# Changelog of Speech to Text — YouTube, TikTok & Media URLs (`lergassy/speech-to-text`) Actor

- **URL**: https://apify.com/lergassy/speech-to-text/changelog.md
- **Full Actor documentation**: https://apify.com/lergassy/speech-to-text.md

## Changelog

### 2026-10-12

- A run stopped by your "Max cost per run" no longer says "No rows were delivered" when transcripts were delivered and paid for; the extra notice row is gone.
- A TikTok or YouTube page refused on one proxy address ("requiring login", "confirm you're not a bot") is tried once more on a new address instead of becoming an error row.

### 2026-10-11

- Speaker labels are now about ten times faster and cheaper to compute. A recording of an hour or more with speaker labels used to run out of time and return nothing; it now finishes in minutes.

### 2026-10-10

- No more silent holes in long transcripts: when the recognition service skips a stretch of speech (it can jump ahead by up to 30 s), that stretch is recognised again and put back. Pauses stay empty.
- Files on Wikimedia Commons (upload.wikimedia.org) download again.

### 2026-10-03

- "Max cost per run" now covers the whole price of a transcript (file, video link, minutes and speaker labels): a file that does not fit returns a free row saying so, and every transcript written is fully paid for.

### 2026-10-02

- Default memory now adapts to your input: a few files use less, larger batches or speaker detection get more.

### 2026-09-27

- Runs stop cleanly at your maximum cost instead of being aborted; nothing beyond your limit is written or charged.

### 2026-09-26

- SRT and WebVTT subtitles are split into cues of up to two 42-character lines and at most 7 seconds.

### 2026-09-24

- Files from hosts that redirect or break the first download are retried before the row reports them unreachable.

### 2026-09-19

- Video page links (YouTube, TikTok and others) download only the audio track, so they are faster.
- A video page larger than 150 MB returns a free row explaining why instead of a charge.
- The input form lists all supported video sites and offers ready-made examples.

### 2026-09-16

- The prefilled example is now a one-minute sample.

### 2026-09-05

- Initial release: audio and video to text in 90+ languages with timed passages, SRT and WebVTT subtitles and retrieval chunks, from MP3, M4A, WAV, MP4, MOV and other media URLs; files of any length; no API key needed.