> ## Documentation Index > Fetch the complete documentation index at: https://docs.sunbird.ai/llms.txt > Use this file to discover all available pages before exploring further. # Speech to Text > Speech-to-Text endpoint. Accepts an uploaded audio file. Replaces /stt, /stt_from_gcs, /org/stt, and /modal/stt. ## OpenAPI ````yaml api-reference/openapi.json POST /tasks/audio/transcriptions openapi: 3.1.0 info: title: Sunbird AI API description: > Welcome to the Sunbird AI API documentation. The Sunbird AI API provides access to Sunbird's language models and AI services for Ugandan languages. ## Supported Languages **English**, **Acholi**, **Ateso**, **Luganda**, **Lugbara**, **Runyankole**, **Swahili** and **20+** more Ugandan languages supported via sunflower. ## Getting Started You can checkout the [usage guide](https://salt.sunbird.ai/API/) for a full tutorial. For quickstart tutorials, visit our [GitHub repository](https://github.com/SunbirdAI/sunbird-ai-api/blob/main/docs/tutorial.md) ### Authentication #### Signing Up If you don't already have an account, use the `/auth/register` endpoint to create one. #### Getting an Access Token Authentication is done via a Bearer token. Use the `/auth/token` endpoint to get your access token. This token lasts for 7 days. Use the `Authorize` button below to login and access the protected endpoints. ## API Endpoints ### Translation - **`POST /tasks/translate`** - Translate text between 32 Ugandan and East African languages using the Sunflower model. Languages are accepted as ISO codes (`lug`) or full names (`Luganda`); `source_language` is optional (auto-detected when omitted). Translation works between any pair of supported languages. ### Language Detection - **`POST /tasks/language_id`** - Auto-detect the language of text input (supports Acholi, Ateso, English, Luganda, Lugbara, Runyankole) ### Speech-to-Text (STT) - **`POST /tasks/audio/transcriptions`** - STT endpoint (OpenAI-style). - Accepts an uploaded audio file (`audio`) or a GCS object (`gcs_blob_name`). - `platform`: `modal` (default, Whisper large-v3) or `runpod`. - RunPod options: `adapter` (language adapter), `whisper`, `recognise_speakers` (diarization), and the `org` organization workflow. - `language`: 3-letter code (e.g. `eng`, `lug`) or full name; improves accuracy. Supports WAV, MP3, OGG, M4A, and more. Auto-detects when omitted. - **Deprecated** → use `POST /tasks/audio/transcriptions`: - `POST /tasks/stt`, `POST /tasks/stt_from_gcs`, `POST /tasks/org/stt`, `POST /tasks/modal/stt` ### Text-to-Speech (TTS) - **`POST /tasks/audio/speech`** - single-synthesis endpoint. - `model`: `orpheus-3b-tts` (default) or `spark-tts`; `platform`: `modal` or `runpod`. - `voice`: speaker tag/name. `response_mode`: `url` (default), `stream`, or `both` (`stream`/`both` require `spark-tts` on `modal`). - Orpheus tuning: `language`, `temperature`, `top_p`, `repetition_penalty`, `max_tokens`, `seed`. Returns a signed GCP Storage URL. - **`POST /tasks/audio/speech/batch`** - Batch synthesis (orpheus-3b-tts only), 1-128 items. - **`GET /tasks/voice/speakers`** - List voices for a `model` (optional orpheus `language`). - **orpheus-3b-tts** covers 20 African languages: Acholi (`ach`), Afrikaans (`afr`), English (`eng`), Ewe (`ewe`), Fulah (`ful`), Hausa (`hau`), Igbo (`ibo`), Kikuyu (`kik`), Kinyarwanda (`kin`), Lugbara (`lgg`), Lingala (`lin`), Luganda (`lug`), Luo (`luo`), Runyankole (`nyn`), Sesotho (`sot`), Swahili (`swa`), Ateso (`teo`), Setswana (`tsn`), Xhosa (`xho`), Yoruba (`yor`). A few (`lgg`, `sot`, `tsn`) are in the training mix but expose no individual voice IDs yet. Full per-speaker catalog: see the [tutorial](https://github.com/SunbirdAI/sunbird-ai-api/blob/main/docs/tutorial.md) or call `GET /tasks/voice/speakers`. - **`GET /tasks/audio/speech/url`** - Refresh an expired signed URL for a stored audio object. - **Deprecated** → superseded