Enter the text content for this dialogue segment.
Select the voice character for this dialogue.
Single speaker
Xavier: [calm] Welcome to the AI studio, where photos come to life with AI Avatar Lip Sync. [excited] Upload an image and an audio file, then watch your avatar speak naturally.
Multi-speaker dialogue
Juniper: [excitedly] Hey James! Have you tried the new ElevenLabs V3?
James: [curiously] Yeah, just got it! The emotion is so amazing. I can actually do whispers now— [whispering] like this!
AI Text to Speech — Multi-Speaker Dialogue with Audio Tag Control
Single-voice TTS with a speed slider is a solved problem. This tool addresses a harder one: producing dialogue audio where multiple speakers interact naturally, each with distinct voice character, and each line shaped by inline Audio Tags that control emotion, delivery style, non-verbal sounds, ambient sound effects, accent, and pacing — mid-sentence if needed. Built on ElevenLabs' text-to-dialogue-v3 model, it processes multi-speaker scripts in a single generation request, outputting one audio file with natural conversational turn-taking. Choose from 113 preset voices with in-browser MP3 preview, select from 75 languages or let the engine auto-detect, and set the Stability parameter (Creative, Natural, or Robust) to control how much expressive variation the voice introduces. The output MP3 feeds directly into the AI Avatar Lip Sync tool on Bernini AI, completing a full script-to-talking-video pipeline.
What Is Multi-Speaker AI Text to Speech?
AI text to speech uses neural voice synthesis to convert written text into natural-sounding speech. ElevenLabs' text-to-dialogue-v3 engine, which powers this tool, models prosody at the phoneme level — shaping pitch contour, stress placement, inter-word timing, and pause duration based on semantic content. The distinction from older TTS systems is not just audio quality: it's the ability to accept structural instructions inline via Audio Tags, and to handle multiple speakers within a single generation request without requiring separate API calls per speaker or manual audio splicing afterward.
The multi-speaker dialogue feature is the primary differentiator from standard TTS tools. Each line of your script gets its own voice assignment; the engine generates a single audio file with natural timing and pacing between speaker turns. Layer in Audio Tags across six categories — emotion, delivery, non-verbal, sound effect, accent, and pacing — and you specify not just what a voice says but precisely how it says it. The output works as a standalone downloadable MP3 or as the audio input to AI Avatar Lip Sync, which maps the audio's phoneme timing to mouth shapes and facial motion on any uploaded portrait photo.
Key Features
ElevenLabs text-to-dialogue-v3 with multi-speaker support, Audio Tags, 113 voices, and 75-language coverage.
Multi-Speaker Dialogue in One Request
Assign a distinct voice to each line of dialogue and submit the entire script in a single generation. The engine handles conversational turn-taking, inter-speaker pacing, and per-line Audio Tag interpretation. Podcasts, game cutscenes, training dialogues, and interview scripts generate as complete audio files — no manual splicing of separately generated clips required.
Inline Audio Tags for Emotional Control
Insert bracketed tags directly into your script text to control delivery at the phrase level. [excited] before a line raises pitch and pace; [whispering] drops volume and reduces breath noise; [sigh] inserts a natural non-verbal before the spoken words begin. Tags are processed during waveform synthesis — not applied as post-processing — so the resulting prosody is organic rather than artificial. All tags work across all voices and all languages.
113 Preset Voices with In-Browser Preview
Browse voices organized by character type — conversational, storytelling, video games, TikTok, Hollywood, announcers, relaxing, and more. Every voice has a cloud-hosted MP3 preview playable in-browser before committing to generation. Voices vary in pitch range, speaking rate, accent, and emotional expressiveness. Combine voice selection with the Stability parameter for fine control over consistency versus variation.
75 Languages with Auto-Detection
Generate speech in English, Mandarin, Japanese, Korean, French, German, Spanish, Portuguese, Arabic, Hindi, Russian, and dozens more — 75 total including an Auto-detect option. Auto-detect identifies the language from your script text without manual selection. Manual language selection is available for mixed-script content or when a specific regional pronunciation is required.
Stability Parameter: Creative, Natural, Robust
The Stability slider has three positions. Creative (0) produces the most expressive, varied output — pitch shifts, emphasis changes, and emotional inflections are pronounced, suited to dramatic content and character dialogue. Natural (0.5, the default) balances expressiveness with consistency, appropriate for podcasts, marketing voiceovers, and general narration. Robust (1) produces the most uniform, predictable output across multiple generations of the same text — essential for e-learning narration and any content where tonal consistency across long scripts is required.
