How do you transcribe audio with timestamps for free?

The free route is the open-source Whisper tool, which you run yourself. It gives you timestamps without a subscription, and it works on audio you do not want to upload to a web service. The trade-off is setup and checking. You install a program, you run it on each file, and you check the times it writes. Our guide to timestamps and speaker labels explains the different kinds of timestamp in more depth, and our overview of transcription covers the whole subject.

Install Whisper and run it on a file

Whisper is released under the MIT License, and the code and model weights are both covered ¹. The Whisper README on GitHub gives the install command, which is pip install -U openai-whisper ². Once it is installed, the README's example transcribes a file with a model named on the command line, for instance whisper audio.mp3 --model turbo.

The command line uses a graphics card from NVIDIA when PyTorch can detect one, and otherwise falls back to the processor ³. A computer without a supported GPU can still run the tool, so you do not need special hardware to start.

Add the flag for word-level timestamps

Plain output gives timestamps at the level of a sentence or chunk. For a stamp on each word, add the --word_timestamps True option to the command. The tool's source describes this option as "(experimental) extract word-level timestamps and refine the results based on them" ⁴. Read the word "experimental" as a warning: the feature is not guaranteed to be precise on every recording.

For a sentence-level stamp, the warning matters less. Whisper's own timestamps are "at the utterance-level, not per word, and can be inaccurate by several seconds" ⁵. That is accurate enough to find a passage. It is not accurate enough to cut a clip.

Use WhisperX when you need each word aligned

If you need every word aligned to the audio, a separate open-source project, WhisperX, adds word-level timestamps "utilising voice activity detection and forced phoneme alignment" ⁶. The alignment step is the part that places each word, and it is what the forced alignment entry in our glossary describes. Expect more setup than the basic Whisper command.

What free timestamps cannot tell you

A free timestamp does not tell you who is speaking. The timestamps show when each piece of speech occurs, not which person said it. If your recording has more than one voice, read our question on transcribing audio with multiple speakers before you start.

Free output also needs checking for words the model adds or leaves out. A 2025 study on dementia speech found that Whisper "tends to omit filler words such as uh and um" in its transcriptions, which means the words in your transcript may not match the audio exactly ⁷. A word the model leaves out has no timestamp, so a reader who searches the transcript for it and then goes to the audio will not find it.

Check the timestamps against the audio before you use them

1. Pick three points spread through the file: near the start, in the middle and near the end.
2. Play the audio at each stamp and listen for the first word of the line. If it is off, the grain is too coarse for your use.
3. Choose the grain by the job. Use sentence stamps to find a passage. Use word stamps only where you have checked them against the audio.
4. Keep the timestamp format in mind when you export. Subtitle files use a comma before the milliseconds in SRT and a full stop in WebVTT.

For a file you only need to transcribe once, the free file transcription in the browser is the quickest route, and our guide to transcribing audio to text covers the other options. For a regular workflow on your own machine, the command-line route above gives you more control, and the check in step 2 is the part that keeps the timestamps trustworthy. If you want to know whether the Whisper model itself is free to use, see whether Whisper AI is free.

For how often to place stamps once you have them, see how often to add timestamps to a transcript.

The short answer is that free timestamps are real and usable. The cost is your time spent checking them, and the checking is what makes them reliable enough to quote, caption or edit from.