> ## Documentation Index > Fetch the complete documentation index at: https://docs.jogg.ai/llms.txt > Use this file to discover all available pages before exploring further. # Synthesize Speech > Generate speech with an optional ElevenLabs model: Flash v2.5 or v4 Turbo. Generate speech audio from text using a public voice or a custom voice you can access. The result includes a task ID, audio URL, and a subtitle URL when available. ## Choose a model `model_id` is **optional** and applies only to ElevenLabs voices. | Value | Behavior | | - | - | | Omitted | Uses `eleven_flash_v2_5` for an ElevenLabs voice | | `eleven_flash_v2_5` | Selects Flash v2.5 | | `eleven_v4_turbo` | Selects v4 Turbo | Empty strings, whitespace, `null`, non-string values, and all other model IDs are rejected. For a voice from another provider, omit `model_id` to keep that provider's existing behavior; explicitly choosing an ElevenLabs model is rejected. Find a `voice_id` in [Get Voices](https://docs.jogg.ai/api-reference/v2/Voice/GetVoices.md) or [Get Custom Voices](https://docs.jogg.ai/api-reference/v2/Voice/GetCustomVoices.md). ## Pauses and audio tags ElevenLabs documents different text controls for these models: | Model | SSML `` | Audio tags | | - | - | - | | `eleven_flash_v2_5` | Supported: ``. Use seconds, with a maximum of 3 seconds per break. | Does not support the square-bracket audio-tag controls used by v3/v4. | | `eleven_v4_turbo` | Not supported. | Supports instructions such as `[whispers]`, `[laughs]`, `[short pause]`, `[pause]`, and `[long pause]` inside the text. | Place the control directly in `text`; no additional tag parameter is needed. Pause tags do not specify a fixed number of seconds, so they are not an exact replacement for a timed SSML break. Delivery can vary with the voice and context. ElevenLabs also warns that excessive SSML breaks can cause faster speech or audio artifacts. Jogg adapts pause syntax when sending the synthesis request: | Selected model | Input control | Sent to ElevenLabs | | - | - | - | | `eleven_v4_turbo` | Timed break greater than 0 and up to 0.5 seconds | `[short pause]` | | `eleven_v4_turbo` | Timed break over 0.5 and up to 1 second | `[pause]` | | `eleven_v4_turbo` | Timed break over 1 second, including breaks over 2 seconds | `[long pause]` | | `eleven_v4_turbo` | Zero-second break | Removed | | `eleven_flash_v2_5` | `[short pause]` | `` | | `eleven_flash_v2_5` | `[pause]` | `` | | `eleven_flash_v2_5` | `[long pause]` | `` | Supported millisecond break values are converted to seconds before selecting the v4 Turbo pause tag. These are Jogg compatibility rules; ElevenLabs does not define an exact duration for each pause tag. Changing models can therefore change the length of the pause. Flash only converts these three pause tags and preserves other bracketed text. Delivery tags such as `[whispers]` need v4 Turbo. Conversion uses a copy for synthesis and preserves your original text. Model selection, request identity, and billing continue to follow the rules on this page. New v4 Turbo subtitles omit complete square-bracket audio tags and SSML breaks. When adapting v4 pause tags for Flash, subtitles omit those controls and their converted breaks while retaining other bracketed text. These v4 compatibility paths split words at explicit pauses and preserve the returned timing gap. Native Flash SSML breaks continue to use the existing subtitle processing. The input-text fallback uses the same filtering when word timestamps are unavailable. A control-only result has no subtitle text. Existing subtitles are not rewritten. Flash v2.5 with a timed break: ```json theme={null} { "text": "I will check the schedule.The next session starts at noon.", "voice_id": "YOUR_ELEVENLABS_VOICE_ID", "model_id": "eleven_flash_v2_5", "idempotency_key": "speech-example-flash-pause" } ``` v4 Turbo with delivery and pause tags: ```json theme={null} { "text": "[whispers] The room is quiet. [short pause] You can begin.", "voice_id": "YOUR_ELEVENLABS_VOICE_ID", "model_id": "eleven_v4_turbo", "idempotency_key": "speech-example-turbo-tags" } ``` Official ElevenLabs references (checked November 5, 2026): * [Pause syntax, supported models, and the 3-second limit](https://elevenlabs.io/docs/help-center/product/core-capabilities/text-to-speech/how-can-i-add-pauses). * [Audio tags, including explicit v4 