How to measure browser text-to-speech speed fairly

Four clocks, four different questions

First load is the time from asking TTS Lines to speak until the local voice model is ready. The browser downloads the model from Hugging Face, so connection and cache state matter. Generation is the wait after you press Make voiceover until the lines are playable. A correction is the time to change and regenerate one line. Export is the wait from choosing MP3 or ZIP until the file is saved.

In our 14 October workflow test, we generated two English lines, exported two MP3 clips, changed only line two, and regenerated it. We did not capture a clean cold-cache stopwatch reading, so there is no first-load result or 'fastest' claim here. The 2- and 3-second labels on the clips are audio lengths, not generation times.

Use a test another person can repeat

Write down your device model, operating system, browser version, connection, language, voice, and whether the model is already cached. Use the same short script for every run: “This is the opening line for the scene.” followed by “This line will be replaced after the edit.” Record the character count and the local or cloud voice setting.

For a cold first-load run, use a browser profile that has never loaded this model. Do not clear an everyday profile just for the test; that would remove other site data. Start a stopwatch when you press Make voiceover and stop when both lines can play. Then run the same script twice more with the model ready. Report all times or the median, and label the first run separately.

Measure the correction your editor actually makes

Change only line two to “The final scene now has a shorter closing line.” Start the clock at Regenerate and stop when that line can play. Time the MP3 or ZIP download separately. The line one audio should remain available; if the whole script starts over, record that too.

If you compare another service, keep the script, export target, connection, and measurement points the same. Cloud tools and local models pay different costs at first load and on repeat runs. Report those costs instead of hiding them in one average.

What we can say today

TTS Lines does its free generation on your device after the model loads. The first download can be large, and slower devices may take longer. A browser cache can be cleared or evicted. We have verified the separate correction workflow, but we have not published a controlled cross-device timing table. If speed determines your purchase or production choice, run the four-clock test on the device you will use.