by the new endpoints above: - `POST /tasks/modal/tts`, `POST /tasks/runpod/tts`, `POST /tasks/modal/orpheus/tts` → `POST /tasks/audio/speech` - `POST /tasks/modal/tts/stream` → `/tasks/audio/speech` (`response_mode=stream`) - `POST /tasks/modal/tts/stream-with-url` → `/tasks/audio/speech` (`response_mode=both`) - `POST /tasks/modal/orpheus/tts/batch` → `POST /tasks/audio/speech/batch` - `GET /tasks/modal/tts/speakers`, `GET /tasks/modal/orpheus/speakers`, `GET /tasks/modal/orpheus/speakers/{language}` → `GET /tasks/voice/speakers` - `GET /tasks/modal/tts/refresh-url` → `GET /tasks/audio/speech/url` ### Inference (Sunflower Chat) - **`POST /tasks/chat/completions`** - OpenAI-compatible chat completions (Sunflower model). Supports single instructions, multi-turn conversations, and SSE streaming (`stream: true`). Use model `Sunbird/Sunflower-14B`. - **Deprecated** → superseded by the unified endpoint above: - `POST /tasks/sunflower_inference` → `POST /tasks/chat/completions` - `POST /tasks/sunflower_simple` → `POST /tasks/chat/completions` (send the instruction as a single user message) ### File Upload - **`POST /tasks/generate-upload-url`** - Generate signed URLs for direct client uploads to GCP Storage - Supports audio files, images, and other content types - Includes path traversal protection and input validation - Returns temporary signed URL valid for 30 minutes ### WhatsApp Integration (Webhooks) - **`POST /tasks/webhook`** - Handle incoming WhatsApp Business API messages - **`GET /tasks/webhook`** - Verify webhook endpoint ownership for WhatsApp ### Legacy Endpoints - **`POST /tasks/summarise`** - *(Deprecated)* Anonymized text summarization (use Sunflower inference instead) ## Rate Limiting API endpoints are rate-limited to ensure fair usage. Authentication is required for most endpoints. version: 0.1.0 servers: - url: https://api.sunbird.ai security: [] tags: - name: Authentication Endpoints description: >- Operations for authentication, including user registration and login. Get access tokens to use protected endpoints. - name: Speech-to-Text description: >- Convert speech audio to text. The /tasks/audio/transcriptions endpoint accepts an uploaded file or a GCS object, routes to the Modal or RunPod backend, and supports optional speaker diarization for Acholi, Ateso, English, Luganda, Lugbara, and Runyankole. - name: Text-to-Speech description: >- Synthesize speech from text. Endpoints for single (/tasks/audio/speech) and batch (/tasks/audio/speech/batch) synthesis across the orpheus-3b-tts and spark-tts models, plus voice listing (/tasks/voice/speakers) and signed-URL refresh (/tasks/audio/speech/url). Returns signed GCS audio URLs. - name: Translation description: >- Translate text using the Sunflower model. Supports 32 languages (e.g. Luganda, Acholi, Ateso, Lugbara, Runyankole, Swahili, Kinyarwanda) accepted as ISO codes or full names; source language is optional and translation works between any supported pair. - name: Language description: >- Language identification and detection. Automatically detect the language of text input from supported languages. - name: Chat description: >- OpenAI-compatible chat completions powered by the Sunflower model. Supports single instructions, multi-turn conversations, and SSE streaming. - name: Upload description: >- File upload utilities. Generate signed URLs for direct client uploads to GCP Storage with security validation. - name: Webhooks description: >- WhatsApp Business API webhook integration. Handle incoming messages and verify webhook endpoints for WhatsApp chatbot functionality. paths: /tasks/audio/transcriptions: post: tags: - Speech-to-Text summary: Transcribe audio (STT endpoint) description: >- Speech-to-Text endpoint. Accepts an uploaded audio file. Replaces /stt, /stt_from_gcs, /org/stt, and /modal/stt. operationId: create_transcription_tasks_audio_transcriptions_post