2CallOptions;
+    details?: string;
+} | {
+    type: 'other';
+    message: string;
+};
+
+/**
+Transcription model specification version 2.
+ */
+type TranscriptionModelV2 = {
+    /**
+  The transcription model must specify which transcription model interface
+  version it implements. This will allow us to evolve the transcription
+  model interface and retain backwards compatibility. The different
+  implementation versions can be handled as a discriminated union
+  on our side.
+     */
+    readonly specificationVersion: 'v2';
+    /**
+  Name of the provider for logging purposes.
+     */
+    readonly provider: string;
+    /**
+  Provider-specific model ID for logging purposes.
+     */
+    readonly modelId: string;
+    /**
+  Generates a transcript.
+     */
+    doGenerate(options: TranscriptionModelV2CallOptions): PromiseLike<{
+        /**
+         * The complete transcribed text from the audio.
+         */
+        text: string;
+        /**
+         * Array of transcript segments with timing information.
+         * Each segment represents a portion of the transcribed text with start and end times.
+         */
+        segments: Array<{
+            /**
+             * The text content of this segment.
+             */
+            text: string;
+            /**
+             * The start time of this segment in seconds.
+             */
+            startSecond: number;
+            /**
+             * The end time of this segment in seconds.
+             */
+            endSecond: number;
+        }>;
+        /**
+         * The detected language of the audio content, as an ISO-639-1 code (e.g., 'en' for English).
+         * May be undefined if the language couldn't be detected.
+         */
+        language: string | undefined;
+        /**
+         * The total duration of the audio file in seconds.
+         * May be undefined if the duration couldn't be determined.
+         */
+        durationInSeconds: number | undefined;
+        /**
+    Warnings for the call, e.g. unsupported settings.
+         */
+        warnings: Array<TranscriptionModelV2CallWarning>;
+        /**
+    Optional request information for telemetry and debugging purposes.
+         */
+        request?: {
+            /**
+      Raw request HTTP body that was sent to the provider API as a string (JSON should be stringified).
+      Non-HTTP(s) providers should not set this.
+             */
+            body?: string;
+        };
+        /**
+    Response information for telemetry and debugging purposes.
+         */
+        response: {
+            /**
+      Timestamp for the start of the generated response.
+            */
+            timestamp: Date;
+            /**
+      The ID of the response model that was used to generate the response.
+            */
+            modelId: string;
+            /**
+      Response headers.
+            */
+            headers?: SharedV2Headers;
+            /**
+      Response body.
+            */
+            body?: unknown;
+        };
+        /**
+    Additional provider-specific metadata. They are passed through
+    from the provider to the AI SDK and enable provider-specific
+    results that can be fully encapsulated in the provider.
+         */
+        providerMetadata?: Record<string, Record<string, JSONValue>>;
+    }>;
+};
+
 /**
  * Provider for language, text embedding, and image generation models.
  */
@@ -2926,4 +3230,4 @@ interface ProviderV2 {
     speechModel?(modelId: string): SpeechModelV2;
 }
 
-export { AISDKError, APICallError, type EmbeddingModelV2, type EmbeddingModelV2Embedding, type EmbeddingModelV3, type EmbeddingModelV3Embedding, EmptyResponseBodyError, type ImageModelV2, type ImageModelV2CallOptions, type ImageModelV2CallWarning, type ImageModelV2ProviderMetadata, type ImageModelV3, type ImageModelV3CallOptions, type ImageModelV3CallWarning, type ImageModelV3ProviderMetadata, InvalidArgumentError, InvalidPromptError, InvalidResponseDataError, type JSONArray, type JSONObject, JSONParseError, type JSONValue, type LanguageModelV2, type LanguageModelV2CallOptions, type LanguageModelV2CallWarning, type LanguageModelV2Content, type LanguageModelV2DataContent, type LanguageModelV2File, type LanguageModelV2FilePart, type LanguageModelV2FinishReason, type LanguageModelV2FunctionTool, type LanguageModelV2Message, type LanguageModelV2Middleware, type LanguageModelV2Prompt, type LanguageModelV2ProviderDefinedTool, type LanguageModelV2Reasoning, type LanguageModelV2ReasoningPart, type LanguageModelV2ResponseMetadata, type LanguageModelV2Source, type LanguageModelV2StreamPart, type LanguageModelV2Text, type LanguageModelV2TextPart, type LanguageModelV2ToolCall, type LanguageModelV2ToolCallPart, type LanguageModelV2ToolChoice, type