I would like this for summary of my long rambling audio ideas . Whisper is good but i want a one stop shop and i pretty much use T3 for all my other LLM needs
Copper Honey
I think one could implement this experimentally with mp3 support, size limit, and a limited range of models, e.g. Gemini 2.5 Flash Lite ($0.3/M tok), and Gemini 3.1 Flash Lite ($0.4/M tok).
For a request to G2.5 FL, I got about 30,000 native prompt tokens when the input was a 7.6MB mp3 file with 15 minutes of audio (0.5MB reduction when quiet parts were removed). This was $0.0106058 of usage for reference.
Scarlet Highlighter
I really would like this, a lot of times I record audios about my advancement on projects I work on and what to do next because I find it easier to do things this way, it would be really useful if I could use T3 Chat to summarize those audios, I know I can use whisper to convert them to text which is what I do know but I believe models like Gemini Flash 3 would do a much better job to avoid transcription issue.
Apricot Gizmo
Would be awesome, especially with flash 3 I'm really impressed by its voice processing and helps me do trans voice training; found it to be extremely helpful at giving instant feedback last session.
Doesn't seem too difficult to implement looking at openrouter’s docs either.
Apricot Honey
i’ve personally built a small n8n flow and linked a telegram bot, which transcribe voice recordings using groq api and the make some activities depending on the content.
As part of my normal work, i usually record very long audio notes inside chatgpt in order to transform them into nice emails and to-do lists for my coworkers.
Indigo Mustard
If this refers to voice transcription as part of the input window, that would be awesome. Just like ChatGPT has. This, to me, is different from real-time voice. I want to be able to read the answer because usually that is faster than listening to the AI read it out, and I get to skim it.
It helps me brainstorm faster because I can just leave an audio note in the chat and then have it help me work through what I’m thinking.
Black Sorbet
I originally thought that transcribing the audio wouldn’t be very useful, but then I saw
Aaron Francis uses it to give an LLM a loose idea of a prompt and have it turn into a proper prompt that you can then feed to the LLM to take action. That let me here to request the future because now that I have seen that way of using it, I will definitely want to use it a lot more rather than typing it all out, whichis what I have been doing
Coffee Asteroid
Having the option to upload an audio file and transcribe it would be excellent. One of the best uses of an LLM is to just talk to yourself about something and, based on that, generate a transcription and process that into a fleshed-out output.
Log in to comment and vote
Comments8
Salmon Yogurt
I would like this for summary of my long rambling audio ideas . Whisper is good but i want a one stop shop and i pretty much use T3 for all my other LLM needs
Copper Honey
I think one could implement this experimentally with mp3 support, size limit, and a limited range of models, e.g. Gemini 2.5 Flash Lite ($0.3/M tok), and Gemini 3.1 Flash Lite ($0.4/M tok).
For a request to G2.5 FL, I got about 30,000 native prompt tokens when the input was a 7.6MB mp3 file with 15 minutes of audio (0.5MB reduction when quiet parts were removed). This was $0.0106058 of usage for reference.
Scarlet Highlighter
I really would like this, a lot of times I record audios about my advancement on projects I work on and what to do next because I find it easier to do things this way, it would be really useful if I could use T3 Chat to summarize those audios, I know I can use whisper to convert them to text which is what I do know but I believe models like Gemini Flash 3 would do a much better job to avoid transcription issue.
Apricot Gizmo
Would be awesome, especially with flash 3 I'm really impressed by its voice processing and helps me do trans voice training; found it to be extremely helpful at giving instant feedback last session.
Doesn't seem too difficult to implement looking at openrouter’s docs either.
Apricot Honey
i’ve personally built a small n8n flow and linked a telegram bot, which transcribe voice recordings using groq api and the make some activities depending on the content.
As part of my normal work, i usually record very long audio notes inside chatgpt in order to transform them into nice emails and to-do lists for my coworkers.
Indigo Mustard
If this refers to voice transcription as part of the input window, that would be awesome. Just like ChatGPT has. This, to me, is different from real-time voice. I want to be able to read the answer because usually that is faster than listening to the AI read it out, and I get to skim it.
It helps me brainstorm faster because I can just leave an audio note in the chat and then have it help me work through what I’m thinking.
Black Sorbet
I originally thought that transcribing the audio wouldn’t be very useful, but then I saw
Aaron Francis uses it to give an LLM a loose idea of a prompt and have it turn into a proper prompt that you can then feed to the LLM to take action. That let me here to request the future because now that I have seen that way of using it, I will definitely want to use it a lot more rather than typing it all out, whichis what I have been doing
Coffee Asteroid
Having the option to upload an audio file and transcribe it would be excellent. One of the best uses of an LLM is to just talk to yourself about something and, based on that, generate a transcription and process that into a fleshed-out output.