Turbo support](https://elevenlabs.io/docs/help-center/product/core-capabilities/text-to-speech/how-do-audio-tags-work-with-eleven-v3-and-v4). * [Prompting guidance and pause-tag examples](https://elevenlabs.io/docs/overview/capabilities/text-to-speech/best-practices#prompting-eleven-v4). * [Official SSML guide: audio-tag compatibility FAQ](https://elevenlabs.io/blog/ssml-tags-explained). ## Example: select v4 Turbo ```bash theme={null} curl --request POST \ --url 'https://api.jogg.ai/v2/audio/tts/synthesize' \ --header 'x-api-key: YOUR_API_KEY' \ --header 'Content-Type: application/json' \ --data '{ "text": "Hello from JoggAI.", "voice_id": "YOUR_ELEVENLABS_VOICE_ID", "model_id": "eleven_v4_turbo", "idempotency_key": "speech-example-turbo" }' ``` To use the default Flash v2.5 model, omit `model_id`: ```json theme={null} { "text": "Hello from JoggAI.", "voice_id": "YOUR_ELEVENLABS_VOICE_ID", "idempotency_key": "speech-example-flash" } ``` ## Retry the same request Keep the returned `task_id`. Use the same optional `idempotency_key` only when retrying the same request. A key already used with a different explicit model causes a conflict; use a new key when intentionally changing the model or text. Synthesis consumes API credits based on the generated audio duration. After a timeout, check the existing [speech task](https://docs.jogg.ai/api-reference/v2/Voice/GetSpeechTask.md) when its ID is known, or retry the unchanged request with its original key. Avoid creating another paid request just because the connection timed out. ## Use the same selection through MCP The `jogg_post_audio_tts_synthesize` tool accepts the same optional field under `body.model_id`. See [MCP speech model selection](https://docs.jogg.ai/mcp/tools-billing-safety#speech-model-selection) for a `tools/call` example. ## OpenAPI ````yaml POST /v2/audio/tts/synthesize openapi: 3.0.1 info: title: JoggAI OpenAPI v2 description: > JoggAI OpenAPI v2 provides a comprehensive suite of endpoints for creating and managing AI-powered video content. ## Authentication All API requests require authentication using an API key provided in the `x-api-key` header. ## Base URL Production: https://api.jogg.ai/v2 ## Rate Limiting API calls are subject to rate limiting based on your subscription plan. ## Error Handling All endpoints return standard HTTP status codes along with detailed error messages in the response body. version: 2.0.0 license: name: MIT servers: - url: https://api.jogg.ai description: Production server security: - apiKeyAuth: [] tags: - name: Video description: Video creation and management operations - name: Avatar description: Avatar management and creation operations - name: Voice description: Voice and timbre management operations - name: Asset description: Asset management including music, scripts, and uploads - name: Template description: Template management operations - name: Product description: Product management operations - name: User description: User account operations - name: Webhook description: Webhook endpoint management for event notifications paths: /v2/audio/tts/synthesize: post: tags: - Voice summary: Synthesize speech description: > Generate speech audio and an optional subtitle file from text. For ElevenLabs voices, model_id optionally selects eleven_flash_v2_5 or eleven_v4_turbo. Omitting model_id uses eleven_flash_v2_5 for ElevenLabs voices and preserves existing behavior for other providers. Explicit model_id values are only supported for ElevenLabs voices. Empty strings, null, non-string values, and unsupported model IDs are rejected. Reuse the same idempotency_key only for the same request; changing an explicit model with the same key causes a conflict. Flash v2.5 supports SSML `` up to 3 seconds and does not support audio tags. v4 Turbo supports audio tags such as `[whispers]`, `[short pause]`, `[pause]`, and `[long pause]`, but does not support SSML ``. Pause tags do not specify an exact duration. Put controls in text; no additional tag parameter is needed. When sending a v4 Turbo synthesis request, Jogg converts positive timed breaks of up to 0.5 seconds to `[short pause]`, over 0.5 and up to 1 second to `[pause]`, and over 1 second to `[long pause]`; zero-second breaks are removed. Supported millisecond inputs are converted to seconds first. For Flash v2.5, Jogg converts `[short pause]`, `[pause]`, and `[long pause]` to SSML breaks of 0.3, 0.75, and 1.5 seconds respectively. These are Jogg compatibility rules, not exact duration equivalences specified by ElevenLabs. Other bracketed text is retained for Flash. Conversion uses a synthesis request copy and preserves the original text and existing request identity and billing rules. Newly generated v4 Turbo subtitles omit audio tags and break tags. When adapting v4 pause tags for Flash, subtitles omit those controls and their converted breaks while retaining other bracketed text. These v4 compatibility paths split words at explicit pauses and retain provider timing gaps. Native Flash breaks continue to use the existing subtitle processing. Input-text subtitle fallback uses model-specific filtering. See ElevenLabs' [pause guide](https://elevenlabs.io/docs/help-center/product/core-capabilities/text-to-speech/how-can-i-add-pauses) and [audio-tag guide](https://elevenlabs.io/docs/help-center/product/core-capabilities/text-to-speech/how-do-audio-tags-work-with-eleven-v3-and-v4). operationId: jogg_post_audio_tts_synthesize requestBody: required: true content: application/json: schema: type: object required: - text - voice_id properties: text: type: string minLength: 1 description: Text to synthesize. voice_id: type: string minLength: 1 description: Public voice or an authorized custom voice ID. model_id: type: string description: >- ElevenLabs-only speech model. Omit to use eleven_flash_v2_5 for ElevenLabs voices and retain the voice provider's existing behavior otherwise. Explicit values for other providers, empty strings, null, and unsupported model IDs are rejected. enum: - eleven_flash_v2_5 - eleven_v4_turbo default: eleven_flash_v2_5 idempotency_key: type: string maxLength: 255 description: >- Stable key for retries of this request. Use a new key for a different model or text. examples: defaultModel: summary: Omit the model to use Flash v2.5 for an ElevenLabs voice value: text: Hello from JoggAI. voice_id: YOUR_ELEVENLABS_VOICE_ID idempotency_key: speech-example-flash turboModel: summary: Select v4 Turbo for an ElevenLabs voice value: text: Hello from JoggAI. voice_id: YOUR_ELEVENLABS_VOICE_ID model_id: eleven_v4_turbo idempotency_key: speech-example-turbo flashPause: summary: Flash v2.5 with a timed SSML pause value: text: >- I will check the schedule.The next session starts at noon. voice_id: YOUR_ELEVENLABS_VOICE_ID model_id: eleven_flash_v2_5 idempotency_key: speech-example-flash-pause turboAudioTags: summary: v4 Turbo with audio tags value: text: '[whispers] The room is quiet. [short pause] You can begin.' voice_id: YOUR_ELEVENLABS_VOICE_ID model_id: eleven_v4_turbo idempotency_key: speech-example-turbo-tags responses: '200': description: Speech task result. Check the business code before reading data. content: application/json: schema: $ref: '#/components/schemas/SpeechTaskResponse' components: schemas: SpeechTaskResponse: allOf: - $ref: '#/components/schemas/BaseResponse' - type: object properties: data: $ref: '#/components/schemas/SpeechTask' BaseResponse: type: object required: - code - msg properties: code: type: integer description: | Business status code: * 0 - Success * 10104 - Record not found * 10105 - Invalid API key * 18020 - Insufficient credit * 18025 - No permission to call APIs * 40000 - Parameter error * 50000 - System error example: 0 msg: type: string description: Response message example: Success SpeechTask: type: object required: - task_id - status - progress - url - subtitle_url - duration_seconds properties: task_id: type: string description: Speech task ID to keep for status queries and retries. example: tts_123 status: type: string enum: - processing - succeeded - failed example: succeeded progress: type: integer minimum: 0 maximum: 100 example: 100 url: type: string description: Generated audio URL when the task succeeds; otherwise empty. subtitle_url: type: string description: Subtitle file URL when available; otherwise empty. duration_seconds: type: number description: Generated audio duration in seconds. example: 2.5 error_message: type: string description: Failure details when the task fails. securitySchemes: apiKeyAuth: type: apiKey in: header name: x-api-key description: API key for authentication. Obtain your key from the JoggAI dashboard. ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.