LanguageModelV2ToolResultOutput, type LanguageModelV2ToolResultPart, type LanguageModelV2Usage, type LanguageModelV3, type LanguageModelV3CallOptions, type LanguageModelV3CallWarning, type LanguageModelV3Content, type LanguageModelV3DataContent, type LanguageModelV3File, type LanguageModelV3FilePart, type LanguageModelV3FinishReason, type LanguageModelV3FunctionTool, type LanguageModelV3Message, type LanguageModelV3Middleware, type LanguageModelV3Prompt, type LanguageModelV3ProviderDefinedTool, type LanguageModelV3Reasoning, type LanguageModelV3ReasoningPart, type LanguageModelV3ResponseMetadata, type LanguageModelV3Source, type LanguageModelV3StreamPart, type LanguageModelV3Text, type LanguageModelV3TextPart, type LanguageModelV3ToolCall, type LanguageModelV3ToolCallPart, type LanguageModelV3ToolChoice, type LanguageModelV3ToolResult, type LanguageModelV3ToolResultOutput, type LanguageModelV3ToolResultPart, type LanguageModelV3Usage, LoadAPIKeyError, LoadSettingError, NoContentGeneratedError, NoSuchModelError, type ProviderV2, type ProviderV3, type SharedV2Headers, type SharedV2ProviderMetadata, type SharedV2ProviderOptions, type SharedV3Headers, type SharedV3ProviderMetadata, type SharedV3ProviderOptions, type SpeechModelV2, type SpeechModelV2CallOptions, type SpeechModelV2CallWarning, TooManyEmbeddingValuesForCallError, type TranscriptionModelV2, type TranscriptionModelV2CallOptions, type TranscriptionModelV2CallWarning, TypeValidationError, UnsupportedFunctionalityError, getErrorMessage, isJSONArray, isJSONObject, isJSONValue };
+export { AISDKError, APICallError, type EmbeddingModelV2, type EmbeddingModelV2Embedding, type EmbeddingModelV3, type EmbeddingModelV3Embedding, EmptyResponseBodyError, type ImageModelV2, type ImageModelV2CallOptions, type ImageModelV2CallWarning, type ImageModelV2ProviderMetadata, type ImageModelV3, type ImageModelV3CallOptions, type ImageModelV3CallWarning, type ImageModelV3ProviderMetadata, InvalidArgumentError, InvalidPromptError, InvalidResponseDataError, type JSONArray, type JSONObject, JSONParseError, type JSONValue, type LanguageModelV2, type LanguageModelV2CallOptions, type LanguageModelV2CallWarning, type LanguageModelV2Content, type LanguageModelV2DataContent, type LanguageModelV2File, type LanguageModelV2FilePart, type LanguageModelV2FinishReason, type LanguageModelV2FunctionTool, type LanguageModelV2Message, type LanguageModelV2Middleware, type LanguageModelV2Prompt, type LanguageModelV2ProviderDefinedTool, type LanguageModelV2Reasoning, type LanguageModelV2ReasoningPart, type LanguageModelV2ResponseMetadata, type LanguageModelV2Source, type LanguageModelV2StreamPart, type LanguageModelV2Text, type LanguageModelV2TextPart, type LanguageModelV2ToolCall, type LanguageModelV2ToolCallPart, type LanguageModelV2ToolChoice, type LanguageModelV2ToolResultOutput, type LanguageModelV2ToolResultPart, type LanguageModelV2Usage, type LanguageModelV3, type LanguageModelV3CallOptions, type LanguageModelV3CallWarning, type LanguageModelV3Content, type LanguageModelV3DataContent, type LanguageModelV3File, type LanguageModelV3FilePart, type LanguageModelV3FinishReason, type LanguageModelV3FunctionTool, type LanguageModelV3Message, type LanguageModelV3Middleware, type LanguageModelV3Prompt, type LanguageModelV3ProviderDefinedTool, type LanguageModelV3Reasoning, type LanguageModelV3ReasoningPart, type LanguageModelV3ResponseMetadata, type LanguageModelV3Source, type LanguageModelV3StreamPart, type LanguageModelV3Text, type LanguageModelV3TextPart, type LanguageModelV3ToolCall, type LanguageModelV3ToolCallPart, type LanguageModelV3ToolChoice, type LanguageModelV3ToolResult, type LanguageModelV3ToolResultOutput, type LanguageModelV3ToolResultPart, type LanguageModelV3Usage, LoadAPIKeyError, LoadSettingError, NoContentGeneratedError, NoSuchModelError, type ProviderV2, type ProviderV3, type SharedV2Headers, type SharedV2ProviderMetadata, type SharedV2ProviderOptions, type SharedV3Headers, type SharedV3ProviderMetadata, type SharedV3ProviderOptions, type SpeechModelV2, type SpeechModelV2CallOptions, type SpeechModelV2CallWarning, type SpeechModelV3, type SpeechModelV3CallOptions, type SpeechModelV3CallWarning, TooManyEmbeddingValuesForCallError, type TranscriptionModelV2, type TranscriptionModelV2CallOptions, type TranscriptionModelV2CallWarning, type TranscriptionModelV3, type TranscriptionModelV3CallOptions, type TranscriptionModelV3CallWarning, TypeValidationError, UnsupportedFunctionalityError, getErrorMessage, isJSONArray, isJSONObject, isJSONValue };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ai-sdk/provider",
-  "version": "2.1.0-beta.4",
+  "version": "2.1.0-beta.5",
   "license": "Apache-2.0",
   "sideEffects": false,
   "main": "./dist/index.js",