requestBody: content: multipart/form-data: schema: $ref: >- #/components/schemas/Body_create_transcription_tasks_audio_transcriptions_post required: true responses: '200': description: Successful Response content: application/json: schema: $ref: '#/components/schemas/STTTranscript' '422': description: Validation Error content: application/json: schema: $ref: '#/components/schemas/HTTPValidationError' security: - OAuth2PasswordBearer: [] components: schemas: Body_create_transcription_tasks_audio_transcriptions_post: properties: language: $ref: '#/components/schemas/SttbLanguage' description: Target language code. default: lug audio: type: string format: binary title: Audio description: Audio file to transcribe. type: object required: - language title: Body_create_transcription_tasks_audio_transcriptions_post STTTranscript: properties: audio_transcription: anyOf: - type: string - type: 'null' title: Audio Transcription description: The transcribed text from the audio diarization_output: anyOf: - additionalProperties: true type: object - type: 'null' title: Diarization Output description: Speaker diarization data formatted_diarization_output: anyOf: - type: string - type: 'null' title: Formatted Diarization Output description: Human-readable diarization output audio_transcription_id: anyOf: - type: integer - type: 'null' title: Audio Transcription Id description: Database ID of the saved transcription audio_url: anyOf: - type: string - type: 'null' title: Audio Url description: URL or path to the processed audio file language: anyOf: - type: string - type: 'null' title: Language description: The language code used for transcription was_audio_trimmed: anyOf: - type: boolean - type: 'null' title: Was Audio Trimmed description: Whether the audio was trimmed to max duration default: false original_duration_minutes: anyOf: - type: number - type: 'null' title: Original Duration Minutes description: Original duration in minutes if audio was trimmed type: object title: STTTranscript description: |- Response model for speech-to-text transcription results. This model represents the output of an STT transcription request, including the transcribed text, diarization data, and metadata. Attributes: audio_transcription: The transcribed text from the audio. diarization_output: Speaker diarization data as a dictionary. formatted_diarization_output: Human-readable diarization output. audio_transcription_id: Database ID of the saved transcription. audio_url: URL or path to the processed audio file. language: The language code used for transcription. was_audio_trimmed: Whether the audio was trimmed to max duration. original_duration_minutes: Original duration if audio was trimmed. HTTPValidationError: properties: detail: items: $ref: '#/components/schemas/ValidationError' type: array title: Detail type: object title: HTTPValidationError SttbLanguage: type: string enum: - ach - teo - eng - lug - lgg - nyn - swa - kin - xog - myx default: eng title: SttbLanguage description: |- Supported languages for Speech-to-Text transcription. This enum defines the language codes supported by the STT service. Each value represents an ISO 639-3 language code. Attributes: acholi: Acholi language (ach). ateso: Ateso language (teo). english: English language (eng). luganda: Luganda language (lug). lugbara: Lugbara language (lgg). runyankole: Runyankole language (nyn). swahili: Swahili language (swa). kinyarwanda: Kinyarwanda language (kin). lusoga: Lusoga language (xog). lumasaba: Lumasaba language (myx). ValidationError: properties: loc: items: anyOf: - type: string - type: integer type: array title: Location msg: type: string title: Message type: type: string title: Error Type type: object required: - loc - msg - type title: ValidationError securitySchemes: OAuth2PasswordBearer: type: oauth2 flows: password: scopes: {} tokenUrl: /auth/token ````