Live AI Speech Translations

The accuracy of speech-to-text conversion has improved drastically, thanks to the use of the best AI and ASR technology on the market.

The result of the speech-to-text conversion, which is translated and then converted to speech, is 99+% accurate for commonly used languages like English, Spanish, French, German, Italian, Portuguese, Dutch, Japanese and others. For less common languages, the accuracy will be somewhat lower.

Factors such as speaking speed, articulation and dialect of the speaker or word usage like jargon and acronyms only reduce accuracy in extreme cases (and only to a very limited extent).

Even though the accuracy is very high, there is always room for improvement. You can do this by using a human operator to make real-time corrections to the source text. The operator doesn’t have to be a professional or someone with experience in the matter.

We expect the accuracy of speech-to-text to continue to improve in the near future. The best-of-class ASR technology used by Clevercast is constantly evolving.

The quality of the voices is excellent: they are clear and sound natural, almost human. If you have specific requirements that differ from our available voices, please contact us.

Sentence segmentation, fluency and intonation depends on the nature of your content and whether or not you’re using a human real-time editor.

For example, speeches with well-structured sentences are ideal for a conversion into AI speech. Conversely, with conversations involving hesitations or searching for words, AI may have difficulty accurately delineating sentences. This problem can be mitigated by using a real-time editor, who can ensure that sentenced are structure properly in the translation source.

Another working point is conversations. Currently, translation is done by one AI interpreter. So in a conversation, it is not always easy to know what is said by whom. In a future upgrade, we intend to remedy this by using two AI voices in such situations.

We are working on improving this. We also expect that AI text-to-speech technology will be able to handle conversational speech better in the near future.

Unlimited. In practice, it depends on your plan.

Yes.

When using AI-generated languages, you can expect the live stream to have a delay of approximately 120 seconds. We are working on a low-latency version of our current solution.

No matter what the delay is, translations are always in sync with the video of the live stream.

Yes, all AI translations are recorded in the cloud. You can download them afterwards as part of a single MP4 and as separate AAC files. Or you can publish a Video on-Demand with audio translations, hosted by Clevercast.

AI hours cost €120/hour, which include streaming in up to 8 languages (= the floor audio plus 7 AI speech translations, as well as 8 AI caption languages). For more info, see our pricing page.

AI hours are used when closed captions or audio translations are generated by speech-to-text conversion, text-to-text translation, or text-to-speech conversion. Usage depends on every set of 8 AI languages. For example, if you broadcast during 1 hour to a single streaming server and have 3 AI languages that are automatically generated, you will use 1 AI hour. If you stream with 10 AI languages, you will consume 2 AI hours.

Please note that this is based on the number of hours you broadcast to Clevercast. So AI hours will also count while your event status is ‘preview’ or ‘paused’.