Audio & Video Transcription: Speech to Text (Whisper)
Pricing
$10.00 / 1,000 audio minute transcribeds
Speech to text and audio to text: transcribes audio and video files with timestamps and SRT/VTT subtitles, offline with OpenAI Whisper inside the Actor: no third-party API. 100 languages, auto-detected, or translation to English. Respects robots.txt. Failed files are free.
Links to audio or video files you are allowed to process (MP3, M4A, MP4, MOV, WAV, FLAC, OGG, Opus, WebM, MKV, AAC, WMA, AVI and more). Each file is downloaded once; the site's robots.txt is respected. Links to web pages (YouTube, podcasts' pages) are not files and fail.
Files sent inline, for files that are not online: a list of base64 strings, or of objects {"name": "call.mp3", "data": ""}. A data: URL prefix is accepted. Apify limits the whole input to a few MB, so use links for larger files.
The language spoken in the files, or 'Detect automatically' (detected once per file, from its first minutes). Naming it avoids a wrong guess on short or noisy files.
Transcribe keeps the spoken language. Translate writes English text from speech in any language (rougher than transcription).
Add a 'segments' list: each phrase with its start and end time in seconds.
Add each word with its start, end and probability to the segments (slower, about 10-20% more time).
Subtitle files written to the run's key-value store for each file; the item links to them. Leave empty for none.
[ "srt", "vtt"]Also put the subtitle text itself in each item (fields 'srt' and 'vtt').
Removes stretches without speech before transcribing (Silero VAD). This reduces made-up text in silences. Timestamps still refer to the original file.
Transcribe at most this many minutes of each file; longer files are marked truncated and charged only for what was transcribed.
Larger downloads are stopped and reported as file-too-large, without charge. Files over 300 MB need a run memory of 2 GB or more.