# Captions from speech and Cut pauses > How to transcribe speech in a manual project into word-by-word captions — in the cloud or free on iPhone, iPad and Mac — and cut pauses for free. Source: https://docs.slicer.media/en/guide/editor/speech/ In an [Edit it myself](https://docs.slicer.media/en/guide/projects/manual/) project speech isn't transcribed when the project is created, so at first there are no captions. Two tools help you get such a reel in shape: - **Captions from speech** — AI transcribes speech in the sources you pick, and word-by-word captions appear on the clips by themselves. In the cloud — at cost; in the app — free, on the device itself. - **Cut pauses** — removes silences longer than a threshold. Free, no AI. In a project with the AI director, speech is transcribed during the montage. There these tools only appear when there is something to transcribe — for example, a file added in the editor — or on a clip without speech. **Demo (Web):** 1. In the main set of a manual project press “Text” — the sheet starts with “Captions from speech” and the price per minute 2. Press “Captions from speech” — sources with sound are checked, the language is “Detect”, the “Transcribe” button shows the price 3. Press “Transcribe” — a minute later “Captions are ready”: phrases on the captions track and in the preview 4. Press “Cut pauses” in the main set — a sheet with the threshold and “The reel gets … shorter” 5. Pick “1 s” — the estimate updates 6. Press “Cut pauses” — “Pauses cut”, the clips on the strip are split and close up ## Captions from speech ### Where to find it - **Text** → **Captions from speech**. Under the row — the price of a minute of transcription. - **Transcribe** in the main tool set of a manual project. - **Transcribe** in the tools of a selected clip — the sheet opens with just that clip's source. - **Text** → **Edit caption**, if the reel has no phrases yet, opens **Captions from speech** right away. The buttons are there while the project still has sources with sound whose speech hasn't been transcribed. ### How to transcribe ### 1. Pick the sources In the **Captions from speech** sheet, the **Sources** list shows videos and voice-overs with sound whose speech hasn't been transcribed; music isn't listed. All of them are checked by default. Uncheck the ones without speech: you pay for the whole length of the checked sources, not only the part that made it into the reel. ### 2. Set the language **Speech language** — **Detect**, and Slicer figures out the language itself. If you know the language, pick it from the list: transcription is more reliable that way. ### 3. Choose where to transcribe In the app for iPhone, iPad and Mac there's a **Where AI runs** row: **On device** for free or **In the cloud** for a price. More in On device or in the cloud. On the web, the cloud transcribes. ### 4. Start In the cloud the “Transcribe” button shows the price of what you picked; on the device it reads **Transcribe · free**. The cloud transcribes in the background: **Transcribing — usually under a minute.** You can keep editing meanwhile. When it's done, you'll see **Captions are ready**. ### On device or in the cloud In the app for iPhone, iPad and Mac speech can be transcribed right on the device — for free. You choose in the sheet, in the **Where AI runs** row: - **On device** · free — Apple's speech model on your phone, tablet or Mac does the work; - **In the cloud** · price — an AI service does it, at cost, just like on the web. The captions come out the same: words with exact timing on the clips. The preselected option comes from **Profile** → **Where AI runs** ([Language, theme and where AI runs](https://docs.slicer.media/en/guide/account/preferences/#ai-where)), and you can change it in the sheet. On the device: - **Set the language.** **Detect** is only available in the cloud; on the device the project or interface language is preselected. The device handles Russian, English and other widely used languages; if a language isn't supported, you'll see **This language can’t be recognized on the device — pick another language or the cloud.** - **Only videos from this device.** If the sources were uploaded from another phone or from the web — **The videos were uploaded from another device — only the cloud is available here.** - **Progress shows above the preview.** The first time, the system may download the language model — **Downloading the language model**, then **Recognizing speech on the device** with a percentage and **Sending the result**. - **Keep the app open:** **This device is recognizing speech — keep the app open. Captions will appear on the clips by themselves.** You can close the sheet and the editor. - **If it fails** — **Couldn’t do it on the device**, and the sheet opens again: try again or pick the cloud. If iOS asks for speech recognition access and you don't allow it — **No access to speech recognition — allow it in Settings.** - On the web the device option is disabled and labeled **In the app**. ### What changes in the reel - **The video doesn't change.** Clips stay where they were, and their edges still move by time. - **Captions appear** — on clips where speech is heard, and as phrases on the **Captions** track. If captions were turned off in the style, Slicer turns them on. - **Everything captions can do works with the words:** fix the text, accent, cut words, emoji on a word, translation ([Captions](https://docs.slicer.media/en/guide/editor/captions/), [Caption translation](https://docs.slicer.media/en/guide/editor/translate/)). - **Elements stay in place** — titles, plaques, emoji and zooms don't move. Each source is transcribed once: after that it's no longer in the list, even if no speech was found. When all of them are done, the sheet says **Speech in every source with sound is already transcribed.** ### Price On the device transcription is free: nothing is reserved or charged. In the cloud transcription is charged at cost — per minute of audio of the checked sources. The price is on the button before you start, and the sheet explains: **Costs exactly what the AI service charges, no markup. The actual price is charged — never more than shown.** - While transcription runs, the price is reserved on your balance, rounded up to a whole token — so you need at least one free token. - If it fails, you'll see **Transcription failed** and the tokens come back. - The project history keeps a **Captions from speech** entry; one done on the device is marked **on the device**. How pricing at cost works — [Tokens, packs and the free reel](https://docs.slicer.media/en/guide/pricing/#at-cost). ## Cut pauses Slicer finds pauses by the silence in the audio while it prepares the files — no AI is needed, so the tool is free. ### Where to find it - **Cut pauses** in the main tool set of a manual project — pauses in the whole reel. - **Cut pauses** in the tools of a selected clip — you can cut pauses in that clip only. The button is there on clips whose edges move by time: in a manual project that's every video. Clips the AI built from words have no pauses left. ### How to cut them ### 1. Pick the threshold **Pauses longer than** — **0.3 s**, **0.5 s** (the default) or **1 s**. The smaller the threshold, the tighter the reel. ### 2. Pick where If the sheet was opened from a selected clip, choose **This clip** or **Whole reel**. ### 3. Check and apply The sheet works out the result in advance: **The reel gets 4.5 s shorter**. If there's nothing to cut — **No pauses longer than 1 s**. Press **Cut pauses** — you'll see **Pauses cut: −4.5 s**. **Free. Undo brings them back.** All pauses come back in a single undo step. ### How clips are cut - Silence is removed with a small margin at the edges so speech isn't clipped. - A clip is split into parts, and each part is a separate clip with the same settings: speed, volume, color, effect. - The transition, the entrance animation and keyframes stay with the first part, the exit animation with the last. - Elements tied to time move to the part that holds their point, or to the nearest one ([Time anchoring](https://docs.slicer.media/en/guide/editor/elements/#timing)). - Freeze frames and reversed clips aren't cut. ## Good to know - Cloud transcription carries on even if you close the editor or the app: the captions will be there when you come back. On the device, keep the app open. - Don't need captions on a particular phrase — **Hide captions** in its tools ([Captions](https://docs.slicer.media/en/guide/editor/captions/)). - It's handy to cut pauses right after creating the project: then all that's left is to trim edges and reorder clips. ## What's next - [Captions: text, accents, style](https://docs.slicer.media/en/guide/editor/captions/) - [Caption translation](https://docs.slicer.media/en/guide/editor/translate/) - [Project without AI](https://docs.slicer.media/en/guide/projects/manual/) - [Operation prices](https://docs.slicer.media/en/guide/help/prices/)