Chapter
Automatic Speech Recognition Software and its Limitations
35:51 - 40:35 (04:43)
The speaker discusses the limitations of automatic speech recognition software such as inaccuracies and low word error rates in comparison to human accuracy.
Clips
The maximum achievable word error rate for human accuracy in transcription is 2-3%.
35:51 - 37:43 (01:51)
Summary
The maximum achievable word error rate for human accuracy in transcription is 2-3%. Rev offers an automated captioning service in addition to their human transcription services.
ChapterAutomatic Speech Recognition Software and its Limitations
Episode#151 – Dan Kokotov: Speech Recognition with AI and Humans
PodcastLex Fridman Podcast
The speaker discusses their experience with Rev, a transcription service that offers dollar per minute captioning and transcription options.
37:43 - 39:17 (01:33)
Summary
The speaker discusses their experience with Rev, a transcription service that offers dollar per minute captioning and transcription options. They note that while they had some issues with integrating the pipeline with Dropbox, the service was overall cheaper and had a user-friendly interface.
ChapterAutomatic Speech Recognition Software and its Limitations
Episode#151 – Dan Kokotov: Speech Recognition with AI and Humans
PodcastLex Fridman Podcast
In this podcast, the host and guest discuss ways to automate transcription downloads from Rev, including using integrations with Dropbox and Zapier or customizing with Rev's API.
39:17 - 40:35 (01:18)
Summary
In this podcast, the host and guest discuss ways to automate transcription downloads from Rev, including using integrations with Dropbox and Zapier or customizing with Rev's API.