YouTube now auto-generates captions in 68 languages and, for eligible channels, dubs videos into multiple languages without any manual work. But the convenience comes with caveats: transcription errors, synthetic voices that may not match the original speaker, and a legal requirement that no one’s voice be cloned without explicit permission. For creators weighing reach against risk, understanding exactly what the platform does—and where it stops—is essential.
- Automatic captions: YouTube generates speech-recognition subtitles in 68 languages, but quality varies and the platform itself recommends professional captions first.
- Auto-dubbing: Eligible channels get translated audio tracks by default—20 target languages from English, but only Arabic, English and Portuguese from Spanish.
- No editing: Automatic dubs cannot be edited; creators can only correct the source language or unpublish, so manual review before publishing is strongly advised.
- Voice cloning tools: Services like ElevenLabs translate into 90+ languages and preserve voice timbre, but raise consent and copyright issues.
YouTube’s automatic captions: useful, but always review
YouTube creates captions via speech recognition and publishes them when ready—not necessarily at upload. Its help page lists 68 available languages, including Spanish, Basque and Galician, but not Catalan. The platform is candid about quality: because captions are machine-generated, results vary from video to video, and YouTube encourages creators to upload professional subtitles first. Typical failure points include poor pronunciation, accents, dialects and background noise. Some videos get no captions at all—bad audio, multiple simultaneous speakers or mixed languages can block generation.
YouTube’s advice is to review captions “always.” In YouTube Studio, go to Subtitles, select the video and click More next to the automatic track to correct or delete errors. Two practical notes: live streams only get automatic captions in English and with normal latency, and by default YouTube replaces words it deems inappropriate with “[__],” a setting that can be changed in advanced options.
How auto-dubbing works and which languages it covers
Auto-dubbing generates translated audio tracks published alongside the original; viewers can switch between the dubbed version and the creator’s voice. According to YouTube’s help documentation, it is enabled by default for eligible creators, and dubbed videos are labeled “Auto-dubbed” in the description.
Target languages depend on the source. An English video can be dubbed into 20 languages: German, Arabic, Bengali, Korean, Spanish, French, Hebrew, Hindi, Indonesian, Italian, Japanese, Malayalam, Dutch, Punjabi, Polish, Portuguese, Russian, Tamil, Telugu and Ukrainian. A Spanish video, by contrast, can currently be dubbed only into Arabic, English and Portuguese; most other source languages go to English only. Seven targets from English—German, Spanish, French, Hindi, Indonesian, Italian and Portuguese—plus English from Spanish support expressive voice, which attempts to replicate the original tone and intonation, though YouTube notes it is not always used.
Not every video qualifies. YouTube excludes videos longer than 120 minutes, those with little speech or only music, those in unsupported or undetectable languages, those with excessively fast speech, and those with copyright claims.
The most important detail is in the fine print: automatic dubs cannot be edited. If a translation goes wrong, the only options are correcting the video’s original language so dubs regenerate, or unpublishing them. That is why enabling Review dubs before publishing under Settings > Channel > Advanced settings in YouTube Studio is advisable. The same menu allows disabling dubbing entirely. Some channels also have experimental features, such as lip sync that adjusts mouth movements to match dubbed audio.
Voice translation tools: when you want it to sound like you
YouTube’s dubbing uses synthetic voices. Specialized platforms offer something different: translating video while making the dubbed voice resemble the original speaker.
ElevenLabs is a prime example. Its dubbing documentation says it translates audio and video into more than 90 languages, detects different speakers and preserves each voice’s characteristics. Users can upload a file or paste a YouTube or TikTok link, and billing is per target language. Two clear advantages over YouTube: a similarity slider from 0 to 10 (default 7 in the Dubbing v2 Alpha model) and the ability to edit the transcript and translation and regenerate only the faulty segment.
ElevenLabs’ own documentation acknowledges the limit: pushing similarity too high can sound less natural between languages with very different phonetics. It is the same tension YouTube admits when it says dubbed voices may not match the original. Magnific, the European platform formerly known as Freepik, also includes voice cloning and lip sync among its audio tools.
The limits: errors, voices and rights
Three constraints deserve attention before publishing.
Translation errors. YouTube warns that dubs can fail on proper nouns, idioms and slang. The risk grows the more technical or colloquial the video. If you do not speak the target language, ask a fluent speaker to listen to at least the first dubbed video.
Other people’s voices. Dubbing your own voice is one thing; dubbing an interviewee’s is another. ElevenLabs’ use policy prohibits cloning or imitating a living person’s voice without explicit and informed consent or other valid authorization. In Spain, Organic Law 1/1982 treats use of a person’s name, voice or image for advertising, commercial or similar purposes as unlawful interference under Article 7.6. A monetized video using a guest’s cloned voice without permission sits on shaky ground.
Rights and labels. Third-party copyrighted content not only blocks YouTube dubbing; ElevenLabs’ policy also prohibits using its service to infringe copyright or image rights. And if a synthetic voice could make the public believe someone said something they did not, transparency obligations apply.
For a small creator, reviewed automatic captions and manually vetted auto-dubbing offer one of the best effort-to-reach ratios AI currently provides. What should not be delegated is the final listen. A mistranscribed caption is forgivable; a voice saying in another language something you never said—with your face and channel behind it—is a credibility problem no algorithm will fix for you.