How to Create Video Transcripts for Free

A transcript turns the speech in a video into text you can search, quote, edit or convert into captions. Free options fall into two groups. Some speech recognition tools run on your own computer. Online services and video platforms process the file on their servers. The choice affects privacy, file limits and how much control you have over the result.

Before you start

First decide where the processing will happen. Open-source speech recognition models such as Whisper can run locally, either from the command line or through a desktop app built on top of them. Do not assume a local app works fully offline. Some apps download models on first use, check for updates or send usage data, so check the app's documentation. Online services and built-in platform captions are usually easier to start with. Their free tiers often limit file length, number of uploads or export formats. Read the current terms on storage, retention and whether uploaded media may be used to train models.

• The original video file, or access to the account where it is published
• Enough free disk space for the video, an extracted audio copy and the exported transcript
• For local tools, a computer that can handle the model; processing time depends on your hardware and the model size
• The spoken language or languages in the recording
• A decision on the output: plain text for reading and notes, or timed captions (SRT or VTT) for playback
• Permission to process the content, especially recordings of calls or meetings with other participants

Step-by-step instructions

1. Listen to the audio before transcribing

Play short sections from the beginning, middle and end. Accuracy depends on the audio track, not on video resolution or frame rate. Note anything that could cause errors: the capture source (a built-in laptop microphone versus a headset), background music, echo, people talking over each other, or long silent stretches.

2. Extract the audio (optional)

Many tools accept video files directly. Extracting the audio first gives a much smaller upload and avoids problems with unsupported video encoding. With the free command-line tool FFmpeg, the following command creates a mono 16 kHz WAV file, a common input for speech models, while leaving the original untouched: ffmpeg -i input.mp4 -vn -ac 1 -ar 16000 audio.wav. Converting the audio standardizes the recording format but does not improve the sound. A higher bitrate will not remove noise that is already in the recording.

3. Choose the tool and its settings

• Language: if the tool lets you set the language manually, do so instead of relying on auto-detection
• Model size (local tools): larger models are generally more accurate but slower and need more memory
• Output format: TXT for plain text, SRT or VTT for captions with timestamps
• Speaker labels: use them only if the tool supports them, and expect to correct them

If the video is already published on a platform you manage, check whether it creates automatic captions you can download. Whether this feature exists, which languages it supports and which export formats it offers depend on the platform and account type, so check its current help pages.

4. Run the transcription

Start with a short clip if you are unsure about the settings, then process the full file. For long recordings, stop the computer from sleeping. With an online service, follow its instructions on whether the browser tab must stay open while processing runs.

5. Export, back up and edit

Save the transcript next to the video with a matching file name. Keep one unedited copy so you can always go back to the raw output. Then correct names, technical terms, numbers and punctuation. Mark speaker changes where they matter. When editing SRT or VTT files, change only the text lines and leave the cue numbers and timestamps as they are.

Check the output

• Accuracy: compare the transcript with the video at several points, including the end; quality often drops in noisy or crowded sections
• Timing: load the SRT or VTT file into a media player that supports external subtitles and confirm captions stay in sync throughout, not just at the start
• Structure: open the file in a plain text editor; SRT uses numbered cues with a comma before milliseconds, while VTT starts with a WEBVTT header and uses a period
• Character encoding: save as UTF-8 so accented or non-Latin characters display correctly
• Completeness: look for skipped passages, repeated lines and text that appears during silence or music, which speech models sometimes generate
• File handling: confirm where the service stored your upload and delete it if the service allows this and you no longer need it there

Common problems and next steps

Many wrong words usually point to the recording, not the tool. Try a larger model or another free tool on the same file. For future recordings, move the microphone closer to the speaker, reduce background noise, and record in a high-quality or lossless format. Noise reduction before transcription can help, but heavy filtering can make speech harder to recognize, so compare results on a short clip.

If the transcript comes out in the wrong language, or switches languages, set the language manually. If the recording genuinely mixes languages, split it into sections and process each one separately.

Repeated phrases or invented sentences usually appear in silent or music-only parts. Trim those parts before transcribing, or remove the extra text during editing.

Captions can drift out of sync after you import them into a video editor. Screen and phone recordings often use a variable frame rate, which some editors handle poorly, so check whether yours does. Re-encoding the video to a constant frame rate before editing is a common fix. Keep the original file in case you need to start again.

If an upload is rejected or a file is too long for a free tier, upload the extracted audio instead of the video, or split the recording into shorter parts. Check the service's stated size and duration limits rather than guessing.

If local processing is too slow, choose a smaller model or process shorter segments. If the tool offers GPU acceleration and your hardware supports it, check its documentation for how to enable it. Many free tools do not label speakers at all, so plan to add speaker names by hand for interviews and meetings.