Latest news with #AmazonTranscribe
CNET
19 hours ago
- Business
- CNET
Your Next Yelp Reviews May Include an AI Video
The next time you visit Yelp on your phone to check out reviews on the newest eatery, you may be greeted by an AI-generated video. Yelp is rolling out artificial intelligence videos to Yelp's home page feed on its iOS app. The AI videos use large language models to grab text from reviews on an establishment and turn it into an AI-voiced narration (courtesy of ElevenLabs) and captions (courtesy of Amazon Transcribe). Then Yelp uses uploaded photos from user reviews to create a slideshow-like display of what you can eat, drink or dance to. As far as oversight of these AI videos go, neither reviewers nor businesses appear to have any say. Companies can't see the AI videos before they are generated or offer input, and users can't decline to have their reviews or photos used, which raises a number of privacy questions. Viewers of the AI videos will have the choice to report a video as inaccurate or containing offensive content. And Yelp has said that it will be doing broad audits of the videos, which will be periodically updated for establishments as more reviews and photos come in. Speaking to The Verge, Yelp CPO Craig Saldanha said that the company wants to create as many videos as possible, although your own activity on Yelp will dictate whether you see the videos and which ones are shown. If we find any way for you to opt out of having your review content -- possibly from years past -- used in these videos, we'll be sure to let you know. Yelp did not immediately respond to a request for comment.
The Verge
2 days ago
- Business
- The Verge
Yelp is creating its own AI videos about restaurants
Yelp is going to use AI to stitch together user-posted content about restaurants, food, and nightlife businesses to make short videos about those businesses. The company initially started testing the AI-stitched videos last year, but they're now available nationwide on the iOS app's TikTok-like vertically scrolling home feed. Business operators can't currently see the videos that are generated for users, and Yelp users also can't currently opt out of having their photos or videos show up in Yelp's AI-stitched videos. Yelp relies on multiple generative AI tools to create the finished product, as OpenAI LLMs write the text descriptions and narrator's script, put together story topics, and proofread, while ElevenLabs is used to generate the narrator's voice and Amazon Transcribe creates the synchronized on-screen captions. You can get an idea of what they're like in the below video shared by Yelp. The vertical video blends together videos and images with an AI-generated voiceover and AI-generated captions to talk about things like the restaurant's food, cocktails, and ambiance. Yelp wants to make 'as many videos as possible,' Yelp CPO Craig Saldanha tells The Verge, but will only make them if a restaurant has enough reviews, photos, and videos to tell a compelling story. Yelp relies on personalized signals to determine when to actually show the videos to you. The videos themselves are not personalized, even though they are eventually refreshed — there is only one active AI-stitched video about a single business live at a time, according to Saldanha. If a user or a business feels that an AI-stitched video is inaccurate or offensive, Saldanha says they can report it by tapping the three dots in the top right corner of the video. Yelp does periodic audits 'at scale' as well. The AI-stitched videos follow other AI-focused features from Yelp, like review summaries and review filters. Posts from this author will be added to your daily email digest and your homepage feed. See All by Jay Peters Posts from this topic will be added to your daily email digest and your homepage feed. See All AI Posts from this topic will be added to your daily email digest and your homepage feed. See All Apps Posts from this topic will be added to your daily email digest and your homepage feed. See All Exclusive Posts from this topic will be added to your daily email digest and your homepage feed. See All News Posts from this topic will be added to your daily email digest and your homepage feed. See All Tech
Scotsman
02-06-2025
- Scotsman
Human transcribers more accurate than AI, study reveals
The typical AI error rate hovered around 38%. | Shutterstock A comprehensive study comparing automated speech recognition to human transcription revealed that AI-based tools significantly lag in accuracy. Sign up to our daily newsletter Sign up Thank you for signing up! Did you know with a Digital Subscription to Edinburgh News, you can get unlimited access to the website including our premium content, as well as benefiting from fewer ads, loyalty rewards and much more. Learn More Sorry, there seem to be some issues. Please try again later. Submitting... The study, conducted by Ditto Transcripts , tested eight AI transcription platforms - including TurboScribe, Notta, Amazon Transcribe, Sonix, Trint, and Microsoft Speech Recognition. They found that AI transcription had an average performance of just 62 per cent against human-made transcripts, which rated 99 per cent. According to the study, fourteen 15-minute audio recordings from legal, medical, and general business settings were submitted to each platform analysis. The audios featured a numbers of speakers, from single-person monologues to multi-speaker board meetings and congressional testimonies. The best-performing platform in this AI test was Sonix, with 73 per cent accuracy when adjusted for margin of error, while the lowest was Trint and Amazon Transcribe at just 61 per cent. What mistakes was AI making? It was revealed that AI struggled significantly with multiple speakers - even when speech was not overlapping - and was especially prone to errors when processing group speech, such as the Pledge of Allegiance. Four out of eight platforms failed to transcribe the pledge altogether in one of the tests. In other instances, single-word responses like 'Aye' were routinely misheard or omitted. Moreover, some platforms introduced