Direct Integration with AI Avatar Lip Sync
The generated MP3 is format-compatible with the AI Avatar Lip Sync tool. Download the audio, upload it alongside a portrait photo in the Avatar tool, and produce a talking head video where the face appears to speak your script. This creates a complete text-to-talking-video pipeline — script, voice, video — without a microphone, camera, recording studio, or voice actor booking.
Audio Tags Reference
Six categories of inline markers that shape how each phrase is delivered.
Audio Tags are plain-text brackets inserted into your dialogue script that instruct the synthesis engine on delivery style, emotional tone, non-verbal sounds, ambient audio, accent, and timing. Place a tag at the start of a dialogue line to set the overall register for that speaker turn, or place it mid-sentence to trigger a shift at a specific word. Tags are independent per line — one speaker can be [whispering] while the next is [shouting] within the same generation. Every tag is compatible with all 113 voices and all 75 supported languages.
Emotion
Controls the underlying emotional register of the voice — affects pitch contour, speaking rate, and breath pattern simultaneously.
[excited] We just hit our launch target! [sad] The results didn't go our way this quarter.
Delivery Style
Controls how the voice physically produces sound — volume level, vocal placement, and articulatory style. Useful for dramatic contrast between lines.
[whispering] Don't let anyone hear this. [shouting] Everyone needs to know right now!
Non-Verbal Sounds
Inserts involuntary or reflexive vocalizations that make dialogue feel unrehearsed and natural — pauses, reactions, and transitions between ideas.
[sigh] I suppose we have no other option. [gasp] You actually pulled it off.
Sound Effects
Embeds ambient or diegetic audio cues directly into the speech output — no separate sound design layer required for short-form content.
[rain] The weather report says conditions worsen overnight. [door knocking] Someone's at the entrance.
Accent
Shifts the phonemic character of the selected voice toward a regional accent without changing the underlying voice identity. Useful for localized content or character differentiation.
[British accent] The meeting's set for half three. [Australian accent] No worries, we'll sort it.
Pacing
Alters the temporal delivery of a phrase — useful for building tension, emphasizing importance, or matching audio to a visual cut point.
[dramatically] The decision rests with one person. [with a pause] And that person is here today.
The TTS-to-Video Pipeline
Script to audio to talking video — no microphone, no camera, no recording setup.
Text to Speech is the first stage in a production pipeline that ends with a lip-synced talking head video. Write a multi-speaker script in the dialogue editor, assign voices from the 113-preset library, insert Audio Tags at emotional beats, and generate the audio. Download the MP3, then upload it alongside a portrait photo in the AI Avatar Lip Sync tool. The lip sync engine maps the audio's phoneme timing to mouth shapes, head movement, and facial expression on the portrait — producing a complete video from text alone, with no recording equipment at any stage.
Write Your Script with Audio Tags
Enter dialogue in the editor, one line per speaker. Assign a voice from the 113-preset library to each line. Insert Audio Tags at emotional beats or delivery transitions. The engine supports up to 5,000 total characters across all dialogue lines in a single generation.
Generate and Download the Audio
Select a language (or use Auto-detect) and choose a Stability setting. Click generate. Processing typically takes seconds to a few minutes depending on total character count. Download the finished MP3 when complete.
Feed into AI Avatar for Lip-Sync Video
Upload the downloaded MP3 alongside a portrait photo in the AI Avatar Lip Sync tool. The lip sync engine maps audio phoneme timing to mouth shapes and facial motion frame by frame, producing a talking head video from the photo and audio alone.
How to Use AI Text to Speech
Three steps from blank script to downloadable audio — all in-browser, no software installation.
1. Write and Tag Your Dialogue
Enter your script in the dialogue editor. Each line represents one speaker turn. Insert Audio Tags like [excited], [whispering], or [sigh] directly into the text at the points where they should take effect. Keep individual lines under 500 characters for optimal prosody within each turn. Total across all lines must not exceed 5,000 characters.
2. Assign Voices and Set Parameters
Open the voice selector for each dialogue line and preview voices in-browser using the cloud-hosted MP3 samples. Assign the voice that fits the character. Set the language — or leave it on Auto-detect. Choose Stability: Creative for dramatic variation, Natural for balanced delivery, Robust for consistent tone across long scripts.
3. Generate and Download
Click Generate Speech. The ElevenLabs text-to-dialogue-v3 engine processes your script and returns a single MP3 file containing all speaker turns with natural conversational timing. Download the file directly or pipe it into AI Avatar Lip Sync for a talking head video.
Text to Speech Use Cases
Multi-speaker dialogue and Audio Tag control open production workflows that single-voice TTS cannot address.
Podcast and Interview Dialogue
Generate dialogue tracks that can guide Seedance scenes.
Assign host and guest voices to alternating dialogue lines, tag natural reactions ([laugh], [gasp], [hmm]), and generate a complete conversational audio track in a single request. A 3,000-character host-guest exchange generates in seconds — revise the script and regenerate without rebooking a co-host or re-recording a session.
Accessibility and Screen Reader Content
Create narration references for visual storytelling.
Generate naturally paced audio narration for written documents, product descriptions, and web content that serves users who consume information through audio. The 75-language library ensures localized audio accessibility for global audiences. Stability at Robust maintains consistent voice character across long-form narration without unexpected pitch variations.
Game Cutscene and Character Voice Prototyping
Prototype character audio before multimodal video tests.
Script a full cutscene dialogue with multiple character voices, assign voices with appropriate dramatic character, add [shouting] battle lines and [whispering] conspiracies, and generate the audio for director review before committing to live voice actor recording sessions. Iterate on dialogue pacing and Audio Tag selections based on what the audio actually sounds like, not what looks right on the page.
E-Learning and Course Narration
Build lesson voice tracks for Seedance-ready modules.
Generate consistent narration across 75 languages from a single master script — translate the text, select the appropriate voice, and regenerate. Set Stability to Robust to ensure tonal consistency across multi-lesson courses. Pair each audio track with AI Avatar Lip Sync to produce on-screen instructor videos that speak every required language.
Voiceover A/B Testing at Scale
Test product voiceovers before using audio references.
Produce five variants of the same ad voiceover — different voices, different Audio Tags, different Stability settings — in under 10 minutes total. Test [excited] versus [calm] delivery, male versus female voice character, or fast versus measured pace against engagement metrics, without rebooking voice talent for each take.
Video and Presentation Voiceover Drafts
Create narration hooks for short Seedance videos.
Generate rough-cut voiceovers for video edits, explainer animations, and presentations before final production decisions are made. Hearing the script spoken reveals pacing problems, awkward phrasing, and tonal mismatches that reading it silently does not. Replace the draft voiceover with live-recorded audio at the end, or keep the AI-generated version if it meets quality requirements.
Best Practices
Script Writing Tips
- Write as spoken language, not formal prose — contractions, sentence fragments, and informal phrasing produce more natural synthesis than grammatically perfect text
- Keep individual dialogue lines under 500 characters — the engine optimizes prosody per segment; very long lines can produce uneven stress and pacing
- Use punctuation deliberately: commas produce brief pauses, em dashes signal abrupt breaks, and ellipses trail off — these timing cues are read literally by the synthesis engine
- Spell out numbers and abbreviations in full: 'forty-two' not '42', 'doctor' not 'Dr.' — the engine may mispronounce abbreviated forms or read digit characters individually
Audio Tag Usage Tips
- Tag key emotional moments rather than every line — over-tagging flattens the contrast that makes tagged moments feel significant
- Stack complementary tags to shape nuanced delivery: [excited] followed by [quickly] in the same line creates urgency with upward energy
- Place non-verbal tags ([sigh], [gasp], [laugh]) at the very start of a line — inserting them mid-sentence interrupts the speech rhythm more than intended
- Test one line with three different emotion tags at Stability 0.5 before choosing — the gap between [sad] and [serious] is wider than it appears on paper
Technical Specifications
AI Engine
- Engine: ElevenLabs text-to-dialogue-v3
- Voice library: 113 preset voices with cloud MP3 preview
- Stability: Creative (0) / Natural (0.5, default) / Robust (1)
Input
- Max characters: 5,000 per generation across all dialogue lines combined
- Speakers: unlimited lines per request, each line assigned its own voice
- Languages: 75 supported including Auto-detect
- Audio Tags: 6 categories — inline bracketed markers embedded directly in script text
Output
- Format: MP3 audio file
- Processing time: seconds to minutes depending on total character count
- Compatible with AI Avatar Lip Sync tool as direct audio input
Related Tools
Text to Speech FAQ
Specific answers about Audio Tags, voice selection, multi-speaker output, and the TTS-to-Avatar pipeline.
Write the Script. Assign the Voices. Hear It.
Type a multi-speaker dialogue, insert Audio Tags for emotional control, choose from 113 voices across 75 languages, and generate a single MP3 — then feed it into AI Avatar Lip Sync to produce a talking head video with no microphone or camera.