Create speech, transform voices, transcribe and dub media, generate sound, and build interactive agents with ElevenLabs.
ElevenLabs
Explore features, practical uses and pricing below.
ElevenLabs provides tools for creating speech, working with recorded voices, transcribing and dubbing media, generating sound and music, and building conversational agents. Its creative workspace can serve a short voiceover or a longer narrated project, while its APIs support integrating the underlying capabilities into an application. The appropriate route depends on whether you need a finished media asset or an interactive system that speaks with users.
It is useful for video producers, audiobook and podcast creators, teams developing narrated lessons, localization workflows, and developers building voice experiences. A voice can be selected from the library, designed from a description, or created through an eligible cloning workflow. The sound still needs to fit the words, audience, and visual timing. A smooth delivery does not establish that the script or translation is correct.
The text-to-speech tools turn written text into audio using supported voices and models. Model choice affects language support, expression, response behavior, and available controls. The creative interface and API provide different ways to use the same general capability, so a production workflow should be planned around the selected model and output requirements.
Write for listening rather than merely reading. A paragraph that is easy to scan can be difficult to follow aloud when it contains several nested clauses. Break an instruction into clear steps, spell out ambiguous numbers, and decide where a listener needs a pause. For a tutorial, align each sentence with the action it describes instead of generating a long narration and forcing the video to fit afterward.
Preview a representative section containing the difficult terms before producing the full script. Listen to the voice's tone and pacing as well as pronunciation. A dramatic voice may suit a story but make a simple customer instruction feel unnecessarily urgent. Save the approved script and voice choice together so a later update can maintain the intended presentation.
The Voice Library provides browseable voices with samples and filters such as language, accent, age, and category. Available voices have their own sharing and access conditions. Voice Design offers another route by creating a synthetic voice from a description.
Choose a voice using the actual script. A sample sentence can show its character, but a passage with your product names and sentence structure gives a more useful comparison. For a recurring course, listen to a calm explanation and an emphasized warning in the same voice. The presenter should remain understandable across both without making every sentence sound equally forceful.
Keep a record of the selected voice and relevant access conditions when continuity matters. A shared library voice and a voice created within your own account have different histories and availability. For a multi-character story, assign voices deliberately so the listener can distinguish roles. The difference should support the narrative rather than rely only on exaggerated accents or volume.
Pronunciation dictionaries provide rules for words or phrases that need special treatment in supported text-to-speech workflows. Supported notation and behavior depend on the model. This is useful for brand names, abbreviations, technical vocabulary, and names that appear repeatedly across a project.
Test the rule in a complete sentence. A term that sounds correct alone can still create an awkward transition in connected speech. Keep a viewer-facing written script separate from any phonetic generation text, and review captions against that written version. Otherwise, an audio correction can accidentally introduce a spelling error into the published transcript.
Build the dictionary around the terms that actually need intervention. A short, reviewed set is easier to maintain than a long list of speculative rules. When a localized version uses a different pronunciation or translation, organize that decision separately. A pronunciation dictionary controls how a term sounds; it does not determine whether the chosen wording is the correct translation.
ElevenLabs distinguishes Instant Voice Cloning and Professional Voice Cloning. Instant cloning uses a shorter sample and requires permission to clone the voice. Professional cloning trains a dedicated model and requires verification of your own voice. The Professional workflow cannot be created for another person on your account merely because they consent; they create and verify it themselves, then use supported sharing.
Prepare clean recordings that represent the delivery you want. A sample containing music, another speaker, or a strongly altered voice gives the workflow less useful source material. Use consistent recording conditions and listen to the source before uploading it. For a recurring narration project, include the speaking style expected in the finished work.
Cloning supplies a reusable voice; it does not approve a new script on behalf of the person represented. Keep the message review separate from the technical voice setup. Compare generated lines with the intended pacing and pronunciation, especially around emotionally important or specialist passages. Plan the source preparation and verification before promising a delivery schedule that depends on the clone.
Voice Changer transforms recorded or uploaded speech into a selected supported voice while using the source performance as guidance. This differs from text-to-speech, where the written words are the primary starting point. It is relevant when timing, emphasis, or expression is easier to perform than to describe.
For a short fictional scene, record the line with the intended pauses and emotional direction, then compare the transformed result. A rushed source performance may remain rushed in another voice. Re-recording a clearer take can be more useful than trying to repair every aspect afterward. Keep the recording focused on one speaker and one intended performance.
Review both identity and delivery. A recognizable voice may still pronounce a word unexpectedly or emphasize a different part of a phrase. For video, compare the transformed timing with the visible action. Preserve the source take so you can identify whether a problem began in the performance or during conversion.
The Voice Isolator reduces unwanted background sound around recorded speech. Its purpose is cleanup of an existing recording, rather than generating missing words. It can support a podcast, voiceover, or supplied video whose sound needs attention before further editing.
Choose a passage with quiet and loud speech to evaluate the treatment. Listen for words that become thin, clipped, or difficult to distinguish. A cleaner background is useful only if the important speech remains understandable. Keep relevant environmental sound in mind: a demonstration may need the sound of the process being explained.
Compare the processed result with the original in the intended mix. A voice can sound different when heard alone and when placed beneath music. If the source contains an inaudible phrase, obtain a correction or pickup where appropriate rather than treating cleanup as confirmation of what was said. Retain the unprocessed recording for comparison and future revisions.