entirely fabricated content when faced with unclear audio. Notta and Amazon Transcribe offered similarly incorrect interpretations - and this tendency to "hallucinate" content underscores AI's limitations in interpreting context and nuance. Speaker labelling was another weak point, with several AI-generated transcripts misidentifying the number of speakers or reassigned dialogue inaccurately mid-sentence. However, when challenged with the same audio bites, Ditto's in-house human transcribers achieved a 99% accuracy rate, with an error rate of around 1 per cent. The typical AI error rate hovered around 38%. Human transcribers achieved a 99% accuracy rate, with an error rate of around 1 per cent. | Shutterstock The consequences of inaccurate transcripts Speaking on the shock result, a Ditto transcripts spokesperson, said: 'Automated systems struggle with overlapping voices, complex dialogue, and contextual interpretation. 'These are common conditions in legal, medical, and law enforcement settings, where accuracy is non-negotiable.' The study also documented user experience issues - with some software requiring certain formats to submit the transcription. These transcription errors, if continued, would lead to serious outcomes. For example, a misidentified voice in a wiretap led to a year-long wrongful imprisonment; a dosage error caused by outsourced transcription resulted in a $140 million lawsuit against a hospital; and a prisoner was released 33 years early due to a misread sentencing transcript. According to the research, AI transcription may be useful for note-taking or casual use, but for professional settings, particularly those involving legal, medical, or regulatory documents, human transcription remains the more dependable choice. The full report, including detailed methodology and comparisons, is available from Ditto Transcripts, a FINRA, HIPAA, and CJIS-compliant transcription provider based in Denver, Colorado. How each AI platform performed after adjusting for margin of error: Sonix: 72.61% TurboScribe: 68.57% Notta: 67.41% 64.86% 64.41% Microsoft Speech-to-Text: 61.93% Amazon Transcribe: 60.81% Trint: 60.77%
Scotsman
08-05-2025
- Scotsman
Human transcribers more accurate than AI, study reveals
The typical AI error rate hovered around 38%. | Shutterstock A comprehensive study comparing automated speech recognition to human transcription revealed that AI-based tools significantly lag in accuracy. Sign up to our daily newsletter – Regular news stories and round-ups from around Scotland direct to your inbox Sign up Thank you for signing up! Did you know with a Digital Subscription to The Scotsman, you can get unlimited access to the website including our premium content, as well as benefiting from fewer ads, loyalty rewards and much more. Learn More Sorry, there seem to be some issues. Please try again later. Submitting... The study, conducted by Ditto Transcripts , tested eight AI transcription platforms - including TurboScribe, Notta, Amazon Transcribe, Sonix, Trint, and Microsoft Speech Recognition. They found that AI transcription had an average performance of just 62 per cent against human-made transcripts, which rated 99 per cent. According to the study, fourteen 15-minute audio recordings from legal, medical, and general business settings were submitted to each platform analysis. The audios featured a numbers of speakers, from single-person monologues to multi-speaker board meetings and congressional testimonies. The best-performing platform in this AI test was Sonix, with 73 per cent accuracy when adjusted for margin of error, while the lowest was Trint and Amazon Transcribe at just 61 per cent. What mistakes was AI making? It was revealed that AI struggled significantly with multiple speakers - even when speech was not overlapping - and was especially prone to errors when processing group speech, such as the Pledge of Allegiance. Four out of eight platforms failed to transcribe the pledge altogether in one of the tests. In other instances, single-word responses like 'Aye' were routinely misheard or omitted. Moreover, some platforms introduced entirely fabricated content when faced with unclear audio. Notta and Amazon Transcribe offered similarly incorrect interpretations - and this tendency to "hallucinate" content underscores AI's limitations in interpreting context and nuance. Speaker labelling was another weak point, with several AI-generated transcripts misidentifying the number of speakers or reassigned dialogue inaccurately mid-sentence. However, when challenged with the same audio bites, Ditto's in-house human transcribers achieved a 99% accuracy rate, with an error rate of around 1 per cent. The typical AI error rate hovered around 38%. Human transcribers achieved a 99% accuracy rate, with an error rate of around 1 per cent. | Shutterstock The consequences of inaccurate transcripts Speaking on the shock result, a Ditto transcripts spokesperson, said: 'Automated systems struggle with overlapping voices, complex dialogue, and contextual interpretation. 'These are common conditions in legal, medical, and law enforcement settings, where accuracy is non-negotiable.' The study also documented user experience issues - with some software requiring certain formats to submit the transcription. These transcription errors, if continued, would lead to serious outcomes. For example, a misidentified voice in a wiretap led to a year-long wrongful imprisonment; a dosage error caused by outsourced transcription resulted in a $140 million lawsuit against a hospital; and a prisoner was released 33 years early due to a misread sentencing transcript. According to the research, AI transcription may be useful for note-taking or casual use, but for professional settings, particularly those involving legal, medical, or regulatory documents, human transcription remains the more dependable choice. The full report, including detailed methodology and comparisons, is available from Ditto Transcripts, a FINRA, HIPAA, and CJIS-compliant transcription provider based in Denver, Colorado. How each AI platform performed after adjusting for margin of error:
