ElevenLabs API

Converts text input (up to 5,000 characters) into natural-sounding speech using the Eleven v3 model, which employs neural vocoding and prosody modeling to generate dramatic, emotionally-expressive audio with support for multiple speaker voices in single dialogue passages. The model handles complex linguistic nuances across 70+ languages and supports streaming output for real-time audio delivery without waiting for full synthesis completion.

Synthesizes speech from text (up to 40,000 characters) using the Eleven Flash v2.5 model, optimized for sub-100ms latency (~75ms excluding network overhead) and 50% lower per-character cost compared to standard models. The model trades some expressiveness for speed and cost efficiency, making it suitable for real-time conversational AI, live streaming, and cost-sensitive applications at scale.

Aligns text transcripts to audio recordings at word-level granularity, producing precise timestamps for each word's start and end times. The alignment system uses acoustic-linguistic models to match text to audio despite pronunciation variations, accents, and speech rate variations, enabling accurate temporal mapping for subtitle generation, audio editing, and downstream NLP tasks requiring precise text-audio synchronization.

Isolates foreground speech from background noise, music, and other audio sources using neural source separation models. The voice isolator analyzes audio spectrograms and applies learned masks to separate speech from non-speech components, producing clean voice-only audio suitable for transcription, re-synthesis, or further processing. Enables high-quality speech extraction from noisy recordings without manual editing.

Modifies voice characteristics (pitch, speed, tone, accent) of existing audio recordings through neural voice transformation, enabling voice customization without re-recording or voice cloning. The voice changer applies learned transformations to match target voice characteristics while preserving original speech content and intelligibility, suitable for accessibility adjustments, creative effects, and voice personalization.

Implements a credit-based pricing model where each API operation consumes credits based on input size and operation type (1 character = 1 credit for standard TTS, 0.5-1 credit per character for Flash models depending on tier). Credits are allocated monthly per subscription tier (10k-6M credits/month), with unused credits rolling over for up to 2 months, enabling cost predictability and budget management. Developers can monitor credit consumption per request and optimize usage patterns to reduce costs.

Maintains a persistent voice library where cloned voices, designed voices, and pre-built voices are stored as reusable profiles with unique identifiers. Developers can create, organize, and manage voice profiles across projects, enabling consistent voice usage across multiple synthesis requests without re-cloning or re-designing. Voice profiles support metadata tagging and organization, facilitating voice discovery and reuse at scale.

Generates speech and text content across 29-90+ languages depending on operation (TTS supports 29-70+ languages, STT supports 90+ languages), with automatic language detection for input content. The system automatically selects appropriate language-specific models and processing pipelines based on detected language, enabling seamless multilingual workflows without explicit language specification. Supports language mixing in some contexts (e.g., code-switching in dialogue).

Synthesizes speech from text (up to 10,000 characters) using the Eleven Multilingual v2 model, optimized for consistent, natural-sounding output across 29 languages with stable prosody and pronunciation accuracy for long-form content like audiobooks and documentation. The model uses language-specific phoneme processing and cross-lingual prosody modeling to maintain voice consistency across language boundaries.

Clones a speaker's voice from a short audio sample (requirements unknown) and generates speech in that voice using the cloned voice profile. The cloning process analyzes acoustic features (pitch, timbre, speaking rate) from the sample and creates a reusable voice model that can be applied to any text input. Instant cloning is available at Starter tier and above, enabling rapid voice customization without professional recording sessions.

Creates high-quality voice clones from longer audio samples using professional-grade voice modeling, available at Creator tier ($11/month) and above. The professional cloning process uses more sophisticated acoustic analysis and voice profile training to produce clones with higher fidelity, better emotional consistency, and improved handling of edge cases compared to instant cloning. Cloned voices are stored as reusable profiles in the user's voice library.

Generates synthetic voices from text descriptions (e.g., 'warm, friendly, slightly accented British English speaker') without requiring audio samples, using a neural voice synthesis model that maps text descriptions to acoustic parameters. The generated voices are stored as reusable profiles and can be applied to any text-to-speech synthesis request, enabling rapid voice experimentation and customization without voice talent or recording equipment.

Transcribes audio (90+ languages) to text using the Scribe v2 model, which combines automatic speech recognition with optional keyterm prompting (up to 1,000 custom terms), entity detection (56 entity types), and speaker diarization (up to 32 speakers). The model produces word-level timestamps, dynamic audio tagging, and automatic language detection, enabling structured extraction of named entities, speaker identification, and precise temporal alignment for downstream processing.

Transcribes audio streams in real-time using Scribe v2 Realtime model, achieving ~150ms latency (excluding network overhead) with word-level timestamps and support for 90+ languages. The model processes audio chunks incrementally and returns partial transcriptions as they become available, enabling live captioning, real-time meeting transcription, and interactive voice applications without waiting for full audio processing.

Automatically dubs video content by extracting dialogue, translating to target languages, and synthesizing speech in the target language while preserving original speaker characteristics and lip-sync timing. The dubbing system uses forced alignment to match synthesized speech duration to original dialogue timing, enabling seamless multilingual video distribution without manual dubbing or voice talent hiring. Available through both API and Dubbing Studio UI.

Transforms and enhances existing voice recordings by applying voice characteristics from reference speakers or voice profiles, enabling voice style transfer without re-recording. The remixing system analyzes acoustic features from source and target voices and applies transformations (pitch, timbre, speaking rate) to match target characteristics while preserving original content intelligibility. Enables rapid voice customization and speaker style adaptation.