
bucolic_frolic
(52,215 posts)Last time I tried to find such I would up with Google Voice, I think, so I can talk to my cell phone but then have to save the files, cut and paste. THIS Is more like a word processor, trouble being getting the talk going as fast and cerebral as my typing.
A definite find!
Goonch
(3,989 posts)I find it to be as powerful and accurate as Dragon WITHOUT any training ;-{)
rog
(867 posts)... Can you use it to transcribe an audio file, like an .mp3, .ogg, etc? From your demo it looks like it only transcribes speaking into the mic in real time.
Thanks for the suggestion about choosing a module!
Goonch
(3,989 posts)rog
(867 posts)I was looking for a (free) application to transcribe recordings of medical appointments. I don't have a lot of those, but this service could cost $4-5/transcription.
I don't know if you're familiar with Google's (I know - I'm sorry) new service called NotebookLM. Basically it's a free application you can use online or 'install' as a web interface on your desktop. There are three panes; you can upload or link any sources you want (mp3, web page, pdf, text, youtube video, etc, etc) on any topic you're researching, the second pane works much like ChatGPT where you can interact with the LM by having a typed conversation (the responses are generated only from your sources - NotebookLM does not scour the web), and the third pane will generate study guides, summaries, flow charts, quizzes, a PowerPoint type presentation, even a 2-host 'podcast' discussing your topic material (again, based only on the sources you upload or link).
I've found this to be very useful, so -- regarding the transcriptions, I thought 'I wonder if NotebookLM can do that', since it accepts mp3 files as a source. I uploaded my doc recording, went to the interaction pane, and said 'generate a word-for-word transcription indicating who is speaking'. I had a (not perfect, but probably 90+% accurate) transcription in seconds, which I could print out and review. It also generated an excellent summary of the discussion with my doc, separating the topics discussed and hitting the high points of the conversation. I should add that - from the context of the recording (I did not supply this info) - the LM was able to discern that this was a medical appointment and that the overall topic was a routine annual exam.
Since "Speechnotes is powered by the leading most accurate speech recognition AI engines by Google & Microsoft," my guess is that NotebookLM uses the same Google speech recognition, so I imagine the accuracy is similar. From what I can tell, Speechnotes online is a lot more robust, but I don't need all those other advanced features. NotebookLM is perfect (for me), and really a game-changer for reviewing long and sometimes complex conversations. The summary serves as a great memory aid; if I need more info I can browse the transcript; if there's any ambiguity in the transcript I can consult the original recording.
I've also used it to aggregate and consolidate information about school board candidates in a recent election. I uploaded and linked news articles, interviews, questionnaire responses, profiles supplied by the candidates, etc, etc. There were often, for example, multiple interview articles in which the candidates all responded to the same questions, but they were published once a week by the local paper. NotebookLM was able to list all responses from each candidate under each question ... and lots more.
Link to the app: https://notebooklm.google.com/
Overview, via Google: https://notebooklm.google/
https://www.revolgy.com/insights/blog/what-is-google-notebooklm-and-why-you-should-start-using-it-right-now
Its a trained model like the others; however, you need to add your own data for it to work and respond. It makes sure all its answers can be verified against your sources, i.e. it wont hallucinate on you.
Unlike general AI tools like ChatGPT, NotebookLM doesnt pull information from the internet or make up details that cant be verified. Instead, it focuses entirely on your materials summarizing key points, finding relevant information, and suggesting ways to build on your ideas.
LPBBEAR
(568 posts)Another handy feature is its text to speech function. Occasionally I'm too tired to read an article. I copy paste the text into Speechnote and have it read the article for me. For extra giggles choose a voice with a English accent or a Indian accent.
Great program.