Skip to main content

Performance

Sending Audio and Response Times​

Quality of results and response times relate to each other. In order to receive a timely response, it is important to send small chunks of data. But for the quality of your results, it's better to send larger chunks. This all relates to the contextual awareness of the system.

Recommendation: Scriptix advises sending chunks between 8KB and 64KB in size.

  • Smaller chunks work but may result in lower transcription quality.
  • Larger chunks are fine too, but will increase response times.

💡 With 16kHz PCM Wave format, 1 second of audio ≈ 32KB.

8kHz Models​

Scriptix provides a few private models for call center purposes. These models are trained on 8kHz, 16-bit little-endian audio data.

If you use these models, make sure to adjust the audio rate and corresponding block size accordingly.