Advertisement

Russian "GigaAM" outperforms OpenAI and Nvidia in performing tasks in the Russian language

Russian "GigaAM" outperforms OpenAI and Nvidia in performing tasks in the Russian language


Russian AI models for speech recognition have shown superiority in performing tasks in the Russian language, making half as many errors as neural network models from OpenAI and Nvidia.

Hot Topics

About half of Libyans drink contaminated water  What is the scientific reason?

Mauritania's blue treasure is in danger Will the map of fish in the Atlantic be redrawn?

The Citrus Secret Everyone's Talking About: Burn More, Crave Less & Feel Great Naturally

This was reported by MTS Web Services on the eve of the Digital Technologies Forum.

A spokesperson for the Russian company said: "According to the test results, the Russian models made half as many errors as the two foreign technologies from OpenAI and Nvidia, and demonstrated sufficient accuracy for use in the most prominent business scenarios specializing in the key sectors of the Russian economy."

The test was conducted on more than 12,000 labeled audio recordings, totaling approximately 38 hours in length. The sample included recordings of speech free from obvious noise, as well as recordings containing background noise and speech.

The neural networks were compared using the "WER" or Word Error Rate index, which is an index that reflects the number of errors in word recognition compared to the number of words in the reference text.

It should be noted that, according to the test results, the average word error rate of the "GigaAM" model from the Russian company "Sper" was approximately 7.4 errors per 100 words of the reference text. This rate is less than half that of the "Nvidia Parakeet" model, which scored 15.8%, and more than three times that of the "Nvidia Nemotron" model, which scored 25.5%.

Furthermore, the difference was even more pronounced in recordings that included background noise and speech, with the index reaching 18% for "GigaAM", compared to 33% for "Nvidia Parakeet" and 62% for "Whisper" from "OpenAI".

The second most accurate development was another Russian one, "Vosk" from the company "Alpha Cephei".

It is worth noting that the "Digital Technologies" forum is being held at the "Russia" National Center from October 6 to 10.

what is your opinion on this news ?do let us know in the comments on tha comment box

 if you liked this news article please share on friends & family & don't forget to follow on website & social media

 

Post a Comment

Previous Post Next Post
Advertisement