> ## Documentation Index > Fetch the complete documentation index at: https://docs.rumik.ai/llms.txt > Use this file to discover all available pages before exploring further. # synthesize speech as json > synthesize an utterance and return a json object whose `audio_base64` field contains the requested audio bytes as base64. omit `audio_format` for wav, or request `opus`, `pcm`, `mulaw`, `alaw`, or `mp3`. the remaining response fields keep their existing behavior. ## OpenAPI ````yaml /openapi.json post /v1/tts/json openapi: 3.1.0 info: title: silk api version: 1.0.0 description: >- rumik ai's speech api. synthesize speech with the expressive `muga` model, the faster `mulberry` model or the multilingual `mulberry-1.6` model over http or a streaming websocket session, run conversational voice agents over webrtc, a realtime pcm socket or the phone, and manage everything the agents playground does — agents, voices, variables, tools, phone numbers, sip trunks and outbound calls — with the same api key. contact: name: rumik ai url: https://rumik.ai servers: - url: https://silk-api.rumik.ai description: production security: - bearerAuth: [] tags: - name: speech description: text-to-speech over HTTP and WebSocket. - name: voice agents description: start calls with a deployed agent from your own app. - name: agents description: build, configure, deploy and inspect your agents. - name: calls description: place outbound phone calls and read call history. - name: voices description: the voices, languages and styles an agent may use. - name: variables description: account-wide `{name}` prompt variables and their defaults. - name: tools description: HTTP tools agents call before or during a call. - name: phone numbers description: 'numbers rented through rumik: search, rent, connect, release.' - name: sip trunks description: bring your own numbers over SIP. paths: /v1/tts/json: post: tags: - speech summary: synthesize speech as json description: >- synthesize an utterance and return a json object whose `audio_base64` field contains the requested audio bytes as base64. omit `audio_format` for wav, or request `opus`, `pcm`, `mulaw`, `alaw`, or `mp3`. the remaining response fields keep their existing behavior. operationId: synthesizeSpeechJson requestBody: required: true content: application/json: schema: $ref: '#/components/schemas/TTSRequest' examples: mp3: summary: mulberry as mp3 value: model: mulberry text: hello from silk. description: a warm conversational voice audio_format: mp3 responses: '200': description: >- json response containing base64-encoded audio in the requested format. content: application/json: schema: $ref: '#/components/schemas/TtsJsonResponse' '400': $ref: '#/components/responses/BadRequest' '401': $ref: '#/components/responses/Unauthorized' '403': $ref: '#/components/responses/Forbidden' '422': $ref: '#/components/responses/ValidationError' '429': $ref: '#/components/responses/CapacityOrRateLimited' '502': $ref: '#/components/responses/GenerationFailed' '503': $ref: '#/components/responses/Unavailable' components: schemas: TTSRequest: type: object required: - text properties: text: type: string maxLength: 2000 description: >- text to synthesize. up to 2000 characters. for `muga`, prefix with a tone tag, e.g. `[happy]`. for `mulberry-1.6`, write each language in its own script, with no tags or markers. example: '[happy] namaste! kaise hain aap?' model: type: string enum: - muga - mulberry - mulberry-1.6 default: muga description: >- which model to use. `muga` is steered by a tone tag in the text; `mulberry` is steered by a natural-language description; `mulberry-1.6` speaks 22 indian languages and is steered by an optional `style, accent, pace` description. audio_format: type: string enum: - opus - pcm - mulaw - alaw - mp3 description: >- optional output format. omit the field for the original wav response. explicit `null`, `wav`, `ogg`, and unsupported values return `400 unsupported_audio_format`. description: type: string description: >- **required for `mulberry`.