Praney Behl (@praneybehl) header
Praney Behl (@praneybehl) profile picture

Building Sustainable AI Products Solo | 20 Years Software Engineering | Voice AI, Devtools, Workflow Autom https://teapawt.net/t.co/Aw97xu1PGF https://teapawt.net/t.co/1OaI0jLNZn https://teapawt.net/t.co/JVVPRa7hHx https://teapawt.net/t.co/FegEBP7Hta https://teapawt.net/t.co/K8qLT4MBDY

Melbourne, Victoria
Joined August 2009
Photo by Praney Behl (@praneybehl)Photo by Praney Behl (@praneybehl)Photo by Praney Behl (@praneybehl)Photo by Praney Behl (@praneybehl)Photo by Praney Behl (@praneybehl)Photo by Praney Behl (@praneybehl)Photo by Praney Behl (@praneybehl)Photo by Praney Behl (@praneybehl)Photo by Praney Behl (@praneybehl)
This one helps me a lot! A short pronunciation sheet keeps each name in book four the same as in book one.
An audiobook series must have more than a list of unusual words. Keep the approved pronunciation, a sample sentence and the scope of each decision. Then the next volume starts with the answers.
1
Vois Studio is in AppSumo's Black Friday sale. From Sunday, Nov. 22 at 4 p.m. CT, get 10% off the AppSumo price. Pay once and keep it. appsumo.com/pl/GCAcUuoKkv/
1
1
16
This check is my favorite! I find each "because" in a script before the voice speaks it. Facts first, then the voice.
Before you generate documentary narration, find each "because" in the script. A smooth voice can give a cause that has no source. Keep a source path adjacent to each line with risk.
5
I want lessons that I can hear while I walk. This guide shows how to write them!
An audio lesson for a listener away from a screen must not refer to the screen. Replace "as you can see" with the relation. Give each section a spoken label. Then do a test with the screen off.
10
People using Gemini, what is your secret? 60+ mins in and Gemini 4 hydrogen their "FASTEST model" ever cannot even make a research plan let alone research anything. 🤯 Gemini fails my year of GenAI, prompt engineering and engineering experience altogether. @GeminiApp
2
93
UPDATE: Great success, I was finally able to get it to kick off the research just had to reload the page a few times after over an hour of head scratching. To my surprise, a single research with Gemini 4 Hydrogen - FAST + CHEAP model is burned a 38% usage. WOW 😱
36
@GeminiApp team, is there a secret manual to using Gemini and Gemini models. BTW I am on Pro plan and so far unable to use any of it. Your "BETTER" models need me to upgrade to Ultra plan with no confidence.
15
OpenAI celebrating fixes to codec capacity errors while their power users move on to anthropic clawed models. 😂🤣 teapawt.net/praneybehl/status/2122…
1
63
A very good direction rule: first the cause of the pause, then its length!
Do not start with "make the pause longer". Start with the task of the listener: complete a step, hear a new speaker or get a result. Then try two lengths in the full paragraph, and keep the length that helps that task.
24
So simply starting up Grok bot can consume 1% of your usage without even a single message send. 🤯 Is someone else seeing this too?
36
Game developers, this one is for you! The line ID is the file name, not the words of the line.
A line changes from "north gate" to "east gate". If the file name comes from the words, the game cannot find the line. Give each line a stable ID. Get approval for the names before an agent exports. Then listen to the files.
10
What a week! 🎉 Five open-source tools for Vois Studio. Thank you to each person who tried them. Tell us what to make next!
One week, five open-source tools (MIT) for Vois Studio: agent-voice, skills, speech-endpoint, video-voiceover and article-narrator. Start with agent-voice. Vois Studio is free for personal and non-commercial use, and agent-voice operates on it. github.com/vois-so
1
17
Voiceovers with no incorrect words! 🎬 video-voiceover examines each line and makes a new take when a word is incorrect.
1
21
Claude Code tells me with a voice when a task is complete. 🔊 Try agent-voice. It operates on the free plan of Vois Studio!
Hear when your coding agent completes a task. agent-voice (MIT) speaks the first sentence of the reply. It also tells you when Claude waits for your approval. Vois Studio is free for personal and non-commercial use, and agent-voice operates on it. github.com/vois-so/agent-voi…
1
38
Praney Behl retweeted
Tell your AI agent how to make a podcast, an audiobook or a voiceover. 16 skills (MIT) for Vois Studio: npx skills add vois-so/skills Vois Studio has a free plan for personal and non-commercial use. Audio files: paid plan or export credits. github.com/vois-so/skills
1
1
1
22
16 skills for your AI agent! 🚀 Ask Claude for a podcast, an audiobook or a voiceover, and it knows how to make it with Vois Studio.
Tell your AI agent how to make a podcast, an audiobook or a voiceover. 16 skills (MIT) for Vois Studio: npx skills add vois-so/skills Vois Studio has a free plan for personal and non-commercial use. Audio files: paid plan or export credits. github.com/vois-so/skills
19
Google Gemini-4-Argon in Pro subscription runs at about 6 tokens/s avg. How do the clever people that use this model work with such speeds? I am curious 🤔
37
Use Vois Studio voices in open-notebook and VoiceMode! 🎙️ speech-endpoint is a small server that operates on your computer.
Use Vois Studio voices in open-notebook and VoiceMode. speech-endpoint (MIT): a small local server for apps that use the OpenAI speech API. Vois Studio has a free plan for personal and non-commercial use. Audio files: paid plan or export credits. github.com/vois-so/speech-en…
1
19
People using GPT-6.1-Terra are you using it and Liking it? What are you using it for?
14