{ "version": "https://jsonfeed.org/version/1.1", "title": "AI Audio blog", "home_page_url": "https://aiaudio.geenius.io/blog/", "feed_url": "https://aiaudio.geenius.io/blog/feed.json", "description": "Practical posts on voice-overs, music beds, effects and dialogue, takes and credits, and being honest about simulated previews, from the team building AI Audio.", "language": "en", "items": [ { "id": "https://aiaudio.geenius.io/blog/a-robot-voice-and-why-we-say-so/", "url": "https://aiaudio.geenius.io/blog/a-robot-voice-and-why-we-say-so/", "title": "A robot voice, and why we say so", "summary": "What the AI Audio Preview voice is, how it is made, why it sounds like a robot and why that is the honest starting point for sound from words.", "date_published": "2026-10-18T09:00:00Z", "authors": [ { "name": "The AI Audio team" } ], "tags": [ "preview", "honesty" ], "content_text": "If you press Generate on a voice-over in AI Audio today, you will hear a robot. We would rather you heard that first from us.\nHow the voice is made\nThe Preview voice is built in your browser in plain code, with no model and no recording. A pulse like the one in a human throat is passed through filters tuned to vowel sounds, with noise for the hissing consonants. The pitch comes from the voice you choose, the contour from the emotion, and the pace from the speed. The result is intelligible and mechanical.\nWhat else the Preview makes\nMusic beds from chords, a bass line, an arpeggio and drums in the mood you choose, with a loop that folds its tail into its start. Sound effects shaped by the word in your description, such as rain, wind, whoosh, beep, boom, click, laser or chime. Dialogue, where each turn gets its place. All real sound files, all simulated.\nWhy ship it\nBecause the product around it is real, and it needs something to run on. The composer, the credits shown before you press, the takes with waveforms, the history of runs, variations, the library, uploads, downloads and saving to Geenius OS Media with provenance all work with the Preview. When a real voice model is switched on, it plugs into the same place and nothing else changes.\nWhat is connected\nGemini speech, ElevenLabs voices and an optional OpenAI voice are connected in code for single voices. They work only where their keys are set on the server or in your workspace's Geenius OS secrets, and whether any of them is on for the hosted app is something we say we have to confirm. Google's Lyria for music and ElevenLabs sound effects are planned and not connected, and two real voices in one dialogue are planned too.\nHow to use the Preview\nUse it as a scratch track, to check timing and length and to try a script in three languages. Replace it before anything goes out. Do not put it in front of a customer as a finished voice.\nThe point\nA product that tells you what its voice is made of is a product you can trust with the next thing it says.\nSee what makes the sound." }, { "id": "https://aiaudio.geenius.io/blog/check-the-cost-first/", "url": "https://aiaudio.geenius.io/blog/check-the-cost-first/", "title": "Check the cost first", "summary": "How AI Audio shows the credits a run costs before anything is made, how credits are counted per take, and what is refunded when a run fails.", "date_published": "2026-10-18T09:00:00Z", "authors": [ { "name": "The AI Audio team" } ], "tags": [ "credits", "product" ], "content_text": "Credits are the part of a plan people worry about, so we show them before you press anything.\nThe rates\nA voice-over or a dialogue costs 1 credit for each 500 characters. A music bed costs 1 credit for each 10 seconds. A sound effect costs 1 credit. Then the total is multiplied by the number of takes you ask for, one to four.\nWhere you see it\nIn the composer, next to the Generate button, before anything is made. Through the door, a queued run answers with the credits it costs. A short voice-over of two hundred characters is one credit a take, and a sixty-second music bed is six.\nThe allowance\nPremium has 1,000 credits a month and Max has 3,000. Pro has none, because Pro does not make sound. The month is a calendar month, and credits are counted from the runs themselves.\nFailed runs\nA run can fail. A provider can refuse, a key can be missing, a tab can close at the wrong time. A failed run is refunded the takes it never made, so you do not pay for what you did not get.\nThe Preview and other credits\nToday the sound comes from the Preview, which costs no Geenius OS credits. If a real model is used in a workspace on a Geenius OS plan, the run is also charged to the Geenius OS ledger and a refusal there stops the run with the Geenius OS sentence. We tell you so because two ledgers can be confusing.\nHabits\nMake one take first and listen before you ask for four. Use a short script to test a voice. Use a shorter music length to test a mood and the long one for the final. And keep the take you like: you will not need to make it again.