Learning a language from video is four file jobs: get a subtitle draft, fix the timing and format, slow the audio without turning voices into chipmunks, and record yourself against the original. Each one uses a different tool, and only one of them uploads anything.
The order matters because the subtitle file is the reference everything else hangs on. Generate or obtain the subtitles first, correct them, then use them while you slow the audio and record. Two limits shape the whole workflow: the speech recogniser supports Chinese and English only, and burned-in captions only cover Latin scripts. For any other language, keep the subtitles as a separate SRT or VTT file.
1. Get a subtitle draft
Video to Subtitles generates a transcript with timecodes from an audio or video file. Pick the language, turn on speaker labels for an interview, and edit names, numbers and wording in the preview before you save. The download comes as SRT, VTT or plain text. A free run is capped at 2 minutes of media per day, a free account receives 5 credits a day, and the upload limit is 30 MB.
Be clear about the trade. This tool is not browser only: the file is uploaded to a server and forwarded to Volcengine AI MediaKit for recognition, so do not send anything confidential. The recogniser also supports Simplified Chinese and English only. For French, Spanish, Japanese or any other language, you need a subtitle file from elsewhere, and the rest of this page still applies once you have one.
Video to Subtitles
Speech to SRT, VTT or text with timecodes. Uploads the file and uses Cloud AI credits.
2. Fix the timing and convert the format
Subtitle Converter reads SRT, WebVTT, ASS and SSA, and writes SRT, VTT and ASS. It also adds a time offset to every cue, which is the fix when subtitles drift ahead of or behind the audio; a negative value such as -1.5 moves them one and a half seconds earlier, and times that would fall before zero are clamped to zero.
Format matters for study because players differ. VTT is the web standard and what most browser players expect, SRT is the most widely accepted upload format, and ASS is useful when a desktop player wants it. Two limits. Styling is discarded on read: karaoke tags, positioning, drawing commands and fonts are dropped, and only start time, end time and text survive. And image based subtitles such as SUB and IDX are not supported, because they store pictures of text rather than text.
Subtitle Converter
Convert SRT, VTT and ASS, and shift every cue by a number of seconds.
3. Slow the audio without chipmunk voices
Audio Pitch & Speed changes tempo and pitch independently. The speed slider covers 0.5x to 2x and changes tempo only, so a voice at 0.5x keeps its normal tone. The pitch slider covers minus 12 to plus 12 semitones and keeps the original running time, which is useful for shadowing a line in a different key. Both controls can be linked if you want the tape recorder effect; for listening practice, leave them independent.
If the audio is inside a video, extract it first with Video to MP3 at 192 kbps, or 128 kbps for speech. The pitch tool opens MP3, WAV, M4A, AAC, OGG and FLAC. Pushing beyond 0.5x or 2x introduces audible artefacts, so 0.5x is the practical floor rather than a starting point for extreme slowdown.
Audio Pitch & Speed
Slow to 0.5x with the natural voice kept, or shift pitch by up to 12 semitones.
4. Record yourself and compare
Voice Recorder captures from any microphone you select, with no time limit and no watermark. The format follows the browser: Chrome and Edge usually produce WebM with Opus, Safari produces MP4 with AAC. A Convert to MP3 button is there when a player or a teacher needs MP3; the first conversion downloads an audio engine of about 30 MB.
One limit is worth planning around. There is no dual-track player here, so you cannot line your recording up against the original waveform. The practical method is to play the slowed source in one tab, record in another, then listen back to back. That is enough for shadowing, but it is not a pronunciation scoring tool.
Voice Recorder
Record from any microphone, download the take, or convert it to MP3.
What this workflow cannot do
- No recognition outside Chinese and English. The speech recogniser supports Simplified Chinese and English only, so other languages need an external subtitle file.
- No non-Latin burned-in captions. The subtitle renderer bundles one Latin font, so CJK, Arabic and Cyrillic text may appear as empty boxes. Keep those as a sidecar file.
- No translation. Nothing here converts a subtitle from one language to another. It converts formats and timing only.
- No automatic per-line alignment. The converter applies one global offset; it does not match each cue to the audio.
- No dual-track comparison. There is no waveform view that plays your take against the original.
- No image-based subtitles. SUB and IDX files store pictures of text and cannot be read.