Researchers propose using large language models to assess the meaning preservation of automatic speech recognition (ASR) transcripts, which is an alternative metric to word error rate (WER) for low-resource scenarios and atypical speech. The current WER and word accuracy metrics do not measure comprehensibility, which is a critical aspect of ASR performance, especially for users with atypical speech patterns.
AIOpenAI | Rating: 86 | 2024-07-09 08:16:44 PM |
Recent text-to-image generation models have made significant progress, but still suffer from issues like artifacts, misalignment with text descriptions, and low aesthetic quality. To improve these models, we propose rich human feedback for text-to-image generation and release a dataset to support this approach.
AI | Rating: 61 | 2024-06-26 09:55:52 PM |
We release the MISeD dataset of information-seeking dialogs focused on meeting transcripts, with corresponding baseline models. Meeting recordings have helped people worldwide catch missed meetings, focus instead of taking notes during calls, and review information. An agent that supports natural language conversations with meeting recordings could enable efficient navigation of recordings.
AI | Rating: 75 | 2024-06-25 05:06:13 PM |
The Kardar-Parisi-Zhang (KPZ) universality class describes the macroscopic behavior common to a variety of randomly growing interfaces, including growing wildfires and snow falling and clumping together. This class was introduced in 1985 by Kardar, Parisi, and Zhang.
AI | Rating: 42 | 2024-06-18 07:55:16 PM |
Researchers evaluated the performance of PaLM2 models in multilingual tasks, comparing direct inference with pre-translation. They found that direct inference in the source language outperformed pre-translation to English. This study focused on generative tasks like text summarization and attributed QA, where the output needs to be in the source language.
AI | Rating: 81 | 2024-06-17 05:15:35 PM |
Human I/O is a unified approach that detects situational impairments and assesses a user's ability to interact in a given situation. These impairments, known as situationally induced impairments and disabilities (SIIDs), can be caused by environmental factors like noise, lighting, temperature, stress, and social norms.
AI | Rating: 72 | 2024-06-14 07:55:14 PM |
We strive to create an environment conducive to many different types of research across many different time scales and levels of risk.
Philosophy | Rating: 42 | 2024-06-14 07:05:26 PM |
We strive to create an environment conducive to many different types of research across many different time scales and levels of risk.
AI | Duplicated with: 1 | Rating: 42 | 2024-06-14 07:05:26 PM |
Our resources are available to everyone, and we regularly share datasets, tools, and services with the broader scientific community to be used, shared, and built on.
Science | Rating: 48 | 2024-06-14 07:05:25 PM |
Google Research strives to create an environment conducive to various research types across different time scales and risk levels.
AI | Duplicated with: 1 | Rating: 42 | 2024-06-14 07:05:25 PM |