Eleven Multilingual Sts V2
Neural Network
Eleven Multilingual Sts V2 by ElevenLabs is a multilingual speech-to-speech model with content and prosody control.
Specifications
- Input
- audio
- Output
- audio
Prompt cost
(per 1M tokens)
How it works Eleven Multilingual Sts V2?
Eleven Multilingual Sts V2 is an ElevenLabs speech-to-speech model: it takes an original recording and re-voices it, preserving the meaning while allowing separate control over content and intonation. Multilingual support means one tool handles tasks in various languages. In BotHub, the model is accessible from Russia without a VPN or foreign card, payment is in rubles based on usage, and over 250 models are available in one window, making it easy to compare results without re-integrating. For teams, there is an API, corporate contracts with invoicing and EDI, an admin panel with limits, and data is protected by AES-GCM encryption and not stored. The model is useful for dubbing videos and social media clips, replacing voices in podcasts, creating multiple versions of a text with different delivery, voicing game and educational materials in multiple languages, and in voice assistant prototypes where prosody control is important.Frequently asked questions about Eleven Multilingual Sts V2
You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.
This is a speech-to-speech model: it takes your audio recording and voices it with a different voice, preserving the tempo, intonation, and emotions of the original. It is suitable for dubbing, voiceovers, video localization, and voice replacement when natural delivery is important, rather than synthesis from scratch.
It does not write text — the output is audio. The data indicates a context and output limit of 10,000 tokens, but for this model, this is a limit on the length of the processed speech fragment, not the volume of text. It is more convenient to submit long recordings in parts.
No. There is no reasoning step here: the model does not think about the answer, but converts the submitted speech into the selected voice. Our data does not indicate reasoning capabilities, and they are not required for speech-to-speech tasks.
No. Function calling and structured JSON output are not indicated in our data, and they do not apply to this model — its output is an audio file, not a text structure. For scenarios involving JSON and functions, a text model is required.
Audio is the input — this is the only format indicated in the data. The model does not accept images or video because it works with sound. Process images and clips separately, and only pass the audio track here.
The model is multilingual, but there is no quality assessment specifically for Russian in our data, so we cannot promise perfect pronunciation. Practical advice: run a short fragment of your recording through it and listen to the result — the timbre and intonation are preserved, but it is better to check the pronunciation by ear.