** natural-language voice/style description. the voice is built from this. send it on every `mulberry` request, including when `speaker` is set. **optional for `mulberry-1.6`:** a style, an accent and a pace, e.g. `happy, Tamil accent, steady pace`. a part you leave out defaults to `professional`, the accent of the text's script, and `fast pace`. not used by `muga`. example: >- a female 30s hindi voice, smooth timbre, conversational pacing, like a podcast host speaker: type: string enum: - emma - mia - sophia - ava - ira - siya - aisha - zoya - lucas - noah - theo - adam example: siya description: >- `mulberry` and `mulberry-1.6` only. optional named voice, case-insensitive. `mulberry`: `emma`, `mia`, `sophia`, `ava`, `ira`, `siya`, `aisha` and `zoya` are female; `lucas`, `noah`, `theo` and `adam` are male. omit and a voice is generated from `description`. `description` is required either way. `mulberry-1.6`: `ira`, `aisha`, `siya` or `zoya`, all female. omit it, or send a name it doesn't know, and you get `ira`. temperature: type: number default: 0.6 description: sampling temperature. `mulberry-1.6` defaults to 0.8. top_p: type: number default: 0.95 description: nucleus sampling. `mulberry-1.6` defaults to 1.0. top_k: type: integer default: 50 description: top-k sampling. `mulberry-1.6` defaults to 30. repetition_penalty: type: number default: 1.2 description: penalize repeated tokens. ignored by `mulberry-1.6`. max_new_tokens: type: integer default: 2048 maximum: 8192 description: >- output length cap. for `mulberry`, if long text comes back truncated, raise this above the default 2048 (up to 8192). for `mulberry-1.6`, it is sized from the text (at least 2048, up to 8192), and a larger value you send is kept. TtsJsonResponse: type: object required: - audio_base64 - audio_duration_ms - request_id - credits_used - usage_cost_nanos properties: audio_base64: type: string format: byte description: >- base64-encoded audio bytes in the requested format; wav when `audio_format` was omitted. audio_duration_ms: type: integer minimum: 0 description: server-derived source audio duration in integer milliseconds. request_id: type: string format: uuid description: identifier for this synthesis request. credits_used: type: integer minimum: 0 description: legacy credit usage. payg and unlimited requests return zero. usage_cost_nanos: type: integer minimum: 0 description: >- payg charge in integer nanounits of the account's billing currency. unlimited and legacy-credit requests return zero. speaker: type: string description: >- the named voice, present when the request set `speaker` on `mulberry` or `mulberry-1.6`. `mulberry-1.6` returns the voice it resolved to, e.g. `Ira`. example: Ira additionalProperties: true Error: type: object properties: error: type: string description: human-readable error message. code: type: string description: machine-readable error code. ConcurrencyLimitExceeded: type: object required: - error - code - active_requests - limit properties: error: type: string example: >- silk is already processing 2 requests for this account. try again when one finishes, or increase your plan capacity. code: type: string const: concurrency_limit_exceeded active_requests: type: integer minimum: 0 example: 2 limit: type: integer minimum: 1 example: 2 responses: BadRequest: description: malformed request body or unknown model. content: application/json: schema: $ref: '#/components/schemas/Error' Unauthorized: description: bearer token missing, invalid, or expired. content: application/json: schema: $ref: '#/components/schemas/Error' Forbidden: description: key revoked or account disabled. content: application/json: schema: $ref: '#/components/schemas/Error' ValidationError: description: validation failed. check the error field. content: application/json: schema: $ref: '#/components/schemas/Error' CapacityOrRateLimited: description: >- either a per-key rate limit or account-wide concurrent-request capacity was reached. headers: Retry-After: description: >- present for a rate-limit response; capacity failures do not include it. schema: type: integer content: application/json: schema: oneOf: - $ref: '#/components/schemas/Error' - $ref: '#/components/schemas/ConcurrencyLimitExceeded' GenerationFailed: description: >- the model could not finish the generation; nothing is charged. `GENERATION_ERROR`: the engine was unreachable or returned invalid audio, safe to retry once. `GENERATION_TOO_LONG` (`mulberry-1.6`): the text was too long for one request (about 80 seconds of audio), returned instead of cut-off audio; split the text into sentences. content: application/json: schema: $ref: '#/components/schemas/Error' Unavailable: description: upstream temporarily unavailable. retry shortly. content: application/json: schema: $ref: '#/components/schemas/Error' securitySchemes: bearerAuth: type: http scheme: bearer description: >- your rumik api key, e.g. `rk_live_...`. create one in the rumik dashboard. ````