Supplied English Transcript Aligner

Get estimated word timestamps for a short English recording using the script you already have. Upload the audio to an Apify key-value store and paste the matching text. JSON keeps your supplied text and includes character spans and word start/end times.

For short narration workflows that need to preserve a script, such as matching a voiceover to text in an application. You supply both the audio and the transcript.

Example output

For the recorded sentence “I read the abstract before the full article.”, the JSON result includes:

{
  "text": "abstract",
  "charStart": 11,
  "charEnd": 19,
  "start": 0.42,
  "end": 0.88,
  "status": "estimated"
}

Times are in seconds. These are acoustic estimates, not frame-exact boundaries. When every word has an estimate and the recognition check agrees, the run also saves SRT and VTT files. Each contains one cue for the whole script; it does not split captions into readable lines or generate karaoke cues.

Input and limits

• One audio record in your Apify key-value store, at most 30 seconds and 32 MiB. WAV, FLAC, MP3, M4A and OGG are supported.
• Matching English text, at most 6000 characters and 440 model tokens. A short script can still exceed the token limit.
• Confirmation that you hold the rights to use the recording and text, and that it is nonpersonal English narration.

Use clear spoken narration. Singing, lyrics, meetings, interviews and personal records are outside the scope. There is no URL downloader or batch input. Longer recordings need another tool. Self-hosted open-source alignment tools are an alternative if you can provide your own setup and compute.

Try the demo

The prefilled input aligns a 1.96-second fiction clip with the text “HOW WOULD THE PAPERS TALK ABOUT IT”. Start it unchanged to try the JSON and caption outputs; the normal completion and start fees apply. The demo audio store is public-read and contains only the licensed sample and attribution.

The rights checkbox stays unchecked for the demo. Demo mode works only while no audio store is selected, the key is AUDIO, and the transcript matches this sample exactly. Selecting any store requires your own rights confirmation, even when demo mode is enabled. To use your audio, select your store, replace the text, and explicitly confirm your rights and the narration requirement. API requests that omit rightsConfirmed keep a false default; they must explicitly set useDemo: true to use the licensed demo.

Demo: unmodified LibriSpeech clip 2277-149896-0004, based on LibriVox audiobooks. Attribution: Vassil Panayotov, Guoguo Chen, Daniel Povey and Sanjeev Khudanpur (2015), LibriSpeech/OpenSLR, CC BY 4.0.

Run it with your audio

1. Upload your audio as a record in your Apify key-value store.
2. Select that store in Audio key-value store and enter its record key. The default key is AUDIO.
3. Paste the matching transcript, confirm the rights and narration requirement, and start the run.
4. Read the result in the dataset or the run's OUTPUT record. Download CAPTIONS.srt and CAPTIONS.vtt from the run's key-value store when captionExportAllowed is true.

The Actor requests read access to the selected audio store. Outputs go to the run's default storage. No external speech-service key is needed.

When captions are withheld

A separate speech-recognition pass checks agreement with your script. Disagreement, invalid alignment or unresolved words withhold the caption files. The JSON result explains the reason and can contain partial results with null times for unresolved words. Silence and invalid input are rejected.

Names, rare words and words the recogniser splits differently (such as 'upon' heard as 'up on') often cause a refusal; refused runs carry no completion fee. Agreement does not prove every word was spoken correctly. Review timings before using them in a finished project.

Recognition can also substitute an ordinary word and refuse a correct script. The result's comparison field shows the recognized text and each difference from your script. Your original script is in suppliedText. Keep your intended script; do not change it just to satisfy the recognition check.

JSON preserves the supplied text exactly. Caption files collapse whitespace and escape markup.

Fees

US$0.01 per complete estimated alignment, plus an Actor start charge of US$0.00005 per run (one start event per GB of memory; this Actor runs at 1 GB). Refused or unaligned results have no completion fee; the start charge still applies. Restarting a run through resurrection adds another start event. No subscription or external speech-service key is required.

Storage and privacy

Speech processing runs inside the Actor. It reads the selected audio record into temporary run disk and removes that temporary file after decoding. Your source audio record stays in your store. Input text, recognized text, hashes and outputs remain in Apify storage under your retention settings. Ending a run does not automatically erase that storage.

Built and maintained with AI agents by Madrasco. For support, use the Actor's Issues tab or support@madrasco.dev. Platform replies may be AI-assisted; the human owner can be reached on request.