ElevenLabs

Positioning Overview

elevenlabs/text-to-dialogue is an interface that directly synthesizes a multi-character dialogue text into speech: the caller describes each line and its voice using an array of { text, voice }, and the model outputs the corresponding human voice per line. All lines submitted in a single request must total no more than 5000 characters. It is suitable for developers who need to produce two-person or multi-person dialogue audio in a single request, such as short-drama dubbing, podcast conversations, voice assistant sample materials, and dialogue performance for interactive content.

Core Highlights

• Assign voices line by line and synthesize multi-person dialogue in one request — each line in the dialogue array is individually bound to a voice_id, with no need to split into multiple single-person syntheses and then splice them together.
• Switch voices per line, covering more than 60 candidate voices — the voice enum in the schema provides dozens of options, and different characters in a single dialogue can use different voices.
• Stability parameter adjustable as a whole — stability supports three levels: 0, 0.5, and 1, used to trade off between emotional variation and voice consistency.
• Language can be automatically detected or specified — when language_code is left empty, it is automatically detected; an ISO language code can also be explicitly specified, covering more than 70 languages.

Capability Description

Multi-character dialogue synthesis: describe lines and voices with a dialogue array, and a single call outputs dialogue audio containing multiple characters, eliminating segmented synthesis and manual splicing.

Per-line voice binding: the voice for each line can be selected independently, which is suitable for assigning different voices to different characters while preserving their respective vocal characteristics.

Stability and language control: stability controls the magnitude of emotional variation in the delivery, while language_code determines the reading language or delegates it to automatic detection, making it convenient to batch-process materials for a fixed language.

Long dialogue text processing: all lines in a single request are limited to a total of 5000 characters, making it suitable for submitting an entire dialogue script at once without worrying about internal segmentation.

Applicable Scenarios

• Short-drama and narrative content creators: multi-character dialogue scripts — submit role-specific lines together with their respective voices to obtain a complete dialogue audio track for plot performance in short dramas, radio dramas, or interactive dramas.
• Podcast and interview content: conversation scripts — convert two-person conversation text into a dialogue track with different voices, directly usable as podcast segments or show demos.
• Product and content teams: multilingual dialogue materials — specify a language code to synthesize dialogue audio in the corresponding language for dubbing and demonstrations of overseas versions of content.
• Developers: voice dialogue prototype materials — quickly generate multi-character sample audio for prototypes such as voice assistants and customer service bots, for process validation and internal demonstrations.

Limitations and Selection Advice

The output is audio and does not produce images or video; when a finished video is needed, visual production must be completed separately. All lines in a single request must not exceed 5000 characters in total; overly long scripts need to be split into multiple calls and spliced by the caller. Voices can only be selected from the voice list provided by the interface; new voices cannot be cloned directly. If the scenario involves only a single narration or single-person reading, the single-person text-to-speech route is sufficient, and there is no need to construct a dialogue array.