# OpenSpeaker generation API guide

API base URL: https://api.openspeaker.ai
Authentication: send the account API key in the xi-api-key header. V3 endpoints also document raw Authorization header support.
Never expose a real key in browser code, a public repository or prompts sent to third parties.
Last source verification: 2026-07-13

## Asynchronous task flow

1. POST to a generation endpoint. Add receive_url when webhook delivery is preferred.
2. Read task_id from the create response.
3. Poll GET /v1/task/{task_id}, or wait for the receive_url webhook.
4. List history with GET /v1/tasks.

## Common endpoints

- GET /v1/task/{task_id} — poll one task
- GET /v1/tasks — list and filter task history
- POST /v1/task/delete — delete a task; refund where available
- GET /v1/credits — current credit balance
- GET /v1/health-check — API availability
- GET /v3/voices — search the unified voice library

## Active generation workflows

## Text to Speech
Turn scripts and long-form text into natural speech. Process up to 1,000,000 characters per task with voice sources from ElevenLabs, MiniMax, Vbee, Fish Audio or low-cost Microsoft Edge voices.
Application: https://openspeaker.ai/app
Public page: https://openspeaker.ai/services/text-to-speech
Endpoints:
- POST /v3/text-to-speech
Verified specifications:
- Input: Text, SRT, TXT, ZIP or folder batch
- Task limit: 1,000,000 characters; SRT timeline up to 5 hours
- Voice controls: Speed 0.5–1.5, pronunciation rules and AI expression tags
- Quality tiers: High quality or Low cost (50% fewer credits in the current UI)
- Low-cost voice library: 318 Microsoft Edge voices across 140 locale codes

## Text to Dialogue
Build multi-speaker audio from written dialogue. Assign a different voice to each speaker and control pauses between lines.
Application: https://openspeaker.ai/app/dialogue
Public page: https://openspeaker.ai/services/text-to-dialogue
Endpoints:
- POST /v3/text-to-speech/dialogue
Verified specifications:
- Speakers: 2–26 speakers labeled A> through Z>
- Timing: Per-speaker speed 0.5–1.5; segment delay 0–5 seconds
- Voice mix: Different connected voice source for each character
- Extras: Pronunciation dictionary, AI expression tags and optional transcript

## Pronunciation Dictionary
Create reusable replacement rules before generating speech or dialogue, so important names and specialist terms are spoken consistently.
Application: https://openspeaker.ai/app/pronunciation-dictionary
Public page: https://openspeaker.ai/services/pronunciation-dictionary
Endpoints:
- GET /v3/dictionaries
- POST /v3/dictionaries
- POST /v3/dictionaries/preview
Verified specifications:
- Capacity: Up to 500 replacement rules per dictionary
- Matching: Whole-word or contains, with optional case sensitivity
- Applies to: Text to Speech and Text to Dialogue
- Behavior: Replaces spoken text before synthesis; submitted text remains unchanged

## Clone Voice
Clone a reusable voice from your own recording or properly consented reference audio, then use it in the unified speech workspace.
Application: https://openspeaker.ai/app/clone-voices
Public page: https://openspeaker.ai/services/voice-cloning
Endpoints:
- POST /v3/text-to-speech/voice-clone
Verified specifications:
- Sample length: 3–30 seconds
- Upload limit: 10 MB per reference recording
- Audio formats: MP3, WAV, M4A, AAC, OGG, WebM, FLAC or Opus
- Cleanup: Optional background-noise removal before cloning

## AI Audio Dubbing
Upload MP3 or M4A audio, choose a target language and create a dubbed version through the ElevenLabs dubbing workflow.
Application: https://openspeaker.ai/app/dubbing
Public page: https://openspeaker.ai/services/audio-dubbing
Endpoints:
- POST /v1/task/dubbing
Verified specifications:
- Input: MP3 or M4A audio; effective direct-request cap 200 MB
- Duration: Up to 5 hours per file
- Languages: Auto-detect source and select one or more of 30 target languages
- Speakers and output: Detect or choose 1–9 speakers; audio and SRT output