\nWhat a credit is not\nIt is not a promise of quality, and it is not a price in euros. Premium and Max are monthly plans with an allowance, and credits are how the allowance is measured.\nSee credits." }, { "id": "https://aiaudio.geenius.io/blog/takes-are-a-habit/", "url": "https://aiaudio.geenius.io/blog/takes-are-a-habit/", "title": "Take four, keep one", "summary": "Why AI Audio makes one to four takes per run, how to listen on a waveform, and how to keep the one you like with its settings.", "date_published": "2026-10-18T09:00:00Z", "authors": [ { "name": "The AI Audio team" } ], "tags": [ "takes", "craft" ], "content_text": "Anyone who has recorded a voice knows the best take is rarely the first. AI Audio builds that habit in.\nOne to four\nA run makes one to four takes. They appear one by one, each on its own waveform, so you can play the first while the others are still being made. Each take has its own seed, so they differ.\nListen on the waveform\nThe waveform shows where the sound is loud and quiet. Look for a voice that rushes a phrase, a bed whose loop point shows a seam, or an effect that is too long. Then play it. The player draws the peaks for every take.\nKeep what you like\nRename the take you like, put it in a collection and download it as WAV or MP3. If you want more like it, make variations: they keep the settings and use new seeds. If you want to change one thing, reuse the run's settings and edit that one.\nThe history helps\nEvery run stays in the history with its words, voice, mood, length and settings, so you can come back to it and make it again. A failed run stays too, with its reason.\nCost\nTakes are multiplied into the cost, so four takes cost four times as much. Listen to one first when you are testing, and ask for more once the words are right.\nWith the Preview\nToday's takes differ in small ways, because the Preview is a synthesis with seeds. That makes it a good place to learn the habit. A real model will differ more between takes, and the habit will matter more.\nSave the recipe\nSaving to Geenius OS Media keeps the words, model, seed and settings with the sound, so a take you keep is a take you can explain and repeat.\nA small example\nImagine you ask for three takes of a welcome line. The first rushes the last word, the second is steady, the third is a little flat. On the waveforms you can see it before you play: the first has a spike at the end, the third has no peaks at all. You keep the second, rename it welcome-final, put it in the project collection and move on, and the history still has the other two if you change your mind.\nSee history and takes." }, { "id": "https://aiaudio.geenius.io/blog/queue-it-let-a-page-make-it/", "url": "https://aiaudio.geenius.io/blog/queue-it-let-a-page-make-it/", "title": "Queue it, and let a page make it", "summary": "How an AI assistant queues AI Audio runs through Geenius OS and why a browser Preview run is made when somebody has the app open.", "date_published": "2026-10-18T09:00:00Z", "authors": [ { "name": "The AI Audio team" } ], "tags": [ "ai", "mcp" ], "content_text": "If you work with an AI assistant, you can ask it to make sound in AI Audio. One detail is easy to miss, and we would rather you heard it from us.\nWhat your AI can do\nThrough Geenius OS your AI gets an audio tool and a collection tool. For sound it can check which providers are ready, queue a run with a kind and a prompt and the options of that kind, list runs and sounds, get one with its waveform, check usage and delete a run with a confirmation. For collections it can list, create and add or remove sounds.\nWhat happens when it queues a run\nThe run goes into the history as queued, and the answer includes the credits it costs. The Preview, today's maker, runs in a browser. So a queued run is made as soon as somebody in the workspace has AI Audio open. Your AI cannot make it on its own; it should tell you to open the app and give you the address.\nWhy a browser\nThe Preview synthesises sound in a page and needs no server and no key. That keeps the whole product runnable without a vendor. A real model runs on the server, and once its key is set a queued run is made without anyone watching.\nWhat your AI should say\nThat it queued a run, what it costs, that the sound comes from a simulated Preview unless a real model is ready, and which provider is ready. It should not promise a result. Asking for the providers first gives that answer.\nPlans\nMaking sound needs Premium or Max, whoever asks. The door is on every plan, and a Pro workspace gets the library and nothing queued.\nA note on waiting\nA queued run that nobody opens just waits. Nothing is lost: it stays queued until a page claims it, and a page that closes hands the run on to another. If it matters that a run finishes tonight, open the app yourself.