Key Highlights
- Google introduces AudioPaLM, a new language model
- It enables zero-shot speech-to-text translations, voice transfer across languages, and more
- A new feature in Google Search that incorporates diverse perspectives from real human beings
In recent times, significant progress has been made in large language models (LLM), which are artificial neural networks trained on vast amounts of text data using self-supervised or semi-supervised learning. These models, such as Google Bard and OpenAI’s ChatGPT, power innovative generative AI tools. Among them, Google researchers have introduced AudioPaLM, a multimodal architecture capable of excelling in listening, speaking, and translating tasks. Also Read | Google Bard CHatGPT Killer Goes Official? What Is It? How To Use?
AudioPaLM: A Combination Of PaLM-2 And AudioLM Models
AudioPaLM integrates the strengths of two existing models, PaLM-2 and AudioLM. PaLM-2 is a text-based language model that exhibits proficiency in understanding text-specific linguistic knowledge. On the other hand, AudioLM excels in retaining paralinguistic information like speaker identity and tone. By combining these models, AudioPaLM achieves a deeper comprehension and generation of both text and speech. It is capable of speech recognition, translation with original voices, zero-shot speech-to-text translations for various languages, and voice transfer across languages based on short prompts.
Applications And Performance Of AudioPaLM
The versatility of AudioPaLM makes it invaluable for real-world applications, especially in real-time multilingual communication. The model displays competitive performance in speech recognition tasks and has obtained impressive results in speech translation benchmarks. With its ability to handle various languages and preserve distinct voices, AudioPaLM proves to be a powerful tool for language-related applications.
Also Read | Microsoft Introduces Bing AI To Edge Sidebar: How To Use?
Introduction Of Google Search’s Perspective Filter
During the Google I/O 2023 annual developers’ conference, Google unveiled a new feature for Google Search called ‘Perspectives.’ This feature aims to provide users with a more human aspect in their search results. Traditionally, search results were influenced by algorithms considering factors such as dates, authors, ratings, and proximity. However, the new Perspective filter introduces views and suggestions from real human beings, expanding the range of perspectives available.
Rollout Of The Perspective Filter
After its announcement at Google I/O 2023, Google has begun the global rollout of the Perspective filter for all Google Search users. The company shared the news through its official social media handles, inviting users to try out the new feature. The Perspective filter is designed to enhance the search experience by including diverse perspectives from experts and everyday people, further enriching the search results.
Also Read | OpenAI ChatGPT iOS Mobile App Launched: Full Details