## Voice Changer
Transform recorded speech into another voice with ElevenLabs speech-to-speech while retaining the original performance and pacing.
Application: https://openspeaker.ai/app/speech-to-speech
Public page: https://openspeaker.ai/services/voice-changer
Endpoints:
- POST /v1/task/voice-changer
Verified specifications:
- Input: MP3, M4A or WAV; effective direct-request cap 200 MB
- Duration: Up to 5 hours per file
- Models: Eleven Multilingual v2 or English v2
- Controls: Stability, similarity, style, speaker boost, noise removal and transcript

## Voice Isolation
Clean up spoken audio by isolating the voice from background noise through ElevenLabs Audio Isolation.
Application: https://openspeaker.ai/app/voice-isolate
Public page: https://openspeaker.ai/services/voice-isolation
Endpoints:
- POST /v1/task/voice-isolate
Verified specifications:
- Input: MP3, M4A or WAV; effective direct-request cap 200 MB
- Length: 5 seconds to 5 hours per file
- Batching: Multiple files processed sequentially
- Output: Clean speech-focused audio with task history

## Speech to Text
Transcribe audio with ElevenLabs Scribe v2 and export readable text, SRT subtitles or structured timing data.
Application: https://openspeaker.ai/app/speech-to-text
Public page: https://openspeaker.ai/services/speech-to-text
Endpoints:
- POST /v1/task/speech-to-text
Verified specifications:
- Upload limit: 200 MB per audio file
- Formats: AAC, AIFF, OGG, MP3, Opus, WAV, WebM, FLAC and M4A
- Outputs: SRT segments plus JSON segment and word timing
- Detection: Optional audio-event tagging; multi-file batch and ZIP download

## Sound Effects
Describe the sound you need and generate a custom effect with ElevenLabs Text to Sound v2.
Application: https://openspeaker.ai/app/sound-effect
Public page: https://openspeaker.ai/services/sound-effects
Endpoints:
- POST /v1/task/sound-effect
Verified specifications:
- Prompt: 3–450 characters
- Duration: Automatic or 0.5–30 seconds
- Variants: 1–4 effects per task
- Controls: Prompt influence 0–100% and seamless loop

## AI Music Generation
Turn a song idea or lyrics into a two-clip generation request with the active Suno v4.5-all workflow. A partial clip failure can occur.
Application: https://openspeaker.ai/app/music2
Public page: https://openspeaker.ai/services/music-generation
Endpoints:
- POST /v1s/task/music-generation
Verified specifications:
- Model: Suno v4.5-all
- Simple mode: 1–500 character description with instrumental option
- Custom mode: Lyrics to 5,000; styles to 1,000; title to 80 characters
- Output: Requests up to two MP3 clips; partial failure may return fewer

## AI Image Generation
Create and edit AI images with 20 active models from OpenAI, Google, ByteDance, Krea, Recraft, Kling, Black Forest Labs, Runway and Wan. Use up to 14 reference images and output up to 4K on supported models.
Application: https://openspeaker.ai/app/imagen2
Public page: https://openspeaker.ai/services/image-generation
Endpoints:
- GET /v1i/models
- POST /v1i/task/price
- POST /v1i/task/generate-image
Verified specifications:
- Model catalog: 20 active models across nine provider families
- Outputs: Up to four images per task and up to 4K where supported
- References: Up to 14 images, 5 MB each, addressed with @img tags
- Prompt and cost: Model-dependent prompts up to 10,000 characters; live credit quote

## Explicit exclusions

Kokoro, Kiki, legacy MiniMax Music, legacy Google Imagen and Video/Veo are not part of the current active public catalog.

## Human and machine documentation

- https://openspeaker.ai/developers
- https://openspeaker.ai/vi/lap-trinh-vien
- https://openspeaker.ai/openapi-generation.json
- https://openspeaker.ai/api/catalog.json
- https://openspeaker.ai/app/api-document