IBM Released a New Speech Recognition Model that Outperformed Others

As artificial intelligence tools become more advanced and easier to use, people want to interact with these tools in the same way that they interact with other people. This is especially true for tools such as customer service chatbots, voice assistants, and those that convert audio into summaries. The better these models are at understanding what people say, the more helpful they will be at home and at work. As a result, tech giant IBM (IBM) recently released a new speech recognition model, Granite Speech 3.3 8B, that is designed to handle business-focused tasks.

Interestingly, this model—which converts English speech into text and can translate it into languages such as French, Spanish, Japanese, and Mandarin—was built by fine-tuning IBM’s text-based Granite 3.3 8B Instruct model. Furthermore, it is open-source and available for anyone to use on Hugging Face, which is a platform that hosts and shares AI models. However, what is most impressive is that since its release, Granite Speech 3.3 8B has topped Hugging Face’s Open ASR leaderboard. In fact, it even beat models from big names like OpenAI and Meta (META) in terms of accuracy and performance.

The model performed well because IBM trained it on many types of audio, such as voicemails, audiobooks, and earnings calls, in order to prepare it for real-world use. In addition, the firm added background noise and intentionally cut parts of the audio to strengthen the model. Nevertheless, speech recognition remains a challenge, as IBM believes that it will take another five to 10 years before AI can match human listening skills.

What Is the Target Price for IBM?

Turning to Wall Street, analysts have a Moderate Buy consensus rating on IBM stock based on seven Buys, five Holds, and two Sells assigned in the past three months, as indicated by the graphic below. Furthermore, the average IBM price target of $269.46 per share implies 6.8% downside risk.

See more IBM analyst ratings