Supported Languages

Understanding language support

Explore VoxSail language options by feature, regional variants, counting rules and voice restrictions before starting recognition, translation or dubbing.

The homepage's 130+ refers to language and regional options across features, not 130 distinct languages in every feature. Regional options are listed separately, not counted as new independent languages. Feature totals cannot simply be added; a language supported by multiple models is counted once.

Counts and verification scope

The configuration lists 125 base-language groups and 14 regional options across features (139 combined options). Recognition lists 99 groups; voiceover lists 100; standard video dubbing lists 84; professional video dubbing lists 93 base groups plus 14 regional options. These are configuration counts, not a claim that every language, voice or translation pair has been tested.

The lists come from the product's recognition capability catalog, voiceover target catalog and video dubbing workspace options. They describe configuration coverage, not a quality certification. Providers, models, voices, feature availability and account access can affect actual use. Check the current workspace options before submitting a task.

This inventory was prepared on 2026-10-27. The production configuration and public page were checked with independent video dubbing enabled. Paid generation has not been tested for every listed language, translation direction, voice combination or regional accent.

Completed sample tests

On 2026-10-27, we used Fish S2.1 Pro in the test environment for 11 text-to-speech generations across 8 languages: Mandarin, English, Japanese, Korean, French, German, Spanish and Cantonese. The initial round included one generation per language, followed by one additional Mandarin test and two additional Cantonese tests. Every run was checked for completion, audio delivery, file format and playback in a real Chrome browser. These checks are not a pronunciation quality pass.

The tests used short passages in each language. The Mandarin and Cantonese comparisons kept their respective original passages and model unchanged, changing only the reference voice. Results are recorded below:

These results apply only to the tested short passages, model and voice combinations. They do not mean all 130+ options have been tested, nor do they establish quality guarantees for speech recognition, subtitle or text translation, audio/video dubbing, voice cloning, long-text stability or every regional accent. Only the method and result summary are published here; test audio, reference recordings and internal task or account details are not public.

• English, Japanese, Korean, French, German and Spanish: one generation and playback check each; human pronunciation review in the corresponding languages has not been completed.
• Mandarin: one generation each with an English-reference voice and a Mandarin-reference voice. The English-reference output failed user listening review because of a noticeable foreign accent. Pronunciation quality for the Mandarin-reference retest is still awaiting user confirmation.
• Cantonese: one generation each with an English-reference voice, a Mandarin-reference voice and a dedicated Cantonese-reference voice. Pronunciation quality for the English-reference output has not been confirmed. The Mandarin-reference output was described by the user as not idiomatic enough. After switching to a voice with a Cantonese tag, a description identifying a Cantonese speaker and a reference recording, the user reported that the Cantonese was much better. This is comparative improvement feedback, not a professional native-speaker review or a quality pass for every Cantonese voice.

Choosing a reference voice

• Check compatibility with the target language, then prefer a voice with a reference recording in that language. Catalog tags help with selection but cannot replace listening.
• For Mandarin, start with a Mandarin-reference voice. For Cantonese, start with a dedicated Cantonese-reference voice; a Chinese or zh label alone does not establish Cantonese pronunciation. An English- or Mandarin-reference voice may retain an unnatural accent when generating Cantonese.
• In this Cantonese comparison, changing the reference voice substantially improved the user's listening impression. This supports testing specific voice combinations separately, not assuming the same outcome for every reference voice or model.
• Before full production, generate a short sample you are authorized to use and ask someone fluent in the target language to check tones, pronunciation, numbers, proper names and naturalness. Listen again after changing the model or voice. Follow the current workspace availability and actual output when making your choice.

Language coverage by feature

“Listed” means a language appears in the feature's configuration or workspace options, not that every voice supports it. “Not listed” means there is no corresponding option in that list; it is not a linguistic assessment.

Recognition covers input languages for audio/video translation and subtitle transcription. Voiceover combines target languages across configured models. Independent video dubbing has separate standard and professional lists; it is not the same as dubbing settings within audio/video translation. These voiceover lists also do not establish subtitle or text translation quality.

Regional variants

These options are listed only for professional video dubbing. They do not imply the same accent choices in other features or every voice. Base languages and regional versions are shown separately; a regional version is not an additional independent language.

Code normalization and counting

• Automatic detection is an input method, not a language, and is excluded.
• Base-language aliases are grouped, including zh/cmn, jw/jv and nya/ny.
• fil and tl have compatibility mappings in the current product and are shown as one Filipino / Tagalog group. This is a product counting convention, not a linguistic classification.
• Cantonese and Mandarin are listed separately. Some voiceover services share a provider code, which does not guarantee Cantonese pronunciation quality for a particular voice.
• Regional options retain their full codes, such as en-US and en-GB, and are not counted again as independent base languages.
• Standard video dubbing follows the current workspace list. A code accepted only by backend validation is not counted as a user-selectable option.

Before choosing a language

Choose a target language, then check compatibility with your selected mode and voice. Recognition support does not imply voiceover support. Voiceover availability does not mean every cloned voice will perform equally well.

Before publishing, preview a short sample you are authorized to use. Check pronunciation, terminology, subtitles and timing. Ask a fluent reviewer to assess less common languages, dialects and regional accents.