Fast transcription vs batch transcription vs real-time transcription vs custom speech
Verdict: One uploaded file needing subtitles fast: fast transcription (synchronous, low latency, diarization built in). A large stored archive with no deadline: batch transcription. Audio as it is spoken live: real-time transcription. Domain vocabulary or noisy audio: custom speech.
| Criterion | Fast transcription | Batch transcription | Real-time transcription | Custom speech |
|---|---|---|---|---|
| Turnaround | Synchronous, predictable low latency | Asynchronous, no immediate deadline | Live, as audio is spoken | Not a turnaround mode - an accuracy tuning layer |
| Input | One already-recorded file | Large volumes of stored audio | A live audio source, e.g. a microphone | Applies on top of any of the above |
| Extra features | Supports language identification and diarization out of the box | - | - | Improves recognition accuracy for domain-specific vocabulary or noisy audio |
| Choose when | Subtitles or a transcript are needed right away for one uploaded file | Thousands of hours of prerecorded audio need a searchable transcript with no urgency | Audio must be transcribed as it is spoken, not from a stored file | Accuracy on specialised vocabulary or noisy audio is the problem, not speed |
Rules
- Fast transcription returns a synchronous result with predictable low latency, matching a need for subtitles on one uploaded video as quickly as possible.
- Batch transcription is built to process large stored audio volumes asynchronously, matching an archive of prerecorded content with no immediate deadline.
- The fast transcription API supports language identification and diarization out of the box, making it the option for near-immediate file transcription with speakers separated.
Traps
- Batch transcription processes large volumes of stored audio asynchronously - it does not suit a request for one file's subtitles right away.
- Fast transcription returns a synchronous result; it is not built to handle a large archive asynchronously despite also being quick per file.
- A microphone AudioConfig for real-time transcription captures live audio - it cannot ingest an already-recorded file.