Some transcripts are needed while the person is still talking. A voice agent has to answer the caller, and live captions have to keep up with the speaker. Others can wait until the recording is finished, and then they can be complete, with speaker numbers and a caption file.
Paxa sells both as separate models with separate rates. Both are multilingual, and both keep each word in the script it was spoken in.
What is the difference between realtime and recorded transcription?#
| Recorded | Realtime | |
|---|---|---|
| Model | Paxa STT Lite | Paxa STT Lite Realtime |
| Input | A finished audio file | Audio streamed as it is spoken |
| Output | The whole transcript at once | One final transcript per turn, as each turn ends |
| Speaker numbers | Yes, with Speakers on | Recorded files only |
| Captions | SRT and VTT files | Word timestamps on each turn |
| Longest input | 120 minutes per file | 4 hours per session |
When should you use recorded transcription?#
Use it whenever the audio already exists and nobody is waiting on the words. The whole file is read before the transcript comes back. That is what makes speaker numbers and caption files possible.
- Meetings, for minutes and follow-up actions. See how to transcribe a Zoom meeting.
- Interviews, for quotes checked against the audio. See how to transcribe an interview.
- Podcasts and videos, for show notes and captions. See how to transcribe a podcast and the subtitle generator.
- Call recordings, for review after the call. See the call recordings page.
When should you use realtime transcription?#
Use it when the text has to arrive while the conversation is still going. A voice agent needs each turn of the caller's speech the moment the caller stops. Live captions and dictation need words on screen as they are said.
Paxa STT Lite Realtime is streaming transcription with automatic turn detection. The session reports when someone starts and stops speaking, then sends one final transcript for that turn. Realtime speech to text shows a session turn by turn, and the voice agents page and the live transcription page show where it is used.
How much does each one cost?#
Each model has its own published rate. Recorded transcription bills the length of the file, and realtime transcription bills the audio it receives, silence included.
| Recorded | Realtime | |
|---|---|---|
| Per minute | 8.33 credits | 12.5 credits |
| Per hour | 500 credits | 750 credits |
| Per hour, in money | $0.50 | $0.75 |
| 100 free credits cover | About 12 minutes | About 8 minutes |
A realtime session bills every minute it is open and receiving audio, pauses included. Close the session when the conversation ends.
Free credits are for personal use and evaluation. Buy any plan or credit pack, and the account's transcripts are yours to use commercially, including those made before the purchase. The Terms set this out.
Can you use both on the same conversation?#
Yes, and many teams do. Realtime transcription runs during the call to drive the agent or the live captions. Afterwards the recording goes through recorded transcription, which adds speaker numbers and a clean transcript for the record.
How can you try each one?#
- RecordedOpen the Transcribe tool, drop in an audio file, and run it. The cost shows before you start.
- RealtimeIn the same tool, open Transcribe live and speak into your microphone. Each turn appears once you pause, and the running cost shows as you go.
Note
Paxa STT Lite and Paxa STT Lite Realtime are research previews. Their output and options may change between releases.
What should you read or try next?#
- Read about recorded transcription on the speech to text page.
- See the realtime session in detail on the realtime speech to text page.
- Read realtime Thai transcription turns for how turns fall in Thai speech.
- Read what an AI voice agent is for where realtime transcription sits in a voice agent.
- Compare transcription services on the speech to text API comparison.
Common questions#
Is realtime transcription less accurate than recorded?#
Paxa publishes no accuracy figure for either model. Test both on your own audio, and measuring transcription accuracy shows how to score a sample.
Why does realtime transcription bill silence?#
The session listens to everything it receives to find where each turn starts and ends, pauses included. Billing follows the audio received, and the rate is published as its own figure.
How does it know when a turn ends?#
The session detects when the speaker stops and closes the turn there. An interim transcript may arrive at a natural pause, and the final transcript replaces it.
Does realtime transcription number the speakers?#
Speaker numbers come with recorded transcription. Transcribe the recording afterwards with Speakers on to get them.