{"title":"自动语音识别和模糊法证音频的转录:新一代系统表现如何?","authors":"Debbie Loakes","doi":"10.3389/fcomm.2024.1281407","DOIUrl":null,"url":null,"abstract":"This study provides an update on an earlier study in the “Capturing Talk” research topic, which aimed to demonstrate how automatic speech recognition (ASR) systems work with indistinct forensic-like audio, in comparison with good-quality audio. Since that time, there has been rapid technological advancement, with newer systems having access to extremely large language models and having their performance proclaimed as being human-like in accuracy. This study compares various ASR systems, including OpenAI’s Whisper, to continue to test how well automatic speaker recognition works with forensic-like audio. The results show that the transcription of a good-quality audio file is at ceiling for some systems, with no errors. For the poor-quality (forensic-like) audio, Whisper was the best performing system but had only 50% of the entire speech material correct. The results for the poor-quality audio were also generally variable across the systems, with differences depending on whether a .wav or .mp3 file was used and differences between earlier and later versions of the same system. Additionally, and against expectations, Whisper showed a drop in performance over a 2-month period. While more material was transcribed in the later attempt, more was also incorrect. This study concludes that forensic-like audio is not suitable for automatic analysis.","PeriodicalId":1,"journal":{"name":"Accounts of Chemical Research","volume":"21 10","pages":""},"PeriodicalIF":16.4000,"publicationDate":"2024-02-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Automatic speech recognition and the transcription of indistinct forensic audio: how do the new generation of systems fare?\",\"authors\":\"Debbie Loakes\",\"doi\":\"10.3389/fcomm.2024.1281407\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"This study provides an update on an earlier study in the “Capturing Talk” research topic, which aimed to demonstrate how automatic speech recognition (ASR) systems work with indistinct forensic-like audio, in comparison with good-quality audio. Since that time, there has been rapid technological advancement, with newer systems having access to extremely large language models and having their performance proclaimed as being human-like in accuracy. This study compares various ASR systems, including OpenAI’s Whisper, to continue to test how well automatic speaker recognition works with forensic-like audio. The results show that the transcription of a good-quality audio file is at ceiling for some systems, with no errors. For the poor-quality (forensic-like) audio, Whisper was the best performing system but had only 50% of the entire speech material correct. The results for the poor-quality audio were also generally variable across the systems, with differences depending on whether a .wav or .mp3 file was used and differences between earlier and later versions of the same system. Additionally, and against expectations, Whisper showed a drop in performance over a 2-month period. While more material was transcribed in the later attempt, more was also incorrect. This study concludes that forensic-like audio is not suitable for automatic analysis.\",\"PeriodicalId\":1,\"journal\":{\"name\":\"Accounts of Chemical Research\",\"volume\":\"21 10\",\"pages\":\"\"},\"PeriodicalIF\":16.4000,\"publicationDate\":\"2024-02-14\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Accounts of Chemical Research\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.3389/fcomm.2024.1281407\",\"RegionNum\":1,\"RegionCategory\":\"化学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q1\",\"JCRName\":\"CHEMISTRY, MULTIDISCIPLINARY\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Accounts of Chemical Research","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.3389/fcomm.2024.1281407","RegionNum":1,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"CHEMISTRY, MULTIDISCIPLINARY","Score":null,"Total":0}
Automatic speech recognition and the transcription of indistinct forensic audio: how do the new generation of systems fare?
This study provides an update on an earlier study in the “Capturing Talk” research topic, which aimed to demonstrate how automatic speech recognition (ASR) systems work with indistinct forensic-like audio, in comparison with good-quality audio. Since that time, there has been rapid technological advancement, with newer systems having access to extremely large language models and having their performance proclaimed as being human-like in accuracy. This study compares various ASR systems, including OpenAI’s Whisper, to continue to test how well automatic speaker recognition works with forensic-like audio. The results show that the transcription of a good-quality audio file is at ceiling for some systems, with no errors. For the poor-quality (forensic-like) audio, Whisper was the best performing system but had only 50% of the entire speech material correct. The results for the poor-quality audio were also generally variable across the systems, with differences depending on whether a .wav or .mp3 file was used and differences between earlier and later versions of the same system. Additionally, and against expectations, Whisper showed a drop in performance over a 2-month period. While more material was transcribed in the later attempt, more was also incorrect. This study concludes that forensic-like audio is not suitable for automatic analysis.
期刊介绍:
Accounts of Chemical Research presents short, concise and critical articles offering easy-to-read overviews of basic research and applications in all areas of chemistry and biochemistry. These short reviews focus on research from the author’s own laboratory and are designed to teach the reader about a research project. In addition, Accounts of Chemical Research publishes commentaries that give an informed opinion on a current research problem. Special Issues online are devoted to a single topic of unusual activity and significance.
Accounts of Chemical Research replaces the traditional article abstract with an article "Conspectus." These entries synopsize the research affording the reader a closer look at the content and significance of an article. Through this provision of a more detailed description of the article contents, the Conspectus enhances the article's discoverability by search engines and the exposure for the research.