Subtitle Generation Guide

From Video Audio to Bilingual Subtitles

Subtitle generation uses automatic speech recognition (ASR), also called speech-to-text (STT). It turns spoken audio into text and adds timestamps so each line appears at the right point in the video.

Translation is a separate step that works on subtitle text. Start it when you need another language. If the video already has readable text subtitles, you can go straight to reviewing, editing and translating them.

Learn where to start, prepare a speech model, generate and edit subtitles, and translate them for original, translated or bilingual playback.

Where to Start

Tap the video to show the player controls, then tap the subtitle speech-bubble button at the top. The panel has Settings, Subtitles and Generate tabs.

When subtitles are available, the panel usually opens on Subtitles. When none are available, it opens on Generate. You can switch tabs at any time.

Settings → Features → Subtitle Generation provides model management and translation management.

Open a local video in the player and choose subtitle generation. Prepare a speech model in model management, then select a model, language and audio track as needed. Review the result, edit text and timing, and export an SRT copy.

First-time model preparation takes time and storage. Processing speed depends on the device, model and video length. Check recognized text before using it.

Player → Subtitles in the top controls → Generate. Settings → Features → Subtitle Generation provides model management and translation management.

Download and Choose a Speech Model

Before the first transcription, download an ASR model in Model Management. Downloads use network data and device storage. Once the model is ready, speech recognition runs on your device.

Within the same model family, larger models are usually more accurate, but need more storage, memory and processing time. Language, recording quality and your device also affect the result.

Whisper Models

Multilingual Whisper models can detect the spoken language automatically or use a language you select. English-only versions are intended for English audio.

Parakeet Models

The current Parakeet model automatically recognizes its supported European languages. It does not support Chinese, Japanese or Korean; choose a suitable multilingual Whisper model for these languages.

Use the device recommendations and short audio test in Model Management to compare options. Test speed is a reference, not a prediction for every video.

Speech models generate text from audio. Apple translation language resources translate existing text. These are different downloads and do not replace each other.

Generate Subtitles and Follow Progress

In Generate, select a downloaded model and the recognition language. For a video with multiple audio tracks, choose the track containing the speech you want, then start generation.

Recognition language means the language spoken in the video. Choose the target language later, when you translate the subtitles.

Tasks may wait in a queue before processing. Follow the current stage, progress and estimated remaining time. You can pause and continue a task, or retry after a failure.

Background processing is limited by iOS. If a task is interrupted, reopen it and continue from its saved progress. Keep the app in the foreground when possible for long tasks.

When transcription finishes, GPlayer normally saves an SRT file with the video’s name and automatically loads it into the player. Open Subtitles to check the result.

The subtitles are loaded as a separate file. Generating them does not write text into the video picture or change the original video.

Review and Edit Subtitles

Tap a subtitle row to jump to its position in the video. The current row is highlighted, making it easier to listen again and check the wording.

Edit the original text and, when a translation exists, its translated text. You can also adjust the start and end timestamps with millisecond precision.

Each tap of a time adjustment button changes the timestamp by 0.1 seconds. Press and hold for continuous adjustment. Keep the end after the start and within the video’s duration.

Closing the editor or switching to the previous or next row automatically saves your changes and updates playback subtitles. If saving fails, follow the message and retry before leaving.

After entering selection mode, tap the rows you want to translate, delete or merge. Use Select all or Deselect all to change the selection quickly. Merging requires at least two consecutive rows.

Choose Original, Translated or Bilingual Display

Use the display menu on the Subtitles tab to choose Original, Translated text or Bilingual. Bilingual places the original above its translation.

Off hides subtitles on the video picture while leaving the subtitle list available. When only some rows are translated, untranslated rows continue to show the original text.

Common Questions and Better Results

Why can’t generation start?

Check that a compatible model has finished downloading, the local video is accessible, and the selected audio track contains audio. Download remote files to the device first.

Why are some words incorrect?

Check the spoken language, the model’s supported languages and the selected audio track. Noise, background music and overlapping voices make recognition harder. Try a larger suitable model and correct remaining errors in the editor.

Why is processing slow?

Long videos and larger models take more time. Device performance, temperature and Low Power Mode also affect speed. The remaining time is an estimate and may change as processing continues.

What should I check if translation fails?

For Apple, check supported languages and downloaded language resources. For DeepL, check the API key, plan and quota. For online engines, check the connection. Use Translation Test before retrying the subtitles.

Why are subtitles garbled or unavailable for editing?

Check the subtitle file’s text encoding and choose a font that supports its language. Image-based subtitles cannot be edited or translated as text directly; generate text subtitles from the video’s audio instead.