Fast Speech to Text: Parallel Whisper Transcription
Pricing
Pay per usage
Transcribes one audio window per run so long videos can be split and transcribed in parallel. Whisper large-v3-turbo on CPU with word timestamps; supports Standby HTTP mode.
https:// URL (public artifact URL from a previous run, or any reachable media) or kv:/. Range-queried when the server allows it.
Offset of this shard on the original timeline.
Shard length before the tail overlap. 240-300s is the measured sweet spot for faster-whisper on CPU (RTF ~4.7x); short 15-45s shards waste fixed overhead.
Extra audio past duration_sec; words there are flagged _overlap so the engine can dedupe.
fast | adaptive | balanced | accurate
Silent windows return 0 words instead of failing.
Baked models. large-v3-turbo = multilingual default (Vietnamese OK). distil-large-v3 is ENGLISH-ONLY and is auto-downgraded to large-v3-turbo for non-English audio. medium = smaller/faster, lower accuracy.
Skip silence before decoding. Cuts compute on sparse audio but can clip word boundaries.
faster-whisper cpu_threads. 4 measured best on a 16 GB Apify run (~4 vCPU).
BatchedInferencePipeline batch size: VAD chunks decoded in parallel so every cpu_thread stays busy. 0/1 = sequential decode.
Empty = transcribe one window per run. 'batch' = one warm worker processes the list in 'windows' (recommended). 'pool_worker' = pull windows from an Apify request queue (same account, full permissions required).
Apify request queue holding the windows. fetch_next_request is atomic, so workers never take the same window twice.
Dataset every worker pushes its window rows into, so the engine reads one stream.
Free-form label echoed back in each row for telemetry.
Exit after this many seconds with an empty queue. 2-120.
Hard stop for one worker. 30-900.
Hard stop after this many windows. 1-512.