Scribe converts spoken audio into text with supported features such as speaker identification, timestamps, and audio-event labels. The Scribe v2 overview describes its use in transcription and Studio caption workflows. Structured output is useful when a transcript needs to connect to a specific moment or speaker in the source.
Use timestamps to review a disputed word directly in the recording. Speaker labels help organize a conversation but should be checked when participants overlap or have similar voices. Correct names and specialist terms before using the transcript for subtitles, summaries, or another language. An error copied into several production stages becomes harder to repair later.
A transcript can support editing and navigation, while the recording remains the reference for tone and context. For a customer interview, retain qualifications attached to a statement when creating an excerpt. For a tutorial, compare the spoken instruction with the visible action. The text is useful production material, not proof that the speaker's claims are true.
ElevenLabs' dubbing documentation distinguishes Automatic Dubbing using the current v2 Alpha model from the v1-based Dubbing Studio route. Supported workflows detect speakers, translate dialogue, and retain background audio. Model, input, review, and licensing conditions differ, so choose the route around the intended use rather than treating all dubbing interfaces as identical.
Dubbing Studio provides transcript and segment controls but is documented as being in maintenance mode. Its fixed and dynamic regeneration settings affect how speech fits a segment. A large wording change inside a fixed duration can change the delivery pace substantially. Preview the revised line with the surrounding material.
Have a fluent reviewer check technical terms, tone, and instructions in the target language. Spoken translation does not automatically update visible screenshots or labels. Current Alpha or beta routes need their service-specific production and commercial conditions checked before use in a released project. A completed dub still requires an editorial and audio review, particularly when several speakers interact quickly.
The sound-effect tools generate audio from descriptions, while Eleven Music develops musical material for eligible creative workflows. These can be used with narration and recorded sound in a broader production. Music-specific rights and access conditions should be checked separately from speech generation.
Describe a sound by its role in the scene. A distant door closing, a close mechanical click, and a room ambience have different timing and perspective requirements. Preview effects alongside the visual action rather than selecting them only in isolation. Keep a sound bed restrained when it needs to support a spoken explanation.
For music, give the cue a purpose and an energy shape. A short introduction may need a clear ending; a narration bed may need fewer competing melodic elements. Adjust the final mix so the voice remains intelligible. A coherent soundtrack comes from the relationship between speech, effects, silence, and music, not from filling every moment with generated audio.
Studio provides a workspace for longer narration, audio, and supported video projects. It combines text-based speech revision, assigned speakers, a timeline, uploaded media, music, effects, captions, and sharing features. Studio Agent can help draft and arrange material, while manual controls let you refine the result.
For a narrated lesson, organize the script into sections that match the visual sequence. Assign a consistent voice, generate and review a representative passage, then develop the remaining sections. Keep music and effects separate enough to adjust them when the narration changes. Use the timeline to align instructions with the moment their result becomes visible.
Review the project from beginning to end after local revisions. A repaired sentence may be correct on its own but no longer connect naturally with the next section. Check captions against the final spoken version and use time-specific feedback to resolve issues. Keep the approved source text together with the project so the production history remains clear.
ElevenAgents supports interactive voice and chat systems with configurable voices, knowledge, instructions, tools, and workflows. It also provides testing and conversation-review features. This is a separate use case from exporting narration: the system responds to input rather than playing a fixed prepared script.
Define a narrow initial job, such as answering questions from an approved product help collection. Decide which questions require a human response and which actions the agent may take through connected tools. Test ordinary requests, ambiguous wording, and missing information before expanding the scope. A suitable voice is only one part of a useful interaction.
Review conversations for the actual outcome. Check whether the answer used the right source and whether the user could complete the task. Keep knowledge current when a product or policy changes. Creative-media subscriptions, agent usage, and API deployment have different requirements, so account access should be planned around the interactive system itself.
A team preparing a short lesson can start with an approved script and screen recording. Choose a voice using a passage containing the difficult terms, add appropriate pronunciation rules, and generate narration in manageable sections. Align those sections with the recorded actions in Studio. Add a small amount of music or effects only where they help explain the sequence.
Generate captions and check them against the approved script. Ask the product expert to review the instructions and the producer to inspect timing and the mix. If a target-language version is needed, choose an eligible dubbing or narration route and use a fluent reviewer. Update visible interface text separately where necessary. Export the approved media under the applicable creative access and retain the script, source recording, and selected voice information.
The creative pricing comparison includes free and paid plans with different credits, cloning, Studio, audio quality, and other features. Agent and enterprise access have separate considerations. Publishing guidance distinguishes free noncommercial creative output from eligible paid commercial output and applies additional service-specific terms. Beta services have separate restrictions, including production and commercial limits.
Budget for regeneration, review, and the actual tools used. Pronunciation errors, transcription mistakes, translation choices, and voice changes can remain in fluent-sounding audio. Shared voices can have access conditions, and a clone has a specific verification history. Confirm the plan and service conditions for the intended release before using an evaluation output as the final asset.
Professional cloning is for your verified own voice. The person can create and verify their clone in their own account and use supported sharing.
It generates a performance of the supplied text. Review factual content, terminology, and pronunciation separately.
No. Interactive agents, APIs, and creative-media production have different usage and access requirements.
Check the chosen model and service-specific conditions. Current Alpha or beta routes have restrictions distinct from regular released creative tools.