\nSee use with your AI." }, { "id": "https://aiaudio.geenius.io/blog/a-sound-library-on-every-plan/", "url": "https://aiaudio.geenius.io/blog/a-sound-library-on-every-plan/", "title": "A sound library, on every plan", "summary": "Why the AI Audio library of uploads, collections, search, downloads and Media is on every plan including Pro, and how to use it without making sound.", "date_published": "2026-10-18T09:00:00Z", "authors": [ { "name": "The AI Audio team" } ], "tags": [ "library", "plans" ], "content_text": "Pro does not make sound. It is a decision, not an omission.\nTwo jobs\nAI Audio does two jobs. One is making sound, which is the AI feature and needs Premium or Max. The other is keeping it: uploads, collections, searching by the words in the prompt, renaming, WAV and MP3 downloads and Geenius OS Media. The second job is useful even if you never press Generate, so it is on every plan.\nWhat Pro is good for\nA place to keep sounds that came from elsewhere, with names and collections. A way to bring sounds in from your Geenius OS Media library and save them back. A searchable history of what you made on a bigger plan if you change plans. And a door for your own AI to keep your collections tidy.\nWhat Pro is not\nIt is not a trial of making sound. If you want to make it, choose Premium or Max. We do not offer a free plan or a trial period, and a promo key from us can grant a short trial to people we invite.\nSearching\nSearch looks at the words in the prompts and the names. If you name your sounds by what they are for, you will find them later.\nProvenance\nSaving to Media puts the words, model, seed and settings with the sound. That is part of the library too.\nCounting\nMaking sound is counted in credits: 1,000 a month on Premium and 3,000 on Max. Pro counts none because it makes none.\nBe honest with yourself\nToday the maker is a simulated Preview. A library is the part of AI Audio that is fully real today.\nA fair question\nWhy pay for a library if you cannot make sound? If you already have sounds, or work from Geenius OS Media, or your team needs one tidy place with search, collections, uploads and downloads, the library alone is enough. If you want to make sound, the bigger plans are there. We would rather offer a plain choice than hide the library behind the maker.\nSee library and plans and limits." }, { "id": "https://aiaudio.geenius.io/blog/three-languages-one-script/", "url": "https://aiaudio.geenius.io/blog/three-languages-one-script/", "title": "Three languages, one script", "summary": "How AI Audio speaks English, German and Dutch, how to test a script in all three with the Preview, and what to check when a real voice is used.", "date_published": "2026-10-18T09:00:00Z", "authors": [ { "name": "The AI Audio team" } ], "tags": [ "languages", "craft" ], "content_text": "Many of the people we build for work in more than one language. AI Audio speaks English, German and Dutch.\nWhat the language setting does\nFor a voice-over or a dialogue you choose English, German or Dutch. The setting tells the maker which language the words are in and, for a real model, which voices are available. The preview voices are available in all three.\nTesting a script\nWrite the script once, then make a take in each language. Listen for length: German and Dutch often run longer than English, which matters if the voice has to fit a picture. The cost is by characters, so the longer language costs a little more.\nWhat the Preview does with language\nThe Preview voice is a robot voice built from vowel formants. It makes the three languages sound like the same robot reading different words. That is enough to test timing and length, and not enough to judge pronunciation.\nWhat a real voice adds\nA real voice model will pronounce names, numbers and abbreviations in its own way. When one is switched on, listen to the names in your script first, and spell out anything that is read wrongly. The history keeps the words of each take, so you can compare.\nSpeed and emotion\nSpeed runs from half to double, and emotion has six settings: neutral, warm, cheerful, calm, serious and excited. Both apply in every language, and both are part of the settings the history keeps.\nA habit\nKeep one collection per project with the three language versions side by side, named by language. When the script changes, make the three again from the same settings.\nWhat is not here\nVoice cloning, and voices in other languages, are not part of AI Audio today.\nAn example\nSuppose you have a thirty-second explainer. Make the English take, then the German and the Dutch from the same settings. If the Dutch runs three seconds longer, you know before you cut the picture, and you can shorten the script or loosen the edit. That is a cheap check that saves a lot of rework later.\nSee voice-overs." } ] }