Speech to text

Transcribe batch and streaming audio through one speech-to-text API, with an explicit model choice and optional downstream processing stages.

Speech-to-text is the first stage of most Ultrafield pipelines. It converts an audio file, audio URL, or live stream into text plus timing and model metadata.

Batch transcription

Batch processing fits recordings that already exist: calls, meetings, interviews, media, and archives. Submit the audio, track the resulting job, and consume the finalized transcript directly or through a webhook.

Use batch processing when completeness and throughput matter more than receiving partial words while audio is still arriving.

Streaming transcription

Streaming processing accepts audio as it is produced and emits interim and final transcript events. Use it for captions, live analytics, assistants, and other latency-sensitive experiences.

See Streaming for connection and product-design considerations.

Output shape

A transcript can include:

  • Final text and segment boundaries
  • Word or segment timestamps when supported by the selected model
  • Language information
  • Speaker labels when diarization is enabled
  • Formatting and understanding results from selected processors
  • Model and processing metadata for observability

Availability can vary by model and processing path. Keep the model explicit when consistent output characteristics matter.

Improve real-world results

Evaluate with representative codecs, sample rates, accents, background noise, domain vocabulary, and overlapping speech. Clean studio audio rarely predicts production performance by